Vollständiges Transkript anzeigen (25.609 Wörter)
Ross Hudgens: JH, we had a breakfast this morning talking B2B content, there is a specific through-line in there that I think is worth leading with. What was that through-line that came top of mind for you?
JH. Scherck: Well, I think I had kind of a different approach to AI and content than a lot of people at the breakfast. A lot of people said, "Like wait, we might be using it a little bit, we aren't using it at all, we really prioritize human writing." And I think, you know, we've spent a lot of time over the last year and a half experimenting with AI content, producing AI content, publishing it, tracking it, and I'd say for getting 85% of the way there, I think AI with the right context and the right execution does a better job than most humans in the content marketing field today. It's that last 10, 15% of getting across the finish line that I don't think AI can do, or not do well, and I think that requires a human to take that over. But for getting it most of the way there, just before the final mile, I think AI is doing a better job than most people that are producing content by hand today.
Ross Hudgens: I don't know if you're comfortable sharing the story you gave of some client situation that comes top of mind for me?
JH. Scherck: Yeah. Yeah, I think, you know, whenever we get feedback on a, you know, a piece of content that we send over for client review and let's say there's like some pretty strong feedback of like, "Hey, we don't want to frame it this way or this seems off or this seems like AI speak." When we go and we look at the edits in the post-history, a lot of the times those edits are made by people and not not AI, and I think that people have this very, um, when you're editing a piece of content from a human, it's like, "Okay, a human did this, I've empathy for them, they're working with me, I don't want to demotivate them." When it's, when they know that AI is not just like when it's even if they know humans have edited it, when they know that AI is sort of the starting point, all of that like empathy and care kind of goes out the window and they just have like very strong feedback, and I think they're almost in some some cases overly critical, and it's because there's just a lot a lack of trust of AI or people think it's inherently not good. But you know, we when we use AI, we have several layers of human review, and we find that oftentimes it's edits that people make that get the strongest feedback, and when it comes to things like research, brand guideline adherence, uh, pulling information from whether it's like gong calls or webinars, fact checking, the AI does a really good job of that. Not like a, not just like a one-shot with Claude, but like an agent team with deterministic checks for like, does the code samples in the article work? Are the links that are going to, do they all function? Are they pointed to the right thing? Are the products described accurately? AI does a phenomenal job of figuring that out and checking that information, more so than like I think a human can put in time to do it. Like if you were to to go through all the steps that AI is doing and you have a human do that, it would be very unprofitable. Now, I think the the real issue is like the the through-line and the narrative and does every every section have a job to be done, because like it is still probabilistic and guessing, it doesn't actually know if this is high-quality content, but it can deterministically figure out, "Do I adhere to brand guidelines? Am I saying the right things? Am I getting this accurately?" And with very technical products, having a freelance writer produce that content, they're not always going to understand the product as well as say like AI could, especially like a developer tool product or an open source product that it actually can write all the code for, pretty much perfectly. So, I think that getting the article 80 or 90% of the way there, with the right inputs and human oversight is honestly like a more efficient way and a and often times like a more guideline adherent way to produce content. Now, there still needs to be that unique angle and that comes from a human typically. Um, and I think making it unique and making it stand out, that's something that AI really hasn't figured out. But in terms of like getting the nuts and bolts and the outline there that a human can then polish up, I think AI is doing a pretty damn good job of getting us 85-90% of the way there. Now, that final mile, that's the most exhausting part. So if you're editing that final mile, that's a lot of mental bandwidth and I think like it is kind of draining to go through it. That being said, I don't think the best use of someone's time is taking like an article outline from an SEO person or a brief from an SEO person and fleshing that out and kind of standing that up. I'd much rather give them something to edit. But what we are seeing, and I see this on my team, I see it with client teams, is like most people want to be the AI doer and not the AI reviewer. I think that there's a big rush to figure out like how to leverage AI and how to use it effectively, and if you are on the the back end and like working with the outputs, that's not as valuable in your career as being like the orchestrator and innovator. So, I think there's like a natural pushback of like, "I don't want to be editing AI slop." And that's fair, but there's a real need in the market for that today, and I think if you can take something that is a BB plus and get it to be an AA plus, that is a very unique skill set, and it's also a mentally draining skill. So if you can get proficient at it and effective at it, you are a very solid force to be reckon with in the content world right now.
Ross Hudgens: Some color we should add for you, in particular for people who don't know you, you run a majority of your clients are dev tools. Is that fair to say?
JH. Scherck: I'd say like 50% of the business is dev tools, and then like data products. Typically things that are highly technical and complex, that have extensive documentation, oftentimes they're open source products which I think LLMs are extremely good at understanding and working with. They've absorbed those code bases, they know it well. So, I'll say like for the technical products, AI generated content tends to work better. For areas where it's more humane and emotional, like I'd say like managing people and and like people management and HR is a good example of that. Like a lot of that content is how to deal with tricky scenarios, like putting someone on a on a performance improvement plan potentially. That not as good. Like I'll just say it's like, but it's not there. But if you're like, "Hey, can I get a a guide to TANCstack table?", yes, it can crush that. So, I think it's like the drier, technical content, it's really good at. But again, that's dry and technical, so how do you make it like interesting? And the thing the thing that's funny is like how AI makes things interesting, it's all like very same patterned, it depends on the model you use. But I'd say like the the scene setting intro is like the biggest tell in AI right now, because it's like dry technical content is a lot of what we do, and you know, Claude specifically will oftentimes be like, "It's 3:00 AM, your micro services fail. No one can tell. Everything's failing silently and you're left in the and it's just like this like stakes get introduced with a strong hook and it's like now this like very easy patterns that we're starting to recognize around like AI trying to make the content more interesting and engaging, but it just reads so fake. So, I think like AI on its own is not great at producing content. You have to have a lot of context and uh, guidance on what to do and what to avoid and what to pull from. So, I'd say like we've been lucky to be able to invest in setting up systems that I would say are are specifically geared for producing best-in-class B2B content, but if you're just like in your, you know, chat window with Claude knocking stuff out, it's not going to be great. Like it goes with a lot of customization, a lot of context engineering, and a lot of experimentation, and that's not going to come from like a like a Claude or ChatGPT web app, it's going to come from like using agentic infrastructure to produce content. Right. Yeah, um, the this situation you described reminds me of a pre-AI time for us, we we were purely a content marketing agency, and there was like an inversion of just how much expertise was needed of what we felt comfortable doing, against now what AI can actually be better than a human at because of that same reality means we were quite mediocre at creating super technical dev content, but AI basically leaps ahead for the average writer pretty significantly, well maybe on the other side of the axis that inversion is not quite there yet, or at least is what some people are saying, maybe consumer audiences, um I know you're talking about even B2B generally, but maybe there's a little bit of more of a What makes you win there?
JH. Scherck: I mean, I just want to, before we even get there, I want to ask a question is like, when was the the SEO content really good? Like when was it really good content that we all like loved and were proud of and put, you know, right front and center for everyone to read? Like we're I think with AI content we're acting like we're now producing this like, there's this narrative from anti-AI folks in the space of like, "Oh, this is just awful, it's slop, no one wants to read it." And it's like, what were we doing 10 years? Like the content we're putting out 10 years ago was this like artisan, amazing, engaging stuff that everyone It was worse. It was bad. We were we were only investing as much as we needed to to win in the channel and nothing more. And now that there is the cost often, and I don't know if I'd say always. Offen. Fine, I'm being I'm being a little hyperbolic, I'm allowed. Uh, but I'd say like, I don't think companies were super proud of their top 10 tips for, you know, running better payroll article or whatever it was. And there was just the bar was lower, and I think the bar has raised as the internet's gotten more mature. And is the AI content perfect? No. Was SEO content ever perfect? Perfect. I would argue no, except for a few exceptional brands who did it very, very well, that weren't just doing SEO, but were also treating it as like a brand play, and making stuff that was really worth your time. It just happened to also get SEO distribution. That was the minority though. Most people were trying to do high-volume cheap content. I mean, when I got my start, we were using like wasn't even Upwork, text broker, Odesk. You text broker, O-lark, and it was just like get the stuff out the door, get the traffic. And now, I think because we have a system that can, or we have technology that can just generate that stuff, we're getting much more critical about the actual quality of it in a way that the industry just never really has before. I'm hearing more about sentence construction, word choice, verbiage on SEO content than I ever did at any point in my career pre-AI. And that's good. I think it's good for us to be thinking about these things and actually like striving for quality. But it's interesting that it took taking humans out of the equation for people to care about the quality of the content. Yeah, it's almost like our retraining the user to never then eventually never come to my website and just do all this in Claude. Well, I think, you know, Mike King talks about Mike King talks about uh, kind of the death of the open web and how things are going to get more and more limited. I don't think he's far off. I think giving out all of your data and just letting LLMs kind of become the abstraction layer that has all the customer interaction and then sending it to whatever they want instead of you getting that traffic is a real risk for some brands, especially brands that are traffic powerhouses that are losing that that steam right now. Like uh, and but, I think closing off your your entire website from LLMs is going to be a very precarious Yes, that's a that's a that's a "damned if you do, damned if you don't" scenario. Um, we're having dinner with someone last night who does that and he's like, "We're still one of the top-cited things, LLMs love us," which I thought was very interesting. Very big consumer website. Yeah. Yeah, yeah, yeah. Um, yeah, it's interesting where this, this all will go, um, with those things. One of the, kind of going circling back to the AI content side, you said there's some perilous aspects of this that people aren't talking about right now. What, what were you referring to there?
JH. Scherck: You know, we've experimented a lot with AI content, and I think we've gotten better every quarter. The stuff is meaningfully better. It's still not perfect, and I think one of the bigger issues of AI content is expectations of what people can expect, and I think like let's say you, uh, set up Agent Memory and you can continuously provide feedback over time. I think people think like, "Oh, you're going to be, uh, refining and improving the outputs over time and they're going to eventually get perfect." Well, they'll never be perfect, or at least with the models that we have today they're always going to require some human intervention. I also think AI content, for the most part, is either pretty dry, or it tries to be exciting and really falls flat. And I, I just don't think it is at a point where it can create something like engaging, unique, and interesting that's going to hold the attention of a B2B buyer without strong levels of human intervention. So, and you know, you do hear about people publishing direct from like, "Hey, yeah we just like use this tool, get it all going, we plug it into sanity and just fire it off and it's great," and I'm like, "Well, how do you know? For the first three months, but maybe not the next three." Yeah. There's a lot of confidence that it's like going great, I've seen this, I've seen this playbook before, like as someone who was pretty aggressive in link building in the pre-penguin days, to see all that traffic come down for article marketing directories, or uh, what was it like "Your name at your keywords", I think it was, where it's like blog comment links and everything and it's like, "Oh, Google just knocked that all out." And I think that whether it's Google, whether it's Bing, whether it's Brave, there's going to be some type of defense mechanism that pops up against like lower-quality AI content. But, some of the content that we've done with AI, it's, it's good, like it's, like it's objectively good content that's helpful. So, um, and that could be like location pages, like state or town pages for like a a large like business that's selling, uh, that people are looking up for like a service near me or something like that. Having that content be automatically updated with deterministic checks and uh, populated by an LLM, it can be just as good as what a human can do if you have the right guard rails, the right setup. But, um, is that interesting and engaging? No, it doesn't need to be because the product is the location service. Yeah, and I think, um, when all you have is a hammer, the the whole world looks like nails. Yeah. And I think people are trying to use LLMs in ways that they just really aren't meant to be used. Like we have a uh, a hookup with a database to be able to like chat with your data and people will ask questions about like, "Why is the data this way? What questions are not being answered by this page, or not being answered well, or aren't visible?" That the users are actually coming in and saying like, "Please tell me this." And that feedback loop is really powerful for figuring out what you need to do next. So, I'm a, I, I do think chat as a a way to interact with a website is going to become more and more popular, and we're seeing that take off a little bit with the uh, you know, like the in the corner little chatbots. Those are now starting to open up like full-on chat windows that are closer to an LLM. I don't see that just being for customer support. I see that for educational content taking off as well. My only like hesitation on that, that I that I thought about was the effort versus a good IA would you'd have the solution pages, the feature pages, if you know the voice of the customer pretty well, two clicks versus writing. Maybe it's not a detailed help center question necessarily, but like two clicks should get you there versus the act, but now we have Whisperflow and the like. So maybe we're just talking to our chatbot. Well, I mean, I think there's still going to be a reason to like go around the website, but when you're like, "Hey, do you integrate with ADP and can you, you know, help me do X, Y, and Z?" Uh, like when you have a very specific question that's not going to be answered by the content and it's going to take together piecing a few different things, I think that's like a a a pretty solid use case for AI. Now there's like the risk of like, "Well, are we going to hallucinate and tell the users the wrong thing?" And I think that's something that to be figured out and systems are going to get better over time, but I I I think people's attention spans are going down, and I think people's open time to learn and explore is going down as well, as requirements to hit numbers and goals get more aggressive. So, just being able to ask your question and get that direct answer, I think is what more people are going to want to do versus like, "Let me read five different articles and see if I can get the functionality I need." Okay, so it's almost like a retreining the user to then, eventually, never come to my website and just do all this in Claude. Well, I think, you know, Mike talks about, Mike King talks about uh, kind of the death of the open web and how things are going to get more and more limited. I don't think he's far off. I think giving out all of your data and just letting LLMs kind of become the abstraction layer that has all the customer interaction and then sending it to whatever they want instead of you getting that traffic is a real risk for some brands, especially brands that are traffic powerhouses that are losing that that steam right now. Like uh, and but, I think closing off your your entire website from LLMs is going to be a very precarious Yeah, it's a a that's a "damned if you do, damned if you don't" scenario. Um, we're having dinner with someone last night who does that and he's like, "We're still one of the top-cited things, LLMs love us," which I thought was very interesting. Very big consumer website. Yeah. Yeah, yeah, yeah. Um, yeah, it's interesting where this, this all will go, um, with those things. One of the, kind of going circling back to the AI content side, you said there's some perilous aspects of this that people aren't talking about right now. What, what were you referring to there?
JH. Scherck: You know, we've experimented a lot with AI content, and I think we've gotten better every quarter. The stuff is meaningfully better. It's still not perfect, and I think one of the bigger issues of AI content is expectations of what people can expect, and I think like let's say you, uh, set up Agent Memory and you can continuously provide feedback over time. I think people think like, "Oh, you're going to be, uh, refining and improving the outputs over time and they're going to eventually get perfect." Well, they'll never be perfect, or at least with the models that we have today they're always going to require some human intervention. I also think AI content, for the most part, is either pretty dry, or it tries to be exciting and really falls flat. And I, I just don't think it is at a point where it can create something like engaging, unique, and interesting that's going to hold the attention of a B2B buyer without strong levels of human intervention. So, and you know, you do hear about people publishing direct from like, "Hey, yeah we just like use this tool, get it all going, we plug it into sanity and just fire it off and it's great," and I'm like, "Well, how do you know? For the first three months, but maybe not the next three." Yeah. There's a lot of confidence that it's like going great, I've seen this, I've seen this playbook before, like as someone who was pretty aggressive in link building in the pre-penguin days, to see all that traffic come down for article marketing directories, or uh, what was it like "Your name at your keywords", I think it was, where it's like blog comment links and everything and it's like, "Oh, Google just knocked that all out." And I think that whether it's Google, whether it's Bing, whether it's Brave, there's going to be some type of defense mechanism that pops up against like lower-quality AI content. But, some of the content that we've done with AI, it's, it's good, like it's, like it's objectively good content that's helpful. So, um, and that could be like location pages, like state or town pages for like a a large like business that's selling, uh, that people are looking up for like a service near me or something like that. Having that content be automatically updated with deterministic checks and uh, populated by an LLM, it can be just as good as what a human can do if you have the right guard rails, the right setup. But, um, is that interesting and engaging? No, it doesn't need to be because the product is the location service. Yeah, and I think, um, when all you have is a hammer, the the whole world looks like nails. Yeah. And I think people are trying to use LLMs in ways that they just really aren't meant to be used. Like we have a uh, a hookup with a database to be able to like chat with your data and people will ask questions about like, "Why is the data this way? What questions are not being answered by this page, or not being answered well, or aren't visible?" That the users are actually coming in and saying like, "Please tell me this." And that feedback loop is really powerful for figuring out what you need to do next. So, I'm a, I, I do think chat as a a way to interact with a website is going to become more and more popular, and we're seeing that take off a little bit with the uh, you know, like the in the corner little chatbots. Those are now starting to open up like full-on chat windows that are closer to an LLM. I don't see that just being for customer support. I see that for educational content taking off as well. My only like hesitation on that, that I that I thought about was the effort versus a good IA would you'd have the solution pages, the feature pages, if you know the voice of the customer pretty well, two clicks versus writing. Maybe it's not a detailed help center question necessarily, but like two clicks should get you there versus the act, but now we have Whisperflow and the like. So maybe we're just talking to our chatbot. Well, I mean, I think there's still going to be a reason to like go around the website, but when you're like, "Hey, do you integrate with ADP and can you, you know, help me do X, Y, and Z?" Uh, like when you have a very specific question that's not going to be answered by the content and it's going to take together piecing a few different things, I think that's like a a a pretty solid use case for AI. Now there's like the risk of like, "Well, are we going to hallucinate and tell the users the wrong thing?" And I think that's something that to be figured out and systems are going to get better over time, but I I I think people's attention spans are going down, and I think people's open time to learn and explore is going down as well, as requirements to hit numbers and goals get more aggressive. So, just being able to ask your question and get that direct answer, I think is what more people are going to want to do versus like, "Let me read five different articles and see if I can get the functionality I need." Okay, so it's almost like a retreining the user to then, eventually, never come to my website and just do all this in Claude. Well, I think, you know, Mike talks about, Mike King talks about uh, kind of the death of the open web and how things are going to get more and more limited. I don't think he's far off. I think giving out all of your data and just letting LLMs kind of become the abstraction layer that has all the customer interaction and then sending it to whatever they want instead of you getting that traffic is a real risk for some brands, especially brands that are traffic powerhouses that are losing that that steam right now. Like uh, and but, I think closing off your your entire website from LLMs is going to be a very precarious Yeah, it's a a that's a "damned if you do, damned if you don't" scenario. Um, we're having dinner with someone last night who does that and he's like, "We're still one of the top-cited things, LLMs love us," which I thought was very interesting. Very big consumer website. Yeah. Yeah, yeah, yeah. Um, yeah, it's interesting where this, this all will go, um, with those things. One of the, kind of going circling back to the AI content side, you said there's some perilous aspects of this that people aren't talking about right now. What, what were you referring to there?
JH. Scherck: You know, we've experimented a lot with AI content, and I think we've gotten better every quarter. The stuff is meaningfully better. It's still not perfect, and I think one of the bigger issues of AI content is expectations of what people can expect, and I think like let's say you, uh, set up Agent Memory and you can continuously provide feedback over time. I think people think like, "Oh, you're going to be, uh, refining and improving the outputs over time and they're going to eventually get perfect." Well, they'll never be perfect, or at least with the models that we have today they're always going to require some human intervention. I also think AI content, for the most part, is either pretty dry, or it tries to be exciting and really falls flat. And I, I just don't think it is at a point where it can create something like engaging, unique, and interesting that's going to hold the attention of a B2B buyer without strong levels of human intervention. So, and you know, you do hear about people publishing direct from like, "Hey, yeah we just like use this tool, get it all going, we plug it into sanity and just fire it off and it's great," and I'm like, "Well, how do you know? For the first three months, but maybe not the next three." Yeah. There's a lot of confidence that it's like going great, I've seen this, I've seen this playbook before, like as someone who was pretty aggressive in link building in the pre-penguin days, to see all that traffic come down for article marketing directories, or uh, what was it like "Your name at your keywords", I think it was, where it's like blog comment links and everything and it's like, "Oh, Google just knocked that all out." And I think that whether it's Google, whether it's Bing, whether it's Brave, there's going to be some type of defense mechanism that pops up against like lower-quality AI content. But, some of the content that we've done with AI, it's, it's good, like it's, like it's objectively good content that's helpful. So, um, and that could be like location pages, like state or town pages for like a a large like business that's selling, uh, that people are looking up for like a service near me or something like that. Having that content be automatically updated with deterministic checks and uh, populated by an LLM, it can be just as good as what a human can do if you have the right guard rails, the right setup. But, um, is that interesting and engaging? No, it doesn't need to be because the product is the location service. Yeah, and I think, um, when all you have is a hammer, the the whole world looks like nails. Yeah. And I think people are trying to use LLMs in ways that they just really aren't meant to be used. Like we have a uh, a hookup with a database to be able to like chat with your data and people will ask questions about like, "Why is the data this way? What questions are not being answered by this page, or not being answered well, or aren't visible?" That the users are actually coming in and saying like, "Please tell me this." And that feedback loop is really powerful for figuring out what you need to do next. So, I'm a, I, I do think chat as a a way to interact with a website is going to become more and more popular, and we're seeing that take off a little bit with the uh, you know, like the in the corner little chatbots. Those are now starting to open up like full-on chat windows that are closer to an LLM. I don't see that just being for customer support. I see that for educational content taking off as well. My only like hesitation on that, that I that I thought about was the effort versus a good IA would you'd have the solution pages, the feature pages, if you know the voice of the customer pretty well, two clicks versus writing. Maybe it's not a detailed help center question necessarily, but like two clicks should get you there versus the act, but now we have Whisperflow and the like. So maybe we're just talking to our chatbot. Well, I mean, I think there's still going to be a reason to like go around the website, but when you're like, "Hey, do you integrate with ADP and can you, you know, help me do X, Y, and Z?" Uh, like when you have a very specific question that's not going to be answered by the content and it's going to take together piecing a few different things, I think that's like a a a pretty solid use case for AI. Now there's like the risk of like, "Well, are we going to hallucinate and tell the users the wrong thing?" And I think that's something that to be figured out and systems are going to get better over time, but I I I think people's attention spans are going down, and I think people's open time to learn and explore is going down as well, as requirements to hit numbers and goals get more aggressive. So, just being able to ask your question and get that direct answer, I think is what more people are going to want to do versus like, "Let me read five different articles and see if I can get the functionality I need." Okay, so it's almost like a retreining the user to then, eventually, never come to my website and just do all this in Claude. Well, I think, you know, Mike talks about, Mike King talks about uh, kind of the death of the open web and how things are going to get more and more limited. I don't think he's far off. I think giving out all of your data and just letting LLMs kind of become the abstraction layer that has all the customer interaction and then sending it to whatever they want instead of you getting that traffic is a real risk for some brands, especially brands that are traffic powerhouses that are losing that that steam right now. Like uh, and but, I think closing off your your entire website from LLMs is going to be a very precarious Yeah, it's a a that's a "damned if you do, damned if you don't" scenario. Um, we're having dinner with someone last night who does that and he's like, "We're still one of the top-cited things, LLMs love us," which I thought was very interesting. Very big consumer website. Yeah. Yeah, yeah, yeah. Um, yeah, it's interesting where this, this all will go, um, with those things. One of the, kind of going circling back to the AI content side, you said there's some perilous aspects of this that people aren't talking about right now. What, what were you referring to there?
JH. Scherck: You know, we've experimented a lot with AI content, and I think we've gotten better every quarter. The stuff is meaningfully better. It's still not perfect, and I think one of the bigger issues of AI content is expectations of what people can expect, and I think like let's say you, uh, set up Agent Memory and you can continuously provide feedback over time. I think people think like, "Oh, you're going to be, uh, refining and improving the outputs over time and they're going to eventually get perfect." Well, they'll never be perfect, or at least with the models that we have today they're always going to require some human intervention. I also think AI content, for the most part, is either pretty dry, or it tries to be exciting and really falls flat. And I, I just don't think it is at a point where it can create something like engaging, unique, and interesting that's going to hold the attention of a B2B buyer without strong levels of human intervention. So, and you know, you do hear about people publishing direct from like, "Hey, yeah we just like use this tool, get it all going, we plug it into sanity and just fire it off and it's great," and I'm like, "Well, how do you know? For the first three months, but maybe not the next three." Yeah. There's a lot of confidence that it's like going great, I've seen this, I've seen this playbook before, like as someone who was pretty aggressive in link building in the pre-penguin days, to see all that traffic come down for article marketing directories, or uh, what was it like "Your name at your keywords", I think it was, where it's like blog comment links and everything and it's like, "Oh, Google just knocked that all out." And I think that whether it's Google, whether it's Bing, whether it's Brave, there's going to be some type of defense mechanism that pops up against like lower-quality AI content. But, some of the content that we've done with AI, it's, it's good, like it's, like it's objectively good content that's helpful. So, um, and that could be like location pages, like state or town pages for like a a large like business that's selling, uh, that people are looking up for like a service near me or something like that. Having that content be automatically updated with deterministic checks and uh, populated by an LLM, it can be just as good as what a human can do if you have the right guard rails, the right setup. But, um, is that interesting and engaging? No, it doesn't need to be because the product is the location service. Yeah, and I think, um, when all you have is a hammer, the the whole world looks like nails. Yeah. And I think people are trying to use LLMs in ways that they just really aren't meant to be used. Like we have a uh, a hookup with a database to be able to like chat with your data and people will ask questions about like, "Why is the data this way? What questions are not being answered by this page, or not being answered well, or aren't visible?" That the users are actually coming in and saying like, "Please tell me this." And that feedback loop is really powerful for figuring out what you need to do next. So, I'm a, I, I do think chat as a a way to interact with a website is going to become more and more popular, and we're seeing that take off a little bit with the uh, you know, like the in the corner little chatbots. Those are now starting to open up like full-on chat windows that are closer to an LLM. I don't see that just being for customer support. I see that for educational content taking off as well. My only like hesitation on that, that I that I thought about was the effort versus a good IA would you'd have the solution pages, the feature pages, if you know the voice of the customer pretty well, two clicks versus writing. Maybe it's not a detailed help center question necessarily, but like two clicks should get you there versus the act, but now we have Whisperflow and the like. So maybe we're just talking to our chatbot. Well, I mean, I think there's still going to be a reason to like go around the website, but when you're like, "Hey, do you integrate with ADP and can you, you know, help me do X, Y, and Z?" Uh, like when you have a very specific question that's not going to be answered by the content and it's going to take together piecing a few different things, I think that's like a a a pretty solid use case for AI. Now there's like the risk of like, "Well, are we going to hallucinate and tell the users the wrong thing?" And I think that's something that to be figured out and systems are going to get better over time, but I I I think people's attention spans are going down, and I think people's open time to learn and explore is going down as well, as requirements to hit numbers and goals get more aggressive. So, just being able to ask your question and get that direct answer, I think is what more people are going to want to do versus like, "Let me read five different articles and see if I can get the functionality I need." Okay, so it's almost like a retreining the user to then, eventually, never come to my website and just do all this in Claude. Well, I think, you know, Mike talks about, Mike King talks about uh, kind of the death of the open web and how things are going to get more and more limited. I don't think he's far off. I think giving out all of your data and just letting LLMs kind of become the abstraction layer that has all the customer interaction and then sending it to whatever they want instead of you getting that traffic is a real risk for some brands, especially brands that are traffic powerhouses that are losing that that steam right now. Like uh, and but, I think closing off your your entire website from LLMs is going to be a very precarious Yeah, it's a a that's a "damned if you do, damned if you don't" scenario. Um, we're having dinner with someone last night who does that and he's like, "We're still one of the top-cited things, LLMs love us," which I thought was very interesting. Very big consumer website. Yeah. Yeah, yeah, yeah. Um, yeah, it's interesting where this, this all will go, um, with those things. One of the, kind of going circling back to the AI content side, you said there's some perilous aspects of this that people aren't talking about right now. What, what were you referring to there?
JH. Scherck: You know, we've experimented a lot with AI content, and I think we've gotten better every quarter. The stuff is meaningfully better. It's still not perfect, and I think one of the bigger issues of AI content is expectations of what people can expect, and I think like let's say you, uh, set up Agent Memory and you can continuously provide feedback over time. I think people think like, "Oh, you're going to be, uh, refining and improving the outputs over time and they're going to eventually get perfect." Well, they'll never be perfect, or at least with the models that we have today they're always going to require some human intervention. I also think AI content, for the most part, is either pretty dry, or it tries to be exciting and really falls flat. And I, I just don't think it is at a point where it can create something like engaging, unique, and interesting that's going to hold the attention of a B2B buyer without strong levels of human intervention. So, and you know, you do hear about people publishing direct from like, "Hey, yeah we just like use this tool, get it all going, we plug it into sanity and just fire it off and it's great," and I'm like, "Well, how do you know? For the first three months, but maybe not the next three." Yeah. There's a lot of confidence that it's like going great, I've seen this, I've seen this playbook before, like as someone who was pretty aggressive in link building in the pre-penguin days, to see all that traffic come down for article marketing directories, or uh, what was it like "Your name at your keywords", I think it was, where it's like blog comment links and everything and it's like, "Oh, Google just knocked that all out." And I think that whether it's Google, whether it's Bing, whether it's Brave, there's going to be some type of defense mechanism that pops up against like lower-quality AI content. But, some of the content that we've done with AI, it's, it's good, like it's, like it's objectively good content that's helpful. So, um, and that could be like location pages, like state or town pages for like a a large like business that's selling, uh, that people are looking up for like a service near me or something like that. Having that content be automatically updated with deterministic checks and uh, populated by an LLM, it can be just as good as what a human can do if you have the right guard rails, the right setup. But, um, is that interesting and engaging? No, it doesn't need to be because the product is the location service. Yeah, and I think, um, when all you have is a hammer, the the whole world looks like nails. Yeah. And I think people are trying to use LLMs in ways that they just really aren't meant to be used. Like we have a uh, a hookup with a database to be able to like chat with your data and people will ask questions about like, "Why is the data this way? What questions are not being answered by this page, or not being answered well, or aren't visible?" That the users are actually coming in and saying like, "Please tell me this." And that feedback loop is really powerful for figuring out what you need to do next. So, I'm a, I, I do think chat as a a way to interact with a website is going to become more and more popular, and we're seeing that take off a little bit with the uh, you know, like the in the corner little chatbots. Those are now starting to open up like full-on chat windows that are closer to an LLM. I don't see that just being for customer support. I see that for educational content taking off as well. My only like hesitation on that, that I that I thought about was the effort versus a good IA would you'd have the solution pages, the feature pages, if you know the voice of the customer pretty well, two clicks versus writing. Maybe it's not a detailed help center question necessarily, but like two clicks should get you there versus the act, but now we have Whisperflow and the like. So maybe we're just talking to our chatbot. Well, I mean, I think there's still going to be a reason to like go around the website, but when you're like, "Hey, do you integrate with ADP and can you, you know, help me do X, Y, and Z?" Uh, like when you have a very specific question that's not going to be answered by the content and it's going to take together piecing a few different things, I think that's like a a a pretty solid use case for AI. Now there's like the risk of like, "Well, are we going to hallucinate and tell the users the wrong thing?" And I think that's something that to be figured out and systems are going to get better over time, but I I I think people's attention spans are going down, and I think people's open time to learn and explore is going down as well, as requirements to hit numbers and goals get more aggressive. So, just being able to ask your question and get that direct answer, I think is what more people are going to want to do versus like, "Let me read five different articles and see if I can get the functionality I need." Okay, so it's almost like a retreining the user to then, eventually, never come to my website and just do all this in Claude. Well, I think, you know, Mike talks about, Mike King talks about uh, kind of the death of the open web and how things are going to get more and more limited. I don't think he's far off. I think giving out all of your data and just letting LLMs kind of become the abstraction layer that has all the customer interaction and then sending it to whatever they want instead of you getting that traffic is a real risk for some brands, especially brands that are traffic powerhouses that are losing that that steam right now. Like uh, and but, I think closing off your your entire website from LLMs is going to be a very precarious Yeah, it's a a that's a "damned if you do, damned if you don't" scenario. Um, we're having dinner with someone last night who does that and he's like, "We're still one of the top-cited things, LLMs love us," which I thought was very interesting. Very big consumer website. Yeah. Yeah, yeah, yeah. Um, yeah, it's interesting where this, this all will go, um, with those things. One of the, kind of going circling back to the AI content side, you said there's some perilous aspects of this that people aren't talking about right now. What, what were you referring to there?
JH. Scherck: You know, we've experimented a lot with AI content, and I think we've gotten better every quarter. The stuff is meaningfully better. It's still not perfect, and I think one of the bigger issues of AI content is expectations of what people can expect, and I think like let's say you, uh, set up Agent Memory and you can continuously provide feedback over time. I think people think like, "Oh, you're going to be, uh, refining and improving the outputs over time and they're going to eventually get perfect." Well, they'll never be perfect, or at least with the models that we have today they're always going to require some human intervention. I also think AI content, for the most part, is either pretty dry, or it tries to be exciting and really falls flat. And I, I just don't think it is at a point where it can create something like engaging, unique, and interesting that's going to hold the attention of a B2B buyer without strong levels of human intervention. So, and you know, you do hear about people publishing direct from like, "Hey, yeah we just like use this tool, get it all going, we plug it into sanity and just fire it off and it's great," and I'm like, "Well, how do you know? For the first three months, but maybe not the next three." Yeah. There's a lot of confidence that it's like going great, I've seen this, I've seen this playbook before, like as someone who was pretty aggressive in link building in the pre-penguin days, to see all that traffic come down for article marketing directories, or uh, what was it like "Your name at your keywords", I think it was, where it's like blog comment links and everything and it's like, "Oh, Google just knocked that all out." And I think that whether it's Google, whether it's Bing, whether it's Brave, there's going to be some type of defense mechanism that pops up against like lower-quality AI content. But, some of the content that we've done with AI, it's, it's good, like it's, like it's objectively good content that's helpful. So, um, and that could be like location pages, like state or town pages for like a a large like business that's selling, uh, that people are looking up for like a service near me or something like that. Having that content be automatically updated with deterministic checks and uh, populated by an LLM, it can be just as good as what a human can do if you have the right guard rails, the right setup. But, um, is that interesting and engaging? No, it doesn't need to be because the product is the location service. Yeah, and I think, um, when all you have is a hammer, the the whole world looks like nails. Yeah. And I think people are trying to use LLMs in ways that they just really aren't meant to be used. Like we have a uh, a hookup with a database to be able to like chat with your data and people will ask questions about like, "Why is the data this way? What questions are not being answered by this page, or not being answered well, or aren't visible?" That the users are actually coming in and saying like, "Please tell me this." And that feedback loop is really powerful for figuring out what you need to do next. So, I'm a, I, I do think chat as a a way to interact with a website is going to become more and more popular, and we're seeing that take off a little bit with the uh, you know, like the in the corner little chatbots. Those are now starting to open up like full-on chat windows that are closer to an LLM. I don't see that just being for customer support. I see that for educational content taking off as well. My only like hesitation on that, that I that I thought about was the effort versus a good IA would you'd have the solution pages, the feature pages, if you know the voice of the customer pretty well, two clicks versus writing. Maybe it's not a detailed help center question necessarily, but like two clicks should get you there versus the act, but now we have Whisperflow and the like. So maybe we're just talking to our chatbot. Well, I mean, I think there's still going to be a reason to like go around the website, but when you're like, "Hey, do you integrate with ADP and can you, you know, help me do X, Y, and Z?" Uh, like when you have a very specific question that's not going to be answered by the content and it's going to take together piecing a few different things, I think that's like a a a pretty solid use case for AI. Now there's like the risk of like, "Well, are we going to hallucinate and tell the users the wrong thing?" And I think that's something that to be figured out and systems are going to get better over time, but I I I think people's attention spans are going down, and I think people's open time to learn and explore is going down as well, as requirements to hit numbers and goals get more aggressive. So, just being able to ask your question and get that direct answer, I think is what more people are going to want to do versus like, "Let me read five different articles and see if I can get the functionality I need." Okay, so it's almost like a retreining the user to then, eventually, never come to my website and just do all this in Claude. Well, I think, you know, Mike talks about, Mike King talks about uh, kind of the death of the open web and how things are going to get more and more limited. I don't think he's far off. I think giving out all of your data and just letting LLMs kind of become the abstraction layer that has all the customer interaction and then sending it to whatever they want instead of you getting that traffic is a real risk for some brands, especially brands that are traffic powerhouses that are losing that that steam right now. Like uh, and but, I think closing off your your entire website from LLMs is going to be a very precarious Yeah, it's a a that's a "damned if you do, damned if you don't" scenario. Um, we're having dinner with someone last night who does that and he's like, "We're still one of the top-cited things, LLMs love us," which I thought was very interesting. Very big consumer website. Yeah. Yeah, yeah, yeah. Um, yeah, it's interesting where this, this all will go, um, with those things. One of the, kind of going circling back to the AI content side, you said there's some perilous aspects of this that people aren't talking about right now. What, what were you referring to there?
JH. Scherck: You know, we've experimented a lot with AI content, and I think we've gotten better every quarter. The stuff is meaningfully better. It's still not perfect, and I think one of the bigger issues of AI content is expectations of what people can expect, and I think like let's say you, uh, set up Agent Memory and you can continuously provide feedback over time. I think people think like, "Oh, you're going to be, uh, refining and improving the outputs over time and they're going to eventually get perfect." Well, they'll never be perfect, or at least with the models that we have today they're always going to require some human intervention. I also think AI content, for the most part, is either pretty dry, or it tries to be exciting and really falls flat. And I, I just don't think it is at a point where it can create something like engaging, unique, and interesting that's going to hold the attention of a B2B buyer without strong levels of human intervention. So, and you know, you do hear about people publishing direct from like, "Hey, yeah we just like use this tool, get it all going, we plug it into sanity and just fire it off and it's great," and I'm like, "Well, how do you know? For the first three months, but maybe not the next three." Yeah. There's a lot of confidence that it's like going great, I've seen this, I've seen this playbook before, like as someone who was pretty aggressive in link building in the pre-penguin days, to see all that traffic come down for article marketing directories, or uh, what was it like "Your name at your keywords", I think it was, where it's like blog comment links and everything and it's like, "Oh, Google just knocked that all out." And I think that whether it's Google, whether it's Bing, whether it's Brave, there's going to be some type of defense mechanism that pops up against like lower-quality AI content. But, some of the content that we've done with AI, it's, it's good, like it's, like it's objectively good content that's helpful. So, um, and that could be like location pages, like state or town pages for like a a large like business that's selling, uh, that people are looking up for like a service near me or something like that. Having that content be automatically updated with deterministic checks and uh, populated by an LLM, it can be just as good as what a human can do if you have the right guard rails, the right setup. But, um, is that interesting and engaging? No, it doesn't need to be because the product is the location service. Yeah, and I think, um, when all you have is a hammer, the the whole world looks like nails. Yeah. And I think people are trying to use LLMs in ways that they just really aren't meant to be used. Like we have a uh, a hookup with a database to be able to like chat with your data and people will ask questions about like, "Why is the data this way? What questions are not being answered by this page, or not being answered well, or aren't visible?" That the users are actually coming in and saying like, "Please tell me this." And that feedback loop is really powerful for figuring out what you need to do next. So, I'm a, I, I do think chat as a a way to interact with a website is going to become more and more popular, and we're seeing that take off a little bit with the uh, you know, like the in the corner little chatbots. Those are now starting to open up like full-on chat windows that are closer to an LLM. I don't see that just being for customer support. I see that for educational content taking off as well. My only like hesitation on that, that I that I thought about was the effort versus a good IA would you'd have the solution pages, the feature pages, if you know the voice of the customer pretty well, two clicks versus writing. Maybe it's not a detailed help center question necessarily, but like two clicks should get you there versus the act, but now we have Whisperflow and the like. So maybe we're just talking to our chatbot. Well, I mean, I think there's still going to be a reason to like go around the website, but when you're like, "Hey, do you integrate with ADP and can you, you know, help me do X, Y, and Z?" Uh, like when you have a very specific question that's not going to be answered by the content and it's going to take together piecing a few different things, I think that's like a a a pretty solid use case for AI. Now there's like the risk of like, "Well, are we going to hallucinate and tell the users the wrong thing?" And I think that's something that to be figured out and systems are going to get better over time, but I I I think people's attention spans are going down, and I think people's open time to learn and explore is going down as well, as requirements to hit numbers and goals get more aggressive. So, just being able to ask your question and get that direct answer, I think is what more people are going to want to do versus like, "Let me read five different articles and see if I can get the functionality I need." Okay, so it's almost like a retreining the user to then, eventually, never come to my website and just do all this in Claude. Well, I think, you know, Mike talks about, Mike King talks about uh, kind of the death of the open web and how things are going to get more and more limited. I don't think he's far off. I think giving out all of your data and just letting LLMs kind of become the abstraction layer that has all the customer interaction and then sending it to whatever they want instead of you getting that traffic is a real risk for some brands, especially brands that are traffic powerhouses that are losing that that steam right now. Like uh, and but, I think closing off your your entire website from LLMs is going to be a very precarious Yeah, it's a a that's a "damned if you do, damned if you don't" scenario. Um, we're having dinner with someone last night who does that and he's like, "We're still one of the top-cited things, LLMs love us," which I thought was very interesting. Very big consumer website. Yeah. Yeah, yeah, yeah. Um, yeah, it's interesting where this, this all will go, um, with those things. One of the, kind of going circling back to the AI content side, you said there's some perilous aspects of this that people aren't talking about right now. What, what were you referring to there?
JH. Scherck: You know, we've experimented a lot with AI content, and I think we've gotten better every quarter. The stuff is meaningfully better. It's still not perfect, and I think one of the bigger issues of AI content is expectations of what people can expect, and I think like let's say you, uh, set up Agent Memory and you can continuously provide feedback over time. I think people think like, "Oh, you're going to be, uh, refining and improving the outputs over time and they're going to eventually get perfect." Well, they'll never be perfect, or at least with the models that we have today they're always going to require some human intervention. I also think AI content, for the most part, is either pretty dry, or it tries to be exciting and really falls flat. And I, I just don't think it is at a point where it can create something like engaging, unique, and interesting that's going to hold the attention of a B2B buyer without strong levels of human intervention. So, and you know, you do hear about people publishing direct from like, "Hey, yeah we just like use this tool, get it all going, we plug it into sanity and just fire it off and it's great," and I'm like, "Well, how do you know? For the first three months, but maybe not the next three." Yeah. There's a lot of confidence that it's like going great, I've seen this, I've seen this playbook before, like as someone who was pretty aggressive in link building in the pre-penguin days, to see all that traffic come down for article marketing directories, or uh, what was it like "Your name at your keywords", I think it was, where it's like blog comment links and everything and it's like, "Oh, Google just knocked that all out." And I think that whether it's Google, whether it's Bing, whether it's Brave, there's going to be some type of defense mechanism that pops up against like lower-quality AI content. But, some of the content that we've done with AI, it's, it's good, like it's, like it's objectively good content that's helpful. So, um, and that could be like location pages, like state or town pages for like a a large like business that's selling, uh, that people are looking up for like a service near me or something like that. Having that content be automatically updated with deterministic checks and uh, populated by an LLM, it can be just as good as what a human can do if you have the right guard rails, the right setup. But, um, is that interesting and engaging? No, it doesn't need to be because the product is the location service. Yeah, and I think, um, when all you have is a hammer, the the whole world looks like nails. Yeah. And I think people are trying to use LLMs in ways that they just really aren't meant to be used. Like we have a uh, a hookup with a database to be able to like chat with your data and people will ask questions about like, "Why is the data this way? What questions are not being answered by this page, or not being answered well, or aren't visible?" That the users are actually coming in and saying like, "Please tell me this." And that feedback loop is really powerful for figuring out what you need to do next. So, I'm a, I, I do think chat as a a way to interact with a website is going to become more and more popular, and we're seeing that take off a little bit with the uh, you know, like the in the corner little chatbots. Those are now starting to open up like full-on chat windows that are closer to an LLM. I don't see that just being for customer support. I see that for educational content taking off as well. My only like hesitation on that, that I that I thought about was the effort versus a good IA would you'd have the solution pages, the feature pages, if you know the voice of the customer pretty well, two clicks versus writing. Maybe it's not a detailed help center question necessarily, but like two clicks should get you there versus the act, but now we have Whisperflow and the like. So maybe we're just talking to our chatbot. Well, I mean, I think there's still going to be a reason to like go around the website, but when you're like, "Hey, do you integrate with ADP and can you, you know, help me do X, Y, and Z?" Uh, like when you have a very specific question that's not going to be answered by the content and it's going to take together piecing a few different things, I think that's like a a a pretty solid use case for AI. Now there's like the risk of like, "Well, are we going to hallucinate and tell the users the wrong thing?" And I think that's something that to be figured out and systems are going to get better over time, but I I I think people's attention spans are going down, and I think people's open time to learn and explore is going down as well, as requirements to hit numbers and goals get more aggressive. So, just being able to ask your question and get that direct answer, I think is what more people are going to want to do versus like, "Let me read five different articles and see if I can get the functionality I need." Okay, so it's almost like a retreining the user to then, eventually, never come to my website and just do all this in Claude. Well, I think, you know, Mike talks about, Mike King talks about uh, kind of the death of the open web and how things are going to get more and more limited. I don't think he's far off. I think giving out all of your data and just letting LLMs kind of become the abstraction layer that has all the customer interaction and then sending it to whatever they want instead of you getting that traffic is a real risk for some brands, especially brands that are traffic powerhouses that are losing that that steam right now. Like uh, and but, I think closing off your your entire website from LLMs is going to be a very precarious Yeah, it's a a that's a "damned if you do, damned if you don't" scenario. Um, we're having dinner with someone last night who does that and he's like, "We're still one of the top-cited things, LLMs love us," which I thought was very interesting. Very big consumer website. Yeah. Yeah, yeah, yeah. Um, yeah, it's interesting where this, this all will go, um, with those things. One of the, kind of going circling back to the AI content side, you said there's some perilous aspects of this that people aren't talking about right now. What, what were you referring to there?
JH. Scherck: You know, we've experimented a lot with AI content, and I think we've gotten better every quarter. The stuff is meaningfully better. It's still not perfect, and I think one of the bigger issues of AI content is expectations of what people can expect, and I think like let's say you, uh, set up Agent Memory and you can continuously provide feedback over time. I think people think like, "Oh, you're going to be, uh, refining and improving the outputs over time and they're going to eventually get perfect." Well, they'll never be perfect, or at least with the models that we have today they're always going to require some human intervention. I also think AI content, for the most part, is either pretty dry, or it tries to be exciting and really falls flat. And I, I just don't think it is at a point where it can create something like engaging, unique, and interesting that's going to hold the attention of a B2B buyer without strong levels of human intervention. So, and you know, you do hear about people publishing direct from like, "Hey, yeah we just like use this tool, get it all going, we plug it into sanity and just fire it off and it's great," and I'm like, "Well, how do you know? For the first three months, but maybe not the next three." Yeah. There's a lot of confidence that it's like going great, I've seen this, I've seen this playbook before, like as someone who was pretty aggressive in link building in the pre-penguin days, to see all that traffic come down for article marketing directories, or uh, what was it like "Your name at your keywords", I think it was, where it's like blog comment links and everything and it's like, "Oh, Google just knocked that all out." And I think that whether it's Google, whether it's Bing, whether it's Brave, there's going to be some type of defense mechanism that pops up against like lower-quality AI content. But, some of the content that we've done with AI, it's, it's good, like it's, like it's objectively good content that's helpful. So, um, and that could be like location pages, like state or town pages for like a a large like business that's selling, uh, that people are looking up for like a service near me or something like that. Having that content be automatically updated with deterministic checks and uh, populated by an LLM, it can be just as good as what a human can do if you have the right guard rails, the right setup. But, um, is that interesting and engaging? No, it doesn't need to be because the product is the location service. Yeah, and I think, um, when all you have is a hammer, the the whole world looks like nails. Yeah. And I think people are trying to use LLMs in ways that they just really aren't meant to be used. Like we have a uh, a hookup with a database to be able to like chat with your data and people will ask questions about like, "Why is the data this way? What questions are not being answered by this page, or not being answered well, or aren't visible?" That the users are actually coming in and saying like, "Please tell me this." And that feedback loop is really powerful for figuring out what you need to do next. So, I'm a, I, I do think chat as a a way to interact with a website is going to become more and more popular, and we're seeing that take off a little bit with the uh, you know, like the in the corner little chatbots. Those are now starting to open up like full-on chat windows that are closer to an LLM. I don't see that just being for customer support. I see that for educational content taking off as well. My only like hesitation on that, that I that I thought about was the effort versus a good IA would you'd have the solution pages, the feature pages, if you know the voice of the customer pretty well, two clicks versus writing. Maybe it's not a detailed help center question necessarily, but like two clicks should get you there versus the act, but now we have Whisperflow and the like. So maybe we're just talking to our chatbot. Well, I mean, I think there's still going to be a reason to like go around the website, but when you're like, "Hey, do you integrate with ADP and can you, you know, help me do X, Y, and Z?" Uh, like when you have a very specific question that's not going to be answered by the content and it's going to take together piecing a few different things, I think that's like a a a pretty solid use case for AI. Now there's like the risk of like, "Well, are we going to hallucinate and tell the users the wrong thing?" And I think that's something that to be figured out and systems are going to get better over time, but I I I think people's attention spans are going down, and I think people's open time to learn and explore is going down as well, as requirements to hit numbers and goals get more aggressive. So, just being able to ask your question and get that direct answer, I think is what more people are going to want to do versus like, "Let me read five different articles and see if I can get the functionality I need." Okay, so it's almost like a retreining the user to then, eventually, never come to my website and just do all this in Claude. Well, I think, you know, Mike talks about, Mike King talks about uh, kind of the death of the open web and how things are going to get more and more limited. I don't think he's far off. I think giving out all of your data and just letting LLMs kind of become the abstraction layer that has all the customer interaction and then sending it to whatever they want instead of you getting that traffic is a real risk for some brands, especially brands that are traffic powerhouses that are losing that that steam right now. Like uh, and but, I think closing off your your entire website from LLMs is going to be a very precarious Yeah, it's a a that's a "damned if you do, damned if you don't" scenario. Um, we're having dinner with someone last night who does that and he's like, "We're still one of the top-cited things, LLMs love us," which I thought was very interesting. Very big consumer website. Yeah. Yeah, yeah, yeah. Um, yeah, it's interesting where this, this all will go, um, with those things. One of the, kind of going circling back to the AI content side, you said there's some perilous aspects of this that people aren't talking about right now. What, what were you referring to there?
JH. Scherck: You know, we've experimented a lot with AI content, and I think we've gotten better every quarter. The stuff is meaningfully better. It's still not perfect, and I think one of the bigger issues of AI content is expectations of what people can expect, and I think like let's say you, uh, set up Agent Memory and you can continuously provide feedback over time. I think people think like, "Oh, you're going to be, uh, refining and improving the outputs over time and they're going to eventually get perfect." Well, they'll never be perfect, or at least with the models that we have today they're always going to require some human intervention. I also think AI content, for the most part, is either pretty dry, or it tries to be exciting and really falls flat. And I, I just don't think it is at a point where it can create something like engaging, unique, and interesting that's going to hold the attention of a B2B buyer without strong levels of human intervention. So, and you know, you do hear about people publishing direct from like, "Hey, yeah we just like use this tool, get it all going, we plug it into sanity and just fire it off and it's great," and I'm like, "Well, how do you know? For the first three months, but maybe not the next three." Yeah. There's a lot of confidence that it's like going great, I've seen this, I've seen this playbook before, like as someone who was pretty aggressive in link building in the pre-penguin days, to see all that traffic come down for article marketing directories, or uh, what was it like "Your name at your keywords", I think it was, where it's like blog comment links and everything and it's like, "Oh, Google just knocked that all out." And I think that whether it's Google, whether it's Bing, whether it's Brave, there's going to be some type of defense mechanism that pops up against like lower-quality AI content. But, some of the content that we've done with AI, it's, it's good, like it's, like it's objectively good content that's helpful. So, um, and that could be like location pages, like state or town pages for like a a large like business that's selling, uh, that people are looking up for like a service near me or something like that. Having that content be automatically updated with deterministic checks and uh, populated by an LLM, it can be just as good as what a human can do if you have the right guard rails, the right setup. But, um, is that interesting and engaging? No, it doesn't need to be because the product is the location service. Yeah, and I think, um, when all you have is a hammer, the the whole world looks like nails. Yeah. And I think people are trying to use LLMs in ways that they just really aren't meant to be used. Like we have a uh, a hookup with a database to be able to like chat with your data and people will ask questions about like, "Why is the data this way? What questions are not being answered by this page, or not being answered well, or aren't visible?" That the users are actually coming in and saying like, "Please tell me this." And that feedback loop is really powerful for figuring out what you need to do next. So, I'm a, I, I do think chat as a a way to interact with a website is going to become more and more popular, and we're seeing that take off a little bit with the uh, you know, like the in the corner little chatbots. Those are now starting to open up like full-on chat windows that are closer to an LLM. I don't see that just being for customer support. I see that for educational content taking off as well. My only like hesitation on that, that I that I thought about was the effort versus a good IA would you'd have the solution pages, the feature pages, if you know the voice of the customer pretty well, two clicks versus writing. Maybe it's not a detailed help center question necessarily, but like two clicks should get you there versus the act, but now we have Whisperflow and the like. So maybe we're just talking to our chatbot. Well, I mean, I think there's still going to be a reason to like go around the website, but when you're like, "Hey, do you integrate with ADP and can you, you know, help me do X, Y, and Z?" Uh, like when you have a very specific question that's not going to be answered by the content and it's going to take together piecing a few different things, I think that's like a a a pretty solid use case for AI. Now there's like the risk of like, "Well, are we going to hallucinate and tell the users the wrong thing?" And I think that's something that to be figured out and systems are going to get better over time, but I I I think people's attention spans are going down, and I think people's open time to learn and explore is going down as well, as requirements to hit numbers and goals get more aggressive. So, just being able to ask your question and get that direct answer, I think is what more people are going to want to do versus like, "Let me read five different articles and see if I can get the functionality I need." Okay, so it's almost like a retreining the user to then, eventually, never come to my website and just do all this in Claude. Well, I think, you know, Mike talks about, Mike King talks about uh, kind of the death of the open web and how things are going to get more and more limited. I don't think he's far off. I think giving out all of your data and just letting LLMs kind of become the abstraction layer that has all the customer interaction and then sending it to whatever they want instead of you getting that traffic is a real risk for some brands, especially brands that are traffic powerhouses that are losing that that steam right now. Like uh, and but, I think closing off your your entire website from LLMs is going to be a very precarious Yeah, it's a a that's a "damned if you do, damned if you don't" scenario. Um, we're having dinner with someone last night who does that and he's like, "We're still one of the top-cited things, LLMs love us," which I thought was very interesting. Very big consumer website. Yeah. Yeah, yeah, yeah. Um, yeah, it's interesting where this, this all will go, um, with those things. One of the, kind of going circling back to the AI content side, you said there's some perilous aspects of this that people aren't talking about right now. What, what were you referring to there?
JH. Scherck: You know, we've experimented a lot with AI content, and I think we've gotten better every quarter. The stuff is meaningfully better. It's still not perfect, and I think one of the bigger issues of AI content is expectations of what people can expect, and I think like let's say you, uh, set up Agent Memory and you can continuously provide feedback over time. I think people think like, "Oh, you're going to be, uh, refining and improving the outputs over time and they're going to eventually get perfect." Well, they'll never be perfect, or at least with the models that we have today they're always going to require some human intervention. I also think AI content, for the most part, is either pretty dry, or it tries to be exciting and really falls flat. And I, I just don't think it is at a point where it can create something like engaging, unique, and interesting that's going to hold the attention of a B2B buyer without strong levels of human intervention. So, and you know, you do hear about people publishing direct from like, "Hey, yeah we just like use this tool, get it all going, we plug it into sanity and just fire it off and it's great," and I'm like, "Well, how do you know? For the first three months, but maybe not the next three." Yeah. There's a lot of confidence that it's like going great, I've seen this, I've seen this playbook before, like as someone who was pretty aggressive in link building in the pre-penguin days, to see all that traffic come down for article marketing directories, or uh, what was it like "Your name at your keywords", I think it was, where it's like blog comment links and everything and it's like, "Oh, Google just knocked that all out." And I think that whether it's Google, whether it's Bing, whether it's Brave, there's going to be some type of defense mechanism that pops up against like lower-quality AI content. But, some of the content that we've done with AI, it's, it's good, like it's, like it's objectively good content that's helpful. So, um, and that could be like location pages, like state or town pages for like a a large like business that's selling, uh, that people are looking up for like a service near me or something like that. Having that content be automatically updated with deterministic checks and uh, populated by an LLM, it can be just as good as what a human can do if you have the right guard rails, the right setup. But, um, is that interesting and engaging? No, it doesn't need to be because the product is the location service. Yeah, and I think, um, when all you have is a hammer, the the whole world looks like nails. Yeah. And I think people are trying to use LLMs in ways that they just really aren't meant to be used. Like we have a uh, a hookup with a database to be able to like chat with your data and people will ask questions about like, "Why is the data this way? What questions are not being answered by this page, or not being answered well, or aren't visible?" That the users are actually coming in and saying like, "Please tell me this." And that feedback loop is really powerful for figuring out what you need to do next. So, I'm a, I, I do think chat as a a way to interact with a website is going to become more and more popular, and we're seeing that take off a little bit with the uh, you know, like the in the corner little chatbots. Those are now starting to open up like full-on chat windows that are closer to an LLM. I don't see that just being for customer support. I see that for educational content taking off as well. My only like hesitation on that, that I that I thought about was the effort versus a good IA would you'd have the solution pages, the feature pages, if you know the voice of the customer pretty well, two clicks versus writing. Maybe it's not a detailed help center question necessarily, but like two clicks should get you there versus the act, but now we have Whisperflow and the like. So maybe we're just talking to our chatbot. Well, I mean, I think there's still going to be a reason to like go around the website, but when you're like, "Hey, do you integrate with ADP and can you, you know, help me do X, Y, and Z?" Uh, like when you have a very specific question that's not going to be answered by the content and it's going to take together piecing a few different things, I think that's like a a a pretty solid use case for AI. Now there's like the risk of like, "Well, are we going to hallucinate and tell the users the wrong thing?" And I think that's something that to be figured out and systems are going to get better over time, but I I I think people's attention spans are going down, and I think people's open time to learn and explore is going down as well, as requirements to hit numbers and goals get more aggressive. So, just being able to ask your question and get that direct answer, I think is what more people are going to want to do versus like, "Let me read five different articles and see if I can get the functionality I need." Okay, so it's almost like a retreining the user to then, eventually, never come to my website and just do all this in Claude. Well, I think, you know, Mike talks about, Mike King talks about uh, kind of the death of the open web and how things are going to get more and more limited. I don't think he's far off. I think giving out all of your data and just letting LLMs kind of become the abstraction layer that has all the customer interaction and then sending it to whatever they want instead of you getting that traffic is a real risk for some brands, especially brands that are traffic powerhouses that are losing that that steam right now. Like uh, and but, I think closing off your your entire website from LLMs is going to be a very precarious Yeah, it's a a that's a "damned if you do, damned if you don't" scenario. Um, we're having dinner with someone last night who does that and he's like, "We're still one of the top-cited things, LLMs love us," which I thought was very interesting. Very big consumer website. Yeah. Yeah, yeah, yeah. Um, yeah, it's interesting where this, this all will go, um, with those things. One of the, kind of going circling back to the AI content side, you said there's some perilous aspects of this that people aren't talking about right now. What, what were you referring to there?
JH. Scherck: You know, we've experimented a lot with AI content, and I think we've gotten better every quarter. The stuff is meaningfully better. It's still not perfect, and I think one of the bigger issues of AI content is expectations of what people can expect, and I think like let's say you, uh, set up Agent Memory and you can continuously provide feedback over time. I think people think like, "Oh, you're going to be, uh, refining and improving the outputs over time and they're going to eventually get perfect." Well, they'll never be perfect, or at least with the models that we have today they're always going to require some human intervention. I also think AI content, for the most part, is either pretty dry, or it tries to be exciting and really falls flat. And I, I just don't think it is at a point where it can create something like engaging, unique, and interesting that's going to hold the attention of a B2B buyer without strong levels of human intervention. So, and you know, you do hear about people publishing direct from like, "Hey, yeah we just like use this tool, get it all going, we plug it into sanity and just fire it off and it's great," and I'm like, "Well, how do you know? For the first three months, but maybe not the next three." Yeah. There's a lot of confidence that it's like going great, I've seen this, I've seen this playbook before, like as someone who was pretty aggressive in link building in the pre-penguin days, to see all that traffic come down for article marketing directories, or uh, what was it like "Your name at your keywords", I think it was, where it's like blog comment links and everything and it's like, "Oh, Google just knocked that all out." And I think that whether it's Google, whether it's Bing, whether it's Brave, there's going to be some type of defense mechanism that pops up against like lower-quality AI content. But, some of the content that we've done with AI, it's, it's good, like it's, like it's objectively good content that's helpful. So, um, and that could be like location pages, like state or town pages for like a a large like business that's selling, uh, that people are looking up for like a service near me or something like that. Having that content be automatically updated with deterministic checks and uh, populated by an LLM, it can be just as good as what a human can do if you have the right guard rails, the right setup. But, um, is that interesting and engaging? No, it doesn't need to be because the product is the location service. Yeah, and I think, um, when all you have is a hammer, the the whole world looks like nails. Yeah. And I think people are trying to use LLMs in ways that they just really aren't meant to be used. Like we have a uh, a hookup with a database to be able to like chat with your data and people will ask questions about like, "Why is the data this way? What questions are not being answered by this page, or not being answered well, or aren't visible?" That the users are actually coming in and saying like, "Please tell me this." And that feedback loop is really powerful for figuring out what you need to do next. So, I'm a, I, I do think chat as a a way to interact with a website is going to become more and more popular, and we're seeing that take off a little bit with the uh, you know, like the in the corner little chatbots. Those are now starting to open up like full-on chat windows that are closer to an LLM. I don't see that just being for customer support. I see that for educational content taking off as well. My only like hesitation on that, that I that I thought about was the effort versus a good IA would you'd have the solution pages, the feature pages, if you know the voice of the customer pretty well, two clicks versus writing. Maybe it's not a detailed help center question necessarily, but like two clicks should get you there versus the act, but now we have Whisperflow and the like. So maybe we're just talking to our chatbot. Well, I mean, I think there's still going to be a reason to like go around the website, but when you're like, "Hey, do you integrate with ADP and can you, you know, help me do X, Y, and Z?" Uh, like when you have a very specific question that's not going to be answered by the content and it's going to take together piecing a few different things, I think that's like a a a pretty solid use case for AI. Now there's like the risk of like, "Well, are we going to hallucinate and tell the users the wrong thing?" And I think that's something that to be figured out and systems are going to get better over time, but I I I think people's attention spans are going down, and I think people's open time to learn and explore is going down as well, as requirements to hit numbers and goals get more aggressive. So, just being able to ask your question and get that direct answer, I think is what more people are going to want to do versus like, "Let me read five different articles and see if I can get the functionality I need." Okay, so it's almost like a retreining the user to then, eventually, never come to my website and just do all this in Claude. Well, I think, you know, Mike talks about, Mike King talks about uh, kind of the death of the open web and how things are going to get more and more limited. I don't think he's far off. I think giving out all of your data and just letting LLMs kind of become the abstraction layer that has all the customer interaction and then sending it to whatever they want instead of you getting that traffic is a real risk for some brands, especially brands that are traffic powerhouses that are losing that that steam right now. Like uh, and but, I think closing off your your entire website from LLMs is going to be a very precarious Yeah, it's a a that's a "damned if you do, damned if you don't" scenario. Um, we're having dinner with someone last night who does that and he's like, "We're still one of the top-cited things, LLMs love us," which I thought was very interesting. Very big consumer website. Yeah. Yeah, yeah, yeah. Um, yeah, it's interesting where this, this all will go, um, with those things. One of the, kind of going circling back to the AI content side, you said there's some perilous aspects of this that people aren't talking about right now. What, what were you referring to there?
JH. Scherck: You know, we've experimented a lot with AI content, and I think we've gotten better every quarter. The stuff is meaningfully better. It's still not perfect, and I think one of the bigger issues of AI content is expectations of what people can expect, and I think like let's say you, uh, set up Agent Memory and you can continuously provide feedback over time. I think people think like, "Oh, you're going to be, uh, refining and improving the outputs over time and they're going to eventually get perfect." Well, they'll never be perfect, or at least with the models that we have today they're always going to require some human intervention. I also think AI content, for the most part, is either pretty dry, or it tries to be exciting and really falls flat. And I, I just don't think it is at a point where it can create something like engaging, unique, and interesting that's going to hold the attention of a B2B buyer without strong levels of human intervention. So, and you know, you do hear about people publishing direct from like, "Hey, yeah we just like use this tool, get it all going, we plug it into sanity and just fire it off and it's great," and I'm like, "Well, how do you know? For the first three months, but maybe not the next three." Yeah. There's a lot of confidence that it's like going great, I've seen this, I've seen this playbook before, like as someone who was pretty aggressive in link building in the pre-penguin days, to see all that traffic come down for article marketing directories, or uh, what was it like "Your name at your keywords", I think it was, where it's like blog comment links and everything and it's like, "Oh, Google just knocked that all out." And I think that whether it's Google, whether it's Bing, whether it's Brave, there's going to be some type of defense mechanism that pops up against like lower-quality AI content. But, some of the content that we've done with AI, it's, it's good, like it's, like it's objectively good content that's helpful. So, um, and that could be like location pages, like state or town pages for like a a large like business that's selling, uh, that people are looking up for like a service near me or something like that. Having that content be automatically updated with deterministic checks and uh, populated by an LLM, it can be just as good as what a human can do if you have the right guard rails, the right setup. But, um, is that interesting and engaging? No, it doesn't need to be because the product is the location service. Yeah, and I think, um, when all you have is a hammer, the the whole world looks like nails. Yeah. And I think people are trying to use LLMs in ways that they just really aren't meant to be used. Like we have a uh, a hookup with a database to be able to like chat with your data and people will ask questions about like, "Why is the data this way? What questions are not being answered by this page, or not being answered well, or aren't visible?" That the users are actually coming in and saying like, "Please tell me this." And that feedback loop is really powerful for figuring out what you need to do next. So, I'm a, I, I do think chat as a a way to interact with a website is going to become more and more popular, and we're seeing that take off a little bit with the uh, you know, like the in the corner little chatbots. Those are now starting to open up like full-on chat windows that are closer to an LLM. I don't see that just being for customer support. I see that for educational content taking off as well. My only like hesitation on that, that I that I thought about was the effort versus a good IA would you'd have the solution pages, the feature pages, if you know the voice of the customer pretty well, two clicks versus writing. Maybe it's not a detailed help center question necessarily, but like two clicks should get you there versus the act, but now we have Whisperflow and the like. So maybe we're just talking to our chatbot. Well, I mean, I think there's still going to be a reason to like go around the website, but when you're like, "Hey, do you integrate with ADP and can you, you know, help me do X, Y, and Z?" Uh, like when you have a very specific question that's not going to be answered by the content and it's going to take together piecing a few different things, I think that's like a a a pretty solid use case for AI. Now there's like the risk of like, "Well, are we going to hallucinate and tell the users the wrong thing?" And I think that's something that to be figured out and systems are going to get better over time, but I I I think people's attention spans are going down, and I think people's open time to learn and explore is going down as well, as requirements to hit numbers and goals get more aggressive. So, just being able to ask your question and get that direct answer, I think is what more people are going to want to do versus like, "Let me read five different articles and see if I can get the functionality I need." Okay, so it's almost like a retreining the user to then, eventually, never come to my website and just do all this in Claude. Well, I think, you know, Mike talks about, Mike King talks about uh, kind of the death of the open web and how things are going to get more and more limited. I don't think he's far off. I think giving out all of your data and just letting LLMs kind of become the abstraction layer that has all the customer interaction and then sending it to whatever they want instead of you getting that traffic is a real risk for some brands, especially brands that are traffic powerhouses that are losing that that steam right now. Like uh, and but, I think closing off your your entire website from LLMs is going to be a very precarious Yeah, it's a a that's a "damned if you do, damned if you don't" scenario. Um, we're having dinner with someone last night who does that and he's like, "We're still one of the top-cited things, LLMs love us," which I thought was very interesting. Very big consumer website. Yeah. Yeah, yeah, yeah. Um, yeah, it's interesting where this, this all will go, um, with those things. One of the, kind of going circling back to the AI content side, you said there's some perilous aspects of this that people aren't talking about right now. What, what were you referring to there?
JH. Scherck: You know, we've experimented a lot with AI content, and I think we've gotten better every quarter. The stuff is meaningfully better. It's still not perfect, and I think one of the bigger issues of AI content is expectations of what people can expect, and I think like let's say you, uh, set up Agent Memory and you can continuously provide feedback over time. I think people think like, "Oh, you're going to be, uh, refining and improving the outputs over time and they're going to eventually get perfect." Well, they'll never be perfect, or at least with the models that we have today they're always going to require some human intervention. I also think AI content, for the most part, is either pretty dry, or it tries to be exciting and really falls flat. And I, I just don't think it is at a point where it can create something like engaging, unique, and interesting that's going to hold the attention of a B2B buyer without strong levels of human intervention. So, and you know, you do hear about people publishing direct from like, "Hey, yeah we just like use this tool, get it all going, we plug it into sanity and just fire it off and it's great," and I'm like, "Well, how do you know? For the first three months, but maybe not the next three." Yeah. There's a lot of confidence that it's like going great, I've seen this, I've seen this playbook before, like as someone who was pretty aggressive in link building in the pre-penguin days, to see all that traffic come down for article marketing directories, or uh, what was it like "Your name at your keywords", I think it was, where it's like blog comment links and everything and it's like, "Oh, Google just knocked that all out." And I think that whether it's Google, whether it's Bing, whether it's Brave, there's going to be some type of defense mechanism that pops up against like lower-quality AI content. But, some of the content that we've done with AI, it's, it's good, like it's, like it's objectively good content that's helpful. So, um, and that could be like location pages, like state or town pages for like a a large like business that's selling, uh, that people are looking up for like a service near me or something like that. Having that content be automatically updated with deterministic checks and uh, populated by an LLM, it can be just as good as what a human can do if you have the right guard rails, the right setup. But, um, is that interesting and engaging? No, it doesn't need to be because the product is the location service. Yeah, and I think, um, when all you have is a hammer, the the whole world looks like nails. Yeah. And I think people are trying to use LLMs in ways that they just really aren't meant to be used. Like we have a uh, a hookup with a database to be able to like chat with your data and people will ask questions about like, "Why is the data this way? What questions are not being answered by this page, or not being answered well, or aren't visible?" That the users are actually coming in and saying like, "Please tell me this." And that feedback loop is really powerful for figuring out what you need to do next. So, I'm a, I, I do think chat as a a way to interact with a website is going to become more and more popular, and we're seeing that take off a little bit with the uh, you know, like the in the corner little chatbots. Those are now starting to open up like full-on chat windows that are closer to an LLM. I don't see that just being for customer support. I see that for educational content taking off as well. My only like hesitation on that, that I that I thought about was the effort versus a good IA would you'd have the solution pages, the feature pages, if you know the voice of the customer pretty well, two clicks versus writing. Maybe it's not a detailed help center question necessarily, but like two clicks should get you there versus the act, but now we have Whisperflow and the like. So maybe we're just talking to our chatbot. Well, I mean, I think there's still going to be a reason to like go around the website, but when you're like, "Hey, do you integrate with ADP and can you, you know, help me do X, Y, and Z?" Uh, like when you have a very specific question that's not going to be answered by the content and it's going to take together piecing a few different things, I think that's like a a a pretty solid use case for AI. Now there's like the risk of like, "Well, are we going to hallucinate and tell the users the wrong thing?" And I think that's something that to be figured out and systems are going to get better over time, but I I I think people's attention spans are going down, and I think people's open time to learn and explore is going down as well, as requirements to hit numbers and goals get more aggressive. So, just being able to ask your question and get that direct answer, I think is what more people are going to want to do versus like, "Let me read five different articles and see if I can get the functionality I need." Okay, so it's almost like a retreining the user to then, eventually, never come to my website and just do all this in Claude. Well, I think, you know, Mike talks about, Mike King talks about uh, kind of the death of the open web and how things are going to get more and more limited. I don't think he's far off. I think giving out all of your data and just letting LLMs kind of become the abstraction layer that has all the customer interaction and then sending it to whatever they want instead of you getting that traffic is a real risk for some brands, especially brands that are traffic powerhouses that are losing that that steam right now. Like uh, and but, I think closing off your your entire website from LLMs is going to be a very precarious Yeah, it's a a that's a "damned if you do, damned if you don't" scenario. Um, we're having dinner with someone last night who does that and he's like, "We're still one of the top-cited things, LLMs love us," which I thought was very interesting. Very big consumer website. Yeah. Yeah, yeah, yeah. Um, yeah, it's interesting where this, this all will go, um, with those things. One of the, kind of going circling back to the AI content side, you said there's some perilous aspects of this that people aren't talking about right now. What, what were you referring to there?
JH. Scherck: You know, we've experimented a lot with AI content, and I think we've gotten better every quarter. The stuff is meaningfully better. It's still not perfect, and I think one of the bigger issues of AI content is expectations of what people can expect, and I think like let's say you, uh, set up Agent Memory and you can continuously provide feedback over time. I think people think like, "Oh, you're going to be, uh, refining and improving the outputs over time and they're going to eventually get perfect." Well, they'll never be perfect, or at least with the models that we have today they're always going to require some human intervention. I also think AI content, for the most part, is either pretty dry, or it tries to be exciting and really falls flat. And I, I just don't think it is at a point where it can create something like engaging, unique, and interesting that's going to hold the attention of a B2B buyer without strong levels of human intervention. So, and you know, you do hear about people publishing direct from like, "Hey, yeah we just like use this tool, get it all going, we plug it into sanity and just fire it off and it's great," and I'm like, "Well, how do you know? For the first three months, but maybe not the next three." Yeah. There's a lot of confidence that it's like going great, I've seen this, I've seen this playbook before, like as someone who was pretty aggressive in link building in the pre-penguin days, to see all that traffic come down for article marketing directories, or uh, what was it like "Your name at your keywords", I think it was, where it's like blog comment links and everything and it's like, "Oh, Google just knocked that all out." And I think that whether it's Google, whether it's Bing, whether it's Brave, there's going to be some type of defense mechanism that pops up against like lower-quality AI content. But, some of the content that we've done with AI, it's, it's good, like it's, like it's objectively good content that's helpful. So, um, and that could be like location pages, like state or town pages for like a a large like business that's selling, uh, that people are looking up for like a service near me or something like that. Having that content be automatically updated with deterministic checks and uh, populated by an LLM, it can be just as good as what a human can do if you have the right guard rails, the right setup. But, um, is that interesting and engaging? No, it doesn't need to be because the product is the location service. Yeah, and I think, um, when all you have is a hammer, the the whole world looks like nails. Yeah. And I think people are trying to use LLMs in ways that they just really aren't meant to be used. Like we have a uh, a hookup with a database to be able to like chat with your data and people will ask questions about like, "Why is the data this way? What questions are not being answered by this page, or not being answered well, or aren't visible?" That the users are actually coming in and saying like, "Please tell me this." And that feedback loop is really powerful for figuring out what you need to do next. So, I'm a, I, I do think chat as a a way to interact with a website is going to become more and more popular, and we're seeing that take off a little bit with the uh, you know, like the in the corner little chatbots. Those are now starting to open up like full-on chat windows that are closer to an LLM. I don't see that just being for customer support. I see that for educational content taking off as well. My only like hesitation on that, that I that I thought about was the effort versus a good IA would you'd have the solution pages, the feature pages, if you know the voice of the customer pretty well, two clicks versus writing. Maybe it's not a detailed help center question necessarily, but like two clicks should get you there versus the act, but now we have Whisperflow and the like. So maybe we're just talking to our chatbot. Well, I mean, I think there's still going to be a reason to like go around the website, but when you're like, "Hey, do you integrate with ADP and can you, you know, help me do X, Y, and Z?" Uh, like when you have a very specific question that's not going to be answered by the content and it's going to take together piecing a few different things, I think that's like a a a pretty solid use case for AI. Now there's like the risk of like, "Well, are we going to hallucinate and tell the users the wrong thing?" And I think that's something that to be figured out and systems are going to get better over time, but I I I think people's attention spans are going down, and I think people's open time to learn and explore is going down as well, as requirements to hit numbers and goals get more aggressive. So, just being able to ask your question and get that direct answer, I think is what more people are going to want to do versus like, "Let me read five different articles and see if I can get the functionality I need." Okay, so it's almost like a retreining the user to then, eventually, never come to my website and just do all this in Claude. Well, I think, you know, Mike talks about, Mike King talks about uh, kind of the death of the open web and how things are going to get more and more limited. I don't think he's far off. I think giving out all of your data and just letting LLMs kind of become the abstraction layer that has all the customer interaction and then sending it to whatever they want instead of you getting that traffic is a real risk for some brands, especially brands that are traffic powerhouses that are losing that that steam right now. Like uh, and but, I think closing off your your entire website from LLMs is going to be a very precarious Yeah, it's a a that's a "damned if you do, damned if you don't" scenario. Um, we're having dinner with someone last night who does that and he's like, "We're still one of the top-cited things, LLMs love us," which I thought was very interesting. Very big consumer website. Yeah. Yeah, yeah, yeah. Um, yeah, it's interesting where this, this all will go, um, with those things. One of the, kind of going circling back to the AI content side, you said there's some perilous aspects of this that people aren't talking about right now. What, what were you referring to there?
JH. Scherck: You know, we've experimented a lot with AI content, and I think we've gotten better every quarter. The stuff is meaningfully better. It's still not perfect, and I think one of the bigger issues of AI content is expectations of what people can expect, and I think like let's say you, uh, set up Agent Memory and you can continuously provide feedback over time. I think people think like, "Oh, you're going to be, uh, refining and improving the outputs over time and they're going to eventually get perfect." Well, they'll never be perfect, or at least with the models that we have today they're always going to require some human intervention. I also think AI content, for the most part, is either pretty dry, or it tries to be exciting and really falls flat. And I, I just don't think it is at a point where it can create something like engaging, unique, and interesting that's going to hold the attention of a B2B buyer without strong levels of human intervention. So, and you know, you do hear about people publishing direct from like, "Hey, yeah we just like use this tool, get it all going, we plug it into sanity and just fire it off and it's great," and I'm like, "Well, how do you know? For the first three months, but maybe not the next three." Yeah. There's a lot of confidence that it's like going great, I've seen this, I've seen this playbook before, like as someone who was pretty aggressive in link building in the pre-penguin days, to see all that traffic come down for article marketing directories, or uh, what was it like "Your name at your keywords", I think it was, where it's like blog comment links and everything and it's like, "Oh, Google just knocked that all out." And I think that whether it's Google, whether it's Bing, whether it's Brave, there's going to be some type of defense mechanism that pops up against like lower-quality AI content. But, some of the content that we've done with AI, it's, it's good, like it's, like it's objectively good content that's helpful. So, um, and that could be like location pages, like state or town pages for like a a large like business that's selling, uh, that people are looking up for like a service near me or something like that. Having that content be automatically updated with deterministic checks and uh, populated by an LLM, it can be just as good as what a human can do if you have the right guard rails, the right setup. But, um, is that interesting and engaging? No, it doesn't need to be because the product is the location service. Yeah, and I think, um, when all you have is a hammer, the the whole world looks like nails. Yeah. And I think people are trying to use LLMs in ways that they just really aren't meant to be used. Like we have a uh, a hookup with a database to be able to like chat with your data and people will ask questions about like, "Why is the data this way? What questions are not being answered by this page, or not being answered well, or aren't visible?" That the users are actually coming in and saying like, "Please tell me this." And that feedback loop is really powerful for figuring out what you need to do next. So, I'm a, I, I do think chat as a a way to interact with a website is going to become more and more popular, and we're seeing that take off a little bit with the uh, you know, like the in the corner little chatbots. Those are now starting to open up like full-on chat windows that are closer to an LLM. I don't see that just being for customer support. I see that for educational content taking off as well. My only like hesitation on that, that I that I thought about was the effort versus a good IA would you'd have the solution pages, the feature pages, if you know the voice of the customer pretty well, two clicks versus writing. Maybe it's not a detailed help center question necessarily, but like two clicks should get you there versus the act, but now we have Whisperflow and the like. So maybe we're just talking to our chatbot. Well, I mean, I think there's still going to be a reason to like go around the website, but when you're like, "Hey, do you integrate with ADP and can you, you know, help me do X, Y, and Z?" Uh, like when you have a very specific question that's not going to be answered by the content and it's going to take together piecing a few different things, I think that's like a a a pretty solid use case for AI. Now there's like the risk of like, "Well, are we going to hallucinate and tell the users the wrong thing?" And I think that's something that to be figured out and systems are going to get better over time, but I I I think people's attention spans are going down, and I think people's open time to learn and explore is going down as well, as requirements to hit numbers and goals get more aggressive. So, just being able to ask your question and get that direct answer, I think is what more people are going to want to do versus like, "Let me read five different articles and see if I can get the functionality I need." Okay, so it's almost like a retreining the user to then, eventually, never come to my website and just do all this in Claude. Well, I think, you know, Mike talks about, Mike King talks about uh, kind of the death of the open web and how things are going to get more and more limited. I don't think he's far off. I think giving out all of your data and just letting LLMs kind of become the abstraction layer that has all the customer interaction and then sending it to whatever they want instead of you getting that traffic is a real risk for some brands, especially brands that are traffic powerhouses that are losing that that steam right now. Like uh, and but, I think closing off your your entire website from LLMs is going to be a very precarious Yeah, it's a a that's a "damned if you do, damned if you don't" scenario. Um, we're having dinner with someone last night who does that and he's like, "We're still one of the top-cited things, LLMs love us," which I thought was very interesting. Very big consumer website. Yeah. Yeah, yeah, yeah. Um, yeah, it's interesting where this, this all will go, um, with those things. One of the, kind of going circling back to the AI content side, you said there's some perilous aspects of this that people aren't talking about right now. What, what were you referring to there?
JH. Scherck: You know, we've experimented a lot with AI content, and I think we've gotten better every quarter. The stuff is meaningfully better. It's still not perfect, and I think one of the bigger issues of AI content is expectations of what people can expect, and I think like let's say you, uh, set up Agent Memory and you can continuously provide feedback over time. I think people think like, "Oh, you're going to be, uh, refining and improving the outputs over time and they're going to eventually get perfect." Well, they'll never be perfect, or at least with the models that we have today they're always going to require some human intervention. I also think AI content, for the most part, is either pretty dry, or it tries to be exciting and really falls flat. And I, I just don't think it is at a point where it can create something like engaging, unique, and interesting that's going to hold the attention of a B2B buyer without strong levels of human intervention. So, and you know, you do hear about people publishing direct from like, "Hey, yeah we just like use this tool, get it all going, we plug it into sanity and just fire it off and it's great," and I'm like, "Well, how do you know? For the first three months, but maybe not the next three." Yeah. There's a lot of confidence that it's like going great, I've seen this, I've seen this playbook before, like as someone who was pretty aggressive in link building in the pre-penguin days, to see all that traffic come down for article marketing directories, or uh, what was it like "Your name at your keywords", I think it was, where it's like blog comment links and everything and it's like, "Oh, Google just knocked that all out." And I think that whether it's Google, whether it's Bing, whether it's Brave, there's going to be some type of defense mechanism that pops up against like lower-quality AI content. But, some of the content that we've done with AI, it's, it's good, like it's, like it's objectively good content that's helpful. So, um, and that could be like location pages, like state or town pages for like a a large like business that's selling, uh, that people are looking up for like a service near me or something like that. Having that content be automatically updated with deterministic checks and uh, populated by an LLM, it can be just as good as what a human can do if you have the right guard rails, the right setup. But, um, is that interesting and engaging? No, it doesn't need to be because the product is the location service. Yeah, and I think, um, when all you have is a hammer, the the whole world looks like nails. Yeah. And I think people are trying to use LLMs in ways that they just really aren't meant to be used. Like we have a uh, a hookup with a database to be able to like chat with your data and people will ask questions about like, "Why is the data this way? What questions are not being answered by this page, or not being answered well, or aren't visible?" That the users are actually coming in and saying like, "Please tell me this." And that feedback loop is really powerful for figuring out what you need to do next. So, I'm a, I, I do think chat as a a way to interact with a website is going to become more and more popular, and we're seeing that take off a little bit with the uh, you know, like the in the corner little chatbots. Those are now starting to open up like full-on chat windows that are closer to an LLM. I don't see that just being for customer support. I see that for educational content taking off as well. My only like hesitation on that, that I that I thought about was the effort versus a good IA would you'd have the solution pages, the feature pages, if you know the voice of the customer pretty well, two clicks versus writing. Maybe it's not a detailed help center question necessarily, but like two clicks should get you there versus the act, but now we have Whisperflow and the like. So maybe we're just talking to our chatbot. Well, I mean, I think there's still going to be a reason to like go around the website, but when you're like, "Hey, do you integrate with ADP and can you, you know, help me do X, Y, and Z?" Uh, like when you have a very specific question that's not going to be answered by the content and it's going to take together piecing a few different things, I think that's like a a a pretty solid use case for AI. Now there's like the risk of like, "Well, are we going to hallucinate and tell the users the wrong thing?" And I think that's something that to be figured out and systems are going to get better over time, but I I I think people's attention spans are going down, and I think people's open time to learn and explore is going down as well, as requirements to hit numbers and goals get more aggressive. So, just being able to ask your question and get that direct answer, I think is what more people are going to want to do versus like, "Let me read five different articles and see if I can get the functionality I need." Okay, so it's almost like a retreining the user to then, eventually, never come to my website and just do all this in Claude. Well, I think, you know, Mike talks about, Mike King talks about uh, kind of the death of the open web and how things are going to get more and more limited. I don't think he's far off. I think giving out all of your data and just letting LLMs kind of become the abstraction layer that has all the customer interaction and then sending it to whatever they want instead of you getting that traffic is a real risk for some brands, especially brands that are traffic powerhouses that are losing that that steam right now. Like uh, and but, I think closing off your your entire website from LLMs is going to be a very precarious Yeah, it's a a that's a "damned if you do, damned if you don't" scenario. Um, we're having dinner with someone last night who does that and he's like, "We're still one of the top-cited things, LLMs love us," which I thought was very interesting. Very big consumer website. Yeah. Yeah, yeah, yeah. Um, yeah, it's interesting where this, this all will go, um, with those things. One of the, kind of going circling back to the AI content side, you said there's some perilous aspects of this that people aren't talking about right now. What, what were you referring to there?
JH. Scherck: You know, we've experimented a lot with AI content, and I think we've gotten better every quarter. The stuff is meaningfully better. It's still not perfect, and I think one of the bigger issues of AI content is expectations of what people can expect, and I think like let's say you, uh, set up Agent Memory and you can continuously provide feedback over time. I think people think like, "Oh, you're going to be, uh, refining and improving the outputs over time and they're going to eventually get perfect." Well, they'll never be perfect, or at least with the models that we have today they're always going to require some human intervention. I also think AI content, for the most part, is either pretty dry, or it tries to be exciting and really falls flat. And I, I just don't think it is at a point where it can create something like engaging, unique, and interesting that's going to hold the attention of a B2B buyer without strong levels of human intervention. So, and you know, you do hear about people publishing direct from like, "Hey, yeah we just like use this tool, get it all going, we plug it into sanity and just fire it off and it's great," and I'm like, "Well, how do you know? For the first three months, but maybe not the next three." Yeah. There's a lot of confidence that it's like going great, I've seen this, I've seen this playbook before, like as someone who was pretty aggressive in link building in the pre-penguin days, to see all that traffic come down for article marketing directories, or uh, what was it like "Your name at your keywords", I think it was, where it's like blog comment links and everything and it's like, "Oh, Google just knocked that all out." And I think that whether it's Google, whether it's Bing, whether it's Brave, there's going to be some type of defense mechanism that pops up against like lower-quality AI content. But, some of the content that we've done with AI, it's, it's good, like it's, like it's objectively good content that's helpful. So, um, and that could be like location pages, like state or town pages for like a a large like business that's selling, uh, that people are looking up for like a service near me or something like that. Having that content be automatically updated with deterministic checks and uh, populated by an LLM, it can be just as good as what a human can do if you have the right guard rails, the right setup. But, um, is that interesting and engaging? No, it doesn't need to be because the product is the location service. Yeah, and I think, um, when all you have is a hammer, the the whole world looks like nails. Yeah. And I think people are trying to use LLMs in ways that they just really aren't meant to be used. Like we have a uh, a hookup with a database to be able to like chat with your data and people will ask questions about like, "Why is the data this way? What questions are not being answered by this page, or not being answered well, or aren't visible?" That the users are actually coming in and saying like, "Please tell me this." And that feedback loop is really powerful for figuring out what you need to do next. So, I'm a, I, I do think chat as a a way to interact with a website is going to become more and more popular, and we're seeing that take off a little bit with the uh, you know, like the in the corner little chatbots. Those are now starting to open up like full-on chat windows that are closer to an LLM. I don't see that just being for customer support. I see that for educational content taking off as well. My only like hesitation on that, that I that I thought about was the effort versus a good IA would you'd have the solution pages, the feature pages, if you know the voice of the customer pretty well, two clicks versus writing. Maybe it's not a detailed help center question necessarily, but like two clicks should get you there versus the act, but now we have Whisperflow and the like. So maybe we're just talking to our chatbot. Well, I mean, I think there's still going to be a reason to like go around the website, but when you're like, "Hey, do you integrate with ADP and can you, you know, help me do X, Y, and Z?" Uh, like when you have a very specific question that's not going to be answered by the content and it's going to take together piecing a few different things, I think that's like a a a pretty solid use case for AI. Now there's like the risk of like, "Well, are we going to hallucinate and tell the users the wrong thing?" And I think that's something that to be figured out and systems are going to get better over time, but I I I think people's attention spans are going down, and I think people's open time to learn and explore is going down as well, as requirements to hit numbers and goals get more aggressive. So, just being able to ask your question and get that direct answer, I think is what more people are going to want to do versus like, "Let me read five different articles and see if I can get the functionality I need." Okay, so it's almost like a retreining the user to then, eventually, never come to my website and just do all this in Claude. Well, I think, you know, Mike talks about, Mike King talks about uh, kind of the death of the open web and how things are going to get more and more limited. I don't think he's far off. I think giving out all of your data and just letting LLMs kind of become the abstraction layer that has all the customer interaction and then sending it to whatever they want instead of you getting that traffic is a real risk for some brands, especially brands that are traffic powerhouses that are losing that that steam right now. Like uh, and but, I think closing off your your entire website from LLMs is going to be a very precarious Yeah, it's a a that's a "damned if you do, damned if you don't" scenario. Um, we're having dinner with someone last night who does that and he's like, "We're still one of the top-cited things, LLMs love us," which I thought was very interesting. Very big consumer website. Yeah. Yeah, yeah, yeah. Um, yeah, it's interesting where this, this all will go, um, with those things. One of the, kind of going circling back to the AI content side, you said there's some perilous aspects of this that people aren't talking about right now. What, what were you referring to there?
JH. Scherck: You know, we've experimented a lot with AI content, and I think we've gotten better every quarter. The stuff is meaningfully better. It's still not perfect, and I think one of the bigger issues of AI content is expectations of what people can expect, and I think like let's say you, uh, set up Agent Memory and you can continuously provide feedback over time. I think people think like, "Oh, you're going to be, uh, refining and improving the outputs over time and they're going to eventually get perfect." Well, they'll never be perfect, or at least with the models that we have today they're always going to require some human intervention. I also think AI content, for the most part, is either pretty dry, or it tries to be exciting and really falls flat. And I, I just don't think it is at a point where it can create something like engaging, unique, and interesting that's going to hold the attention of a B2B buyer without strong levels of human intervention. So, and you know, you do hear about people publishing direct from like, "Hey, yeah we just like use this tool, get it all going, we plug it into sanity and just fire it off and it's great," and I'm like, "Well, how do you know? For the first three months, but maybe not the next three." Yeah. There's a lot of confidence that it's like going great, I've seen this, I've seen this playbook before, like as someone who was pretty aggressive in link building in the pre-penguin days, to see all that traffic come down for article marketing directories, or uh, what was it like "Your name at your keywords", I think it was, where it's like blog comment links and everything and it's like, "Oh, Google just knocked that all out." And I think that whether it's Google, whether it's Bing, whether it's Brave, there's going to be some type of defense mechanism that pops up against like lower-quality AI content. But, some of the content that we've done with AI, it's, it's good, like it's, like it's objectively good content that's helpful. So, um, and that could be like location pages, like state or town pages for like a a large like business that's selling, uh, that people are looking up for like a service near me or something like that. Having that content be automatically updated with deterministic checks and uh, populated by an LLM, it can be just as good as what a human can do if you have the right guard rails, the right setup. But, um, is that interesting and engaging? No, it doesn't need to be because the product is the location service. Yeah, and I think, um, when all you have is a hammer, the the whole world looks like nails. Yeah. And I think people are trying to use LLMs in ways that they just really aren't meant to be used. Like we have a uh, a hookup with a database to be able to like chat with your data and people will ask questions about like, "Why is the data this way? What questions are not being answered by this page, or not being answered well, or aren't visible?" That the users are actually coming in and saying like, "Please tell me this." And that feedback loop is really powerful for figuring out what you need to do next. So, I'm a, I, I do think chat as a a way to interact with a website is going to become more and more popular, and we're seeing that take off a little bit with the uh, you know, like the in the corner little chatbots. Those are now starting to open up like full-on chat windows that are closer to an LLM. I don't see that just being for customer support. I see that for educational content taking off as well. My only like hesitation on that, that I that I thought about was the effort versus a good IA would you'd have the solution pages, the feature pages, if you know the voice of the customer pretty well, two clicks versus writing. Maybe it's not a detailed help center question necessarily, but like two clicks should get you there versus the act, but now we have Whisperflow and the like. So maybe we're just talking to our chatbot. Well, I mean, I think there's still going to be a reason to like go around the website, but when you're like, "Hey, do you integrate with ADP and can you, you know, help me do X, Y, and Z?" Uh, like when you have a very specific question that's not going to be answered by the content and it's going to take together piecing a few different things, I think that's like a a a pretty solid use case for AI. Now there's like the risk of like, "Well, are we going to hallucinate and tell the users the wrong thing?" And I think that's something that to be figured out and systems are going to get better over time, but I I I think people's attention spans are going down, and I think people's open time to learn and explore is going down as well, as requirements to hit numbers and goals get more aggressive. So, just being able to ask your question and get that direct answer, I think is what more people are going to want to do versus like, "Let me read five different articles and see if I can get the functionality I need." Okay, so it's almost like a retreining the user to then, eventually, never come to my website and just do all this in Claude. Well, I think, you know, Mike talks about, Mike King talks about uh, kind of the death of the open web and how things are going to get more and more limited. I don't think he's far off. I think giving out all of your data and just letting LLMs kind of become the abstraction layer that has all the customer interaction and then sending it to whatever they want instead of you getting that traffic is a real risk for some brands, especially brands that are traffic powerhouses that are losing that that steam right now. Like uh, and but, I think closing off your your entire website from LLMs is going to be a very precarious Yeah, it's a a that's a "damned if you do, damned if you don't" scenario. Um, we're having dinner with someone last night who does that and he's like, "We're still one of the top-cited things, LLMs love us," which I thought was very interesting. Very big consumer website. Yeah. Yeah, yeah, yeah. Um, yeah, it's interesting where this, this all will go, um, with those things. One of the, kind of going circling back to the AI content side, you said there's some perilous aspects of this that people aren't talking about right now. What, what were you referring to there?
JH. Scherck: You know, we've experimented a lot with AI content, and I think we've gotten better every quarter. The stuff is meaningfully better. It's still not perfect, and I think one of the bigger issues of AI content is expectations of what people can expect, and I think like let's say you, uh, set up Agent Memory and you can continuously provide feedback over time. I think people think like, "Oh, you're going to be, uh, refining and improving the outputs over time and they're going to eventually get perfect." Well, they'll never be perfect, or at least with the models that we have today they're always going to require some human intervention. I also think AI content, for the most part, is either pretty dry, or it tries to be exciting and really falls flat. And I, I just don't think it is at a point where it can create something like engaging, unique, and interesting that's going to hold the attention of a B2B buyer without strong levels of human intervention. So, and you know, you do hear about people publishing direct from like, "Hey, yeah we just like use this tool, get it all going, we plug it into sanity and just fire it off and it's great," and I'm like, "Well, how do you know? For the first three months, but maybe not the next three." Yeah. There's a lot of confidence that it's like going great, I've seen this, I've seen this playbook before, like as someone who was pretty aggressive in link building in the pre-penguin days, to see all that traffic come down for article marketing directories, or uh, what was it like "Your name at your keywords", I think it was, where it's like blog comment links and everything and it's like, "Oh, Google just knocked that all out." And I think that whether it's Google, whether it's Bing, whether it's Brave, there's going to be some type of defense mechanism that pops up against like lower-quality AI content. But, some of the content that we've done with AI, it's, it's good, like it's, like it's objectively good content that's helpful. So, um, and that could be like location pages, like state or town pages for like a a large like business that's selling, uh, that people are looking up for like a service near me or something like that. Having that content be automatically updated with deterministic checks and uh, populated by an LLM, it can be just as good as what a human can do if you have the right guard rails, the right setup. But, um, is that interesting and engaging? No, it doesn't need to be because the product is the location service. Yeah, and I think, um, when all you have is a hammer, the the whole world looks like nails. Yeah. And I think people are trying to use LLMs in ways that they just really aren't meant to be used. Like we have a uh, a hookup with a database to be able to like chat with your data and people will ask questions about like, "Why is the data this way? What questions are not being answered by this page, or not being answered well, or aren't visible?" That the users are actually coming in and saying like, "Please tell me this." And that feedback loop is really powerful for figuring out what you need to do next. So, I'm a, I, I do think chat as a a way to interact with a website is going to become more and more popular, and we're seeing that take off a little bit with the uh, you know, like the in the corner little chatbots. Those are now starting to open up like full-on chat windows that are closer to an LLM. I don't see that just being for customer support. I see that for educational content taking off as well. My only like hesitation on that, that I that I thought about was the effort versus a good IA would you'd have the solution pages, the feature pages, if you know the voice of the customer pretty well, two clicks versus writing. Maybe it's not a detailed help center question necessarily, but like two clicks should get you there versus the act, but now we have Whisperflow and the like. So maybe we're just talking to our chatbot. Well, I mean, I think there's still going to be a reason to like go around the website, but when you're like, "Hey, do you integrate with ADP and can you, you know, help me do X, Y, and Z?" Uh, like when you have a very specific question that's not going to be answered by the content and it's going
Video