Loading summary
Transcript79 lines
- [00:00]
Robert
Foreign.
- [00:05]
Alessio
Welcome back to the Latent Space podcast. This is Alessio, partner and CTO at Decibel. And I'm joined by Swix, founder of Small AI.
- [00:12]
Swix
Hello. Hello. Today we're actually diving into a topic that I've wanted to dive into for a while. AI search engines have caused a lot of the sort of rising opposition force of AI search engine optimization. And today we have Robert from Scrunchai.
- [00:26]
Robert
Welcome. Thank you. Yeah, great to be here.
- [00:29]
Swix
Alessio. I think you are probably the best place to introduce Robert.
- [00:33]
Alessio
Robert, as I guess disclosure. Decibel invested in the round that we're going to announce with this podcast. So congrats, Robert, and I guess congrats to us for being picked by you. You were previously the CTO of Hearsay, which was started by Clara. She, who's also a friend of the podcast. And my partner John was on the board of Hearsay. So you guys knew each other from there. So we kind of go back. And even back then you were already working on the space of I have a business. How do I reach people in a way that is very tailored and that was more on the financial services side. And then you started Scrunch a couple years ago, so you were pretty early to the space. And then now everything is blowing up, everybody's getting traffic from ChatGPT. So maybe talk a bit about when you first realized that LLMs and search was not just normal search and maybe the idea maze that got you there and we can take it from there.
- [01:29]
Robert
Yeah, absolutely. That's a. That's a great starting question. And the origin story. Hopefully I don't get in trouble for telling this on the pod. The origin story is a little bit convoluted, as I think it is for most startups. With Scrunch, my co founder Chris and I, we had left Hearsay. Chris was actually also a longtime executive at Hearsay. He was the chief product officer for a long time. He's the first employee. So we both know all those folks really, really well. We were looking for essentially what to do next in our careers. You know, thinking about starting a company and kicking around ideas, talking to people we knew in the industry. And as we were doing that, the elephant in the room was AI Right. Like, it just became really clear I think, a couple of years ago that, like, it would be sort of foolish to start a company without thinking about what the impact of, you know, LLMs would be on the world and you'd already be behind if you sort of started a company that wasn't in Some way LLM native. So we went through an exploration process there and as we were talking to like the people we knew the best, who are often like, you know, sort of enterprise like large banks and insurance companies, that was a primary kind of audience for hearsay. It was a tough sell back then to sell them anything related to AI. And the number one thing that people were probably asking us for, quite honestly was like, can you put a chatbot on my website? Right. Like I'd like to have a little widget. I'd like to have my version of ChatGPT. I want people to come to my website and a widget will pop up and then people can type into it and like ask my website questions instead of going to ChatGPT. And there are good versions of that now, the versions that were out at that time of that functionality, I'd say we're mostly not so great. But basically my core feeling about that at the time was like, nobody wants to use a chatbot on your website. I don't want to use chatbots on people's website. It's like if things pop up on a website when I go to visit it, but I just like, I fly into a visceral rage. I'm definitely closing the window if not closing the tab. So we didn't want to do that.
- [03:20]
Swix
I have that feeling. But also I wonder if the normies respond fine to it. You know, like Intercom is doing fantastic as a business, right? So I have a self doubt about how representative we are.
- [03:31]
Robert
That's a fair question, right? And I always ask myself that as well as maybe kind of like a weird technologist person, you know, I think there's some evidence that people don't love it. I think Intercom's a little bit different because by the time you're engaging with like fin or what have you, hopefully you're in a relationship to a certain extent with the business. So I do think that people have a little bit more tolerance for using these AI tools, like when they're already kind of feeling some rapport with a website or with a business, they're more willing to like use your tools instead of saying instead of staying in the environment they like to use. But if you're just coming to a website and it pops up like people mostly I think don't want to do that. People have responded really positively to ChatGPT. Most people I've talked to and I'm by the way, I live in la. I've lived in LA for a couple of years since Pandemic So I'd say, like, more of my social circle is normies than it used to be. Like, people love ChatGPT. The experience people have of using ChatGPT versus, like browsing the Internet, sort of the old Google 10 blue links and clicking around is like vastly more positive compared to the past. So people really feel, I think, like an affinity for the tool and they want to use the tools they like. They want to use, you know, whether it's ChatGPT or it's Claude or Puxity or what have you. Like, they really like that environment. It's convenient, it's productive. They don't have to browse websites that maybe aren't well designed for them. People are not excited about the idea of like, okay, let me click around, find a website and then engage with the chatbot there. In the same way that I think people aren't like, let me go to the website and use their search bar. Right? People just want to use Google. I do think that's, that's true. And when we have that realization of like, okay, people are asking us for this thing, but we don't really want to build it, we don't necessarily think that's like the durable value. We started asking ourselves, you know, what is it? What are they trying to get to here? And it came down, I think, to discovery and basically this realization that like AI was going to change something about their website, it was going to change something about their customer journey and how people sort of interacted with their business, but like not knowing exactly what it was. And as we started showing people as ChatGPT search was coming out, like how ChatGPT search was referencing their business, like how it was surfacing their content or like what it was saying about their brand. We just had the experience of like having a lot of folks, you know, folks who are like CMOs at like large insurance companies or enterprise software companies or large, you know, e commerce marketplaces. That's really scary. Like, it's saying stuff about my business, it's saying stuff about my brand, it's surfacing content from my website and like, I don't really know what's happening here. And so that's sort of the genesis of how we got into this business. More than focusing on like SEO and performance marketing specifically. More like there's just a change in the way consumers kind of interact with the Internet. And I'd say that's still true today. Like, that's still sort of the core thesis of the business is like the consumer experience of the Internet. Is changing really, really quickly for the first time in like quite a long time. Yeah, so I'd stop, maybe I'll stop there. But that's sort of the origin story.
- [06:24]
Alessio
Yeah. And I know, as you know, most people here are technical in the audience, so it would be nice to dive into that. I guess. Everything starts with a prompt in your world, which is what the consumer asks ChatGPT about some sort of product or market or whatever. And then how are you guys doing the monitoring? Are you monitoring every LLM, are you monitoring every model plus search? Are you monitoring the AI native tools like exsign, things like that, maybe talk people through the pieces of the stack here?
- [06:53]
Robert
Yeah. So what we do starts with what you said, which is we monitor prompts. Basically we go out and we sort of simulate consumer interactions with the major AI platforms. So we kind of go just by basically like usage. So ChatGPT obviously is among the highest usage AI overviews could be considered up there as well, if you consider that in the same category, AI mode perplexity. And you can kind of go down the list from there. So we definitely don't cover everything. I'd say our ambition is ultimately like to sort of be wherever people are, wherever the consumer is, wherever our customers audiences are. But today we primarily focus on sort of the biggies. So ChatGPT, whether it's search or trained in model knowledge, Gemini, Perplexity, Claude, which is important for, you know, not necessarily huge in terms of raw number of users, but the people who do use Claude tend to be like a very valuable audience, especially for some types of company. And then also things like meta AI for example, which I don't think gets necessarily remarked upon a ton in the like AI enthusiast AI developer community. But like it's a silent monster, right. Because of Meta's distribution and reach. So it's expanding all the time. You know, we'll have more platforms live like probably by the time this podcast comes out. But we sort of start by focusing on like what are kind of, you said it before, what are the normies using, what are they doing, what are they seeing in these tools?
- [08:09]
Alessio
So now that you have searched, they have the backlinks, but when you don't have search on, there's no way to figure out why a model saying something is that a battle that basically people cannot fight? Like there's something in the pre training, is there something that you advise people to do or you just flag it and that's it.
- [08:26]
Robert
It's definitely less actionable. Right. And I think that the Good thing for brands is like, generally speaking, by the way, when I say brands like throughout this conversation, that's just like our term of art. But I just mean like businesses, right? Businesses that have products and services and like websites and want you to do stuff on the Internet to make money. I think it is less actionable. So most people I think are focused more squarely on AI search. So chatgpt with a search box ticked. The good news is I think most of the queries that lead to action that are sort of of commercial interest are increasingly using search across these systems, like for a bunch of reasons. Right. Timeliness and avoiding hallucinations and so on. So I'd say like most people are focused on that there are things you can do even in pre training mode. And an example of that is just thinking about like what are you exposing to these models when they do come and crawl your website? Right. Everybody knows that GPT bot and CC bot and all these various AI training data collectors are hitting tons of websites on the Internet. Some people are fine with that, some people are upset about it, but they're identifiable and you can think about the half life of the information you're presenting and whether or not that's something that really makes sense to give to a training data crawler. So for example, like if you're an E commerce company and you're running a seasonal sale, should that go into the trained knowledge of a model? I would argue it's probably counterproductive. Right. So maybe you don't want to expose information about that to those crawlers hitting your website. And that's something we advise our customers on.
- [09:55]
Swix
While we're here on the topic, any quick takes on what cloudflare did last week, which was apparently start introducing paywalls for all these bots?
- [10:05]
Robert
Yeah, right. I mean it's really interesting. I kind of agree with the sentiment I think behind it. Right. Like, I mean obviously it's tough to be, it's tough to be a writer on the Internet. It's tough to be like a content publisher on the Internet and it's getting tougher and tougher I think over time, whether it's ChatGPT or it's Google AI mode. So the idea that where there's some sort of maybe like grand bargain we need to make as a society to compensate content providers, which I think is kind of what Matthew Prince is getting at with his blog post, does make some sense to me. I think there's two things to think about there and one is technical mechanism, which is how does this paywall work. And I would just say the bridge doesn't connect on both sides yet. Maybe we'll get there. So I don't think there's a lot of actual use of those systems. And there's prior art. Right. There's things like tolbit, for example, that has some similar ideas. I would say there's some uptake in traditional media. It's not a huge piece of infrastructure on the Internet yet. And then the second thing I think is about power dynamics, which is like, who wears the pants between website operators and CDNs and AI in particular. AI search or search engines in general. Right. Because websites obviously are sensitive to how their content's being used, but they also need traffic, and traffic, by and large comes from Google is increasingly coming from ChatGPT, as we've seen. There's a fine line, I think, for people to walk where unless you've really got a strong native, sort of like organic audience of people who, like, know you, love you, like, have a relationship with you, which is still like one of the most important things you can do on the Internet, you need to be careful about how restrictive you are with your content, because distributing your content through these systems is. Is how you're going to get users. So I don't know the right answer, but I think it's. It's tricky.
- [11:51]
Swix
So I find it a really interesting divide in society between, like, some content creators want their everything behind a paywall, want to be paid for every single use, and in others, work very, very hard to make everything free so that, you know, like, train on me. Right? Like, and it's. If anyone cares about our position as content creators, everything in the inspace is free. We have some, like, sort of soft gates, but it's not really actually something that we care about. Like your content. Content wants to be free. The cost of reproduction is zero. Like, just get it out there, right? And actually, you know, maybe the ChatGPT will start referring more people to you and you'll be more of an authority in your. In your space. All that good stuff.
- [12:29]
Robert
That's very real. And I think one thing I would just mention here is they also have the AI browser announcements this week. There's Dia, there's Comet, there's the Untitled OpenAI browser. And I think that's a really interesting addition to this mix too, because what I kind of think about is you sort of just said it, but it's that 90s phrase of information wants to be free and people want to connect models to the Internet. And so the Way people are using these today is going through ChatGPT or something which feels like, you know, Google is an input box in the middle of a pretty Spartan webpage and ChatGPT is like an input box in the middle of a pretty spartan web page. So they feel pretty similar. They're both sort of like these Internet platforms, Internet gatekeepers and people think about them the same way. But like there's no reason that has to be the way you consume Internet content and get it into these models. And I think like the AI browser form factor is like a really interesting evolution there, there's gotta be others, right? There could be like local tools, local models. And what I would just say, right is you can fight the battle. I think of saying OpenAI shouldn't consume your content without paying for it or shouldn't consume it at all. But I think it's going to be really tough to fight the battle of saying LLMs writ large shouldn't consume my content because it's just not that hard to hook these things up to Internet content to some degree. And once it's in somebody's browser, how are you going to stop them? That's a whole different mechanism. Some people I think are thinking about this change. When it comes to AI, search is sort of one platform replacing another. So instead of Google it's OpenAI or maybe it's Google disrupting itself. But I think there's sort of a more fundamental change which is like people just have more powerful tools, have these AI tools right, to consume web content. And I think what you're seeing with people who have access to the tools is like they don't want to browse websites the same way they used to in the past, right? So like I actually think like this is probably like a medium spicy take, but it's like thinking about what's going on in AI that's really replacing search is the wrong take or maybe not that interesting. It's really more like AI is replacing web browsing. I think that's actually like a more meaningful like more fundamental shift. And the reason it's happening I think is because people just like it better, right? It's not because of some like top down platform mandate. Like people like using these tools to replace what they were doing before by the clicking around the Internet and consuming content. Not like for every single use case, but for a lot of things that people are using websites today for that they prefer to use, you know, an.
- [14:51]
Alessio
Agent well and deep research probably being the biggest example of that. On the content side it's interesting because. So Lenny, who runs a very big, you know, product newsletter, he posted his stats from May 10 to last week and ChatGPT drove more traffic than Twitter for him, which I'm surprised because he has 250,000 followers on Twitter and he only had 9,000 views from it. So for us, Twitter is much higher than ChatGPT, even though we do get a good amount of them. But I'm curious how much you want to tell the models about who you are and not as much about the content. You kind of want to be present in the curation, but not in the details, and then appeal in the details for you. This is something that people have been debating on shopping, which is when I'm using Chat chatgpt to find something to buy, how much of the information should I put in the webpage so that it gets put in the context versus on podcasts? If I'm searching for the best AI podcast or whatever, how much does it need to know before it suggests me? Or is the model simply just using the same ranking as the SEO that Google API, Bing API are using?
- [16:07]
Robert
So I mean, this gets into the mechanism of it, right? And what I would say is AI search is still fundamentally search, right? So there is still a search ranking. And by the way, to the best of my knowledge, which I think is pretty good, most of the search that's in AI search is still traditional search, right? There's an AI model in front of it, but you're still using your sort of traditional, like, you know, text based, like BM25 or TF IDF or PageRank or what have you search algorithms under the hood to locate pages in a search index. And that is kind of a fundamental component of all these AI search systems. So far, I know people like EXA are doing something different and I think that's really exciting. But just like that's not what's powering most of the products that people are using today. So traditional search ranking does still matter and traditional SEO techniques do still work. Now where it gets tricky after that, right, is like you're not just getting links for discovery, you're also consuming the content behind the links. And how much of that are you giving away? What does it say? How well is the system able to interpret it? And so the way we think about this for most models, most platforms, and the details vary a little bit between ChatGPT versus Claude search or something like that, is you're generating queries, you can generate one query, you could generate multiple. If you're looking at AI mode or some of the more recent O3 search integrations that ChatGPT has. So you're generating keywords using an LLC, right? You're running searches in parallel or potentially serially. If you're using a reasoning model, they're going out and retrieving results. And then there is sort of re ranking that happens once you've got those raw search results. So the raw search results are again traditional search. But then what you're doing after that is it's re ranking them for relevance and then it is only then starting to look at the actual content behind the pages. So this is where we see people I think get tripped up a lot in terms of small things that go wrong in terms of getting cited or Getting mentioned inside ChatGPT is like people have these pages that rank really well for search and then when you look at the title of the page, like you don't look at the page content, you just sort of look at the metadata about the page. The metadata has been super optimized for human click through, right? So think about like clickbait. Think about things that are like, you know, like creating urgency, especially true on E commerce right around sales and things like that. But they're not very informational about what actually is on the page. And then what we see is like regardless of search ranking, often like a system like ChatGPT chooses not to use those, right? So it will re rank them out of the consideration set and then it'll just kind of go on to the next best result that has a more descriptive metadata saying hey, this page actually has something that's relevant to your question on it. That sort of re ranking stuff is critical. And you know, since this is a technical podcast, they'll say like, you know, I'm not saying re rank or model, I just mean there is a more intelligent step of sort of reordering what it looks at coming out of the traditional search index before you get to the part where it's actually consuming content and like summarizing and generating. That's I think like an important technical mechanism to understand in terms of like the strategy of like how much to expose. I think it varies based on what you want to do. For shopping, I would say typically they're going to buy the product from somewhere. If you can get a link to your page and your page has a checkout on it, that's what you want to achieve. Obviously that's starting to change with like ChatGPT shopping and some of the more agentic shopping integrations we're seeing in terms of being able to buy right within the AI experience. But I would still err on the side of disclosure. Right. And I especially err on the side of in your content talking more about like who the product is for and what it's good for and being descriptive about it. A lot of shopping pages are like really highly visual, for example, so they're, they're tuned for people who are like looking visually at the webpage and being like, I want to buy that. And that still works because it's still super important if you're like buying f fashion, for example. Right. That's just inherently a visual aesthetic experience. But even within like apparel, we see that like just being a little bit clearer about like here's why, like here's who buys these type of leggings and for what can make a big difference in getting surfaced in ChatGPT and then not only getting surfaced, but actually getting people to click through and ultimately purchase the product from that source. So our general stance is disclose more.
- [20:15]
Alessio
Yeah, I was going to say that makes sense. And I think there's a lot of people that do things like I need to buy a gift for my friend that loves to hike, that lives in LA and blah, blah, blah, blah, blah. So I'm curious if that's going to be visible, if that's going to be again what you guys want to do later too. It's just like help rewrite these things. My idea. And we haven't really talked about this, even offline. Do you get the query that the ChatGPT used to find your page when they come to your page, like, is there some sort of way to understanding what brought them there? And then can you real time rewrite the page to fit that query better?
- [20:55]
Robert
Yeah, I mean people really want that to happen. It doesn't happen today. Right. And I think there's a, there's a bunch of reasons why that's the case. The reason I think it probably won't be the case in the future is for privacy reasons. So I don't think that you will see, at least in the context of the way search is working today inside like a system like ChatGPT, I don't think you'll see that it's going to pass like the prompt the user is using to the website. Right. That it's getting content from. Because like, if you think about how people use ChatGPT, people who get used to using ChatGPT start putting like all sorts of like protected health information and like crazy personal stuff into it. And that's one of the reasons it works so well. But obviously you don't want to be passing that to, you know, websites you're finding through search on the Internet. So right now the answer is, like, people really get almost no information about what led somebody from chatbeat to their website. But I mean, that being said, I think if you look at where people are going to, from ChatGPT to your website, you can get a lot of signal there. Right. People are coming to certain product categories. If they're coming to certain blog articles, you can get a sense of like, okay, what are people interested in who are using these tools? And especially if you have a good feeling, if you're. If you have a good feel for your audience, I think it can be really powerful. We have customers, for example, who have, like, seen that people arrive on certain topics. Like, these are like developer infrastructure companies. They see that people arrive from ChatGPT on certain blog posts about certain, you know, certain problems they're trying to solve. And they're like, clearly there are people trying to solve this problem on ChatGPT because this was surfaced and it came up, let's write more content about that. And it's been a really effective strategy. Right. They're just realizing that there is demand, even though they're sort of indirectly detecting it, because ChatGPT is sending people their way, and then they're just creating additional supply for that demand. Yeah.
- [22:44]
Swix
I think Alessia was very struck by. There was an example this week on Hacker News, I think, where ChatGPT hallucinated a feature that they didn't have. Then they were like, screw it, let's build it. Because they could be.
- [22:56]
Robert
I saw that. I think I saw that one come through. Yeah. I mean, it does that all the time, right? I mean, I think that if you're. If you're like a cursor enthusiast, I think many of us have had the experience, for example, of being like, this thing keeps hallucinating methods in my code. I probably should just implement the methods so that, you know, I can stop having to, like, try to prompt engineer the model not to do that. And there's sort of a similar vibe there. Right. But, you know, I think that's also an example of. My guess is without having looked a ton at the actual example of that particular company, that's a case where doing some sort of context engineering, doing some engineering of, like, the actual content they had on their websites, on their pages, might. How to improve that problem, maybe it would have been a mistake to improve it. Because they might have like lost this funnel of users who were like, yeah, we're really interested if you have this feature. But one thing I'll say about startups, and we've done like a lot of reviews with like Y Combinator startups, for example, is like startups are really, really bad at describing what they do on their homepage. You know, the number of homepages where you're like, there's a really cool like parallax scroll animation and you're like, what does this company actually do is super high. And I think that's tricky if you're trying to make the best use of these tools.
- [24:04]
Alessio
Is there a metagame already of like this AI SEO algorithm? I remember back in the day there was like, oh, the Google Penguin algorithm update or like all these different things that they were doing. Is there something similar happening? Or there's kind of like the ground truth of the links which are the search engines. But then on the RE ranking side are people tracking how the different LLMs do the re ranking and go through it or that it's just a very native space still?
- [24:32]
Robert
Yeah, I mean, I think that people are definitely doing. People are paying very close attention to the specifics of what the different AI platforms are doing in terms of like ranking and what kind of content they consume and like testing all sorts of experiments for what works best. And I'd say like, our role in terms of what we do at Scrunch, right, is like we do a couple of different things, but one of the things we do is like we give people a feedback loop to understand how what they're doing is changing their performance. And so we've had some great case studies of people doing things like that that are like pretty in the weeds and getting good performance boosts out of it in terms of like some of the more gray area SEO tactics like that, you know, Penguin was addressing with Google back in the day. I mean, I think the reality is a lot of those things do work today. Like, I can't tell you they don't work, but I would say is like, you know, as a company I wouldn't encourage anybody to do them because I think it'll change eventually. And one thing you mentioned this, but like one thing that's definitely true about all the AI search companies is like, everything's super early. 24 months ago, ChatGPT was basically like, you know, ChatGPT was the original ChatGPT wrapper. It was just a wrapper around like GPT 3.5 and obviously it's gotten way more sophisticated over the past year or two. But none of these products have as much attention to detail or engineering put into them in terms of, like, moderation and abuse detection and like being defensive against the fact that the Internet's a wild place as Google does. Right. And Google's been around for 25 plus years, so no surprise there. A lot of things that work the way they work in AI search today, I would definitely say work that way sort of by accident. And so that's part of the challenge, I think, of being in the space, whether you were us or one of our customers that's trying to figure out how to make this work is like everything changes super quickly. Of course, that's always true in AI, but it's not just about, like models getting better. It's really about, like all of the underlying kind of glue that's connecting these models to search and to the Internet. Just like being pretty rudimentary and like getting more sophisticated over time. I don't know if that answers your question, but that's the way I think about it.
- [26:36]
Alessio
What's the name of the category in your mind? So there's AI, SEO aeo, people I come up with, then the Geo Generative engine optimization that a16z posted about. Do you think any of these are winners? Do you think we need something new?
- [26:51]
Robert
I'm terrible at naming things, so I'm the last person who should name the category. I would say people ask us, like, are we a geo company? And I say yes, right, because you can only fight so many battles. But again, I mean, I think if you think about people have different definitions of what SEO means, but I think most common understanding, right, is it's basically about like showing up higher in search results. It's about showing up more frequently. And that is an important part of optimizing how your business and your products and services perform in these tools like ChatGPT. But again, I think that's only like the tip of the spear because it's not just replacing discovery, it's actually replacing how people actually consume content about your business and increasingly how they interact with your business. It's not just like the entry point, it's more of the full journey of how somebody engages with you or engages with your website. And so I think like SEO aeo, Geo, that's generally been more focused on that, like, kind of entry point. And the real prize is, again, I think more and more traditional web browsing is going away. Like, how do you optimize the complete customer experience for a world where like most people are doing most things in a tool like ChatGPT instead of browsing around your website. So I don't, I don't know what a good category name for that is, but I think that's the one we're in. The way we think about it is like agent experience. Right? So like by analogy to the customer experience, there's tons and tons of technology around making sure that customers have a good experience when they do come to your website and like go down funnels and like have a high NPS score and ultimately convert and are happy with the service they get. You know, I think it's going to be really important to be that data driven about how these like AI systems are interacting with your content, with your website, with your infrastructure, because ultimately they are serving your ultimate customer, serving the user. And you have to do a good job with the AI systems if you want to do a good job for the user. So maybe agent experience is what I would call it.
- [28:47]
Alessio
Let's go forward to a feature where like Sam Altman says ChatGPT is kind of like the all encompassing personal assistant. I have all my memories, I have all my preferences. Do you see that playing a big part and kind of like how the re ranking and the selection of these websites gets done? Is it used at all today?
- [29:05]
Robert
It definitely is. Right. Like so chatgpt personalization memories, just explicit preferences, if you set them up, do definitely affect the results you get. Now again, obviously they're still using traditional search, so the search index itself is not necessarily personalized, but you can definitely tell ChatGPT. Right? Like I don't want to read anything from these sources and it will try to obey. That can affect both like what it's searching for and also which sources it prefers and ultimately how it presents information from those sources to you. Right. I mean I have the classic set of ChatGPT instructions in my ChatGPT config, which is like I'm a senior engineer, like be concise, like don't over explain things and don't glaze me too much. And I certainly get different results using AI search with that profile that I do if I'm like in an incognito window with like the default consumer experience of ChatGPT. So I think like again, understanding your audience matters a lot. Ultimately maybe where this ends up is like no single chatgpt. Every single person kind of has their own, right, their own fully personalized experience. And I think if you're like a marketer who's like, how do I Measure this stuff, that's kind of a nightmare. But I just think that's the reality we might be living in. So like again, understanding like your audience, understanding your Personas, like your ideal customer Persona, thinking about how they want to use these tools, like what they're trying to accomplish and sort of modeling that out and measuring it is like a really important thing you should start doing. And that's something again that we try to help with in terms of what we're doing with monitoring and scrunch. Not just like monitoring prompts, but trying to group them into like customer Personas and actually monitoring the complete experience. For somebody who you know is a senior engineer, is a product manager, is whatever, ICP kind of makes sense for your business.
- [30:45]
Swix
Speaking of clusters, I'm always curious to mine for data. You don't have to talk in specifics, but what are the major clusters? Anything in there that surprises people?
- [30:55]
Robert
Clusters in terms of prompts, usage, things.
- [30:59]
Swix
Your, things your customers really care about. For example, shopping big cluster. But if you don't use ChatGPT for shopping, you don't really care. Coding big cluster. Again, if you're not a coder, you don't care.
- [31:10]
Robert
What else is like that in terms of what we see as important? Like there's, there's clusters of how people use LLMs that are definitely a little, little bit less commercially relevant. So for example, there's, there are a lot of people who are very high volume Users of ChatGPT who are like doing role playing, for example. And then there's some other systems that have even more of that. And if you're us, that's maybe a little bit less interesting. The biggest thing I would say is like if you expand from coding a little bit, what I would really say is it's about problem solving. I don't know if you guys have seen this, but there was a study that's gone around, there's been like a couple versions of it around. Like what is basically the prompt intent of what people type into ChatGPT, right? Are they using it to find information? Is it navigational? Which doesn't really make sense. Was it generative? Like they're trying to create some text or create an artifact? Are they doing research? And you know, they mined all this data from like, like panel data, like clickstream data of like a bunch of consumers who had opted into having their data collected, which is like the classically this stuff happens in marketing research. One of the takeaways that the study had was like people use ChatGPT a lot for these sort of generative tasks, right? They're trying to accomplish something. They're not necessarily trying, they're not necessarily trying to do a search in the sense that somebody's, you know, trying to go to Google and search for a topic. And their takeaway was like, maybe Google safe people are not necessarily using ChatGPT that much for these search tasks. So Google's still the king for search. You don't have to worry like Google, you know, nobody is moving your cheese too much. But actually what I would say is when you look deeper and like study how people use the tools, we've done some like qualitative studies and things like that. And then we have, you know, we have the same data everybody else has as well. People go to ChatGPT, especially in a business context, to solve problems. They're like, I need help with something, I need to accomplish something. I need to like write this proposal. I need to come up with this financial model, like for coders. It's really like, it's really, really potent. You know, if you, if you've used AI coding tools, like they write the code for you, you get a solution, you can run the solution, like, problem solved, you're done, right? So that's very tactile for most of us. But even outside of coding, I think people do a lot of that. And so what I would say, right, is like, you think about search again in a commercial context is usually like somebody being like, how should I start to think about solving a problem? Like, how should I go, like find options I could consider to solve this problem I have, right? How can I use Excel to build my financial model? How can I write a proposal that'll look good to my boss? These are like search intent informational queries in chat search people, people don't really do that. People just say, solve the problem. For me, what I think is really interesting about that is it's actually the highest intent, it's the most valuable thing you could be doing because you're going from being like, I think I need to solve this problem and maybe I will, to being like, I'm actually in the middle of trying to solve it inside ChatGPT or cursor or what have you, depending on what you're doing. So I'd say that's kind of the biggest cluster if you want to get a little bit more abstract. One specific example that takes people a little bit by surprise is like, there are a huge number of people doing like B2B software Bake Offs in ChatGPT. You know, if you're at a big company and your boss is like, hey, we need to buy something to solve this problem. But like procurement makes us get three examples. Like we can't just go buy something, we have to like, you know, like run a bake off. People are outsourcing a lot of that type of work to ChatGPT and particularly to Deep Research, but even just with regular search. And it's really good at it, to be honest. Right. Like, it's really good at being like, here's your comparison table of like different CRM systems or something like that.
- [34:39]
Swix
Yeah. Make the checkboxes focus on what I'm looking for, not what they're selling, you know.
- [34:43]
Robert
Right, right. And so for, for B2B software companies. Yeah, you can imagine as a quick.
- [34:48]
Swix
Note, how do you know? Because you don't directly see the traffic, but like, how do you know?
- [34:54]
Robert
Yeah, so there's two parts to it. Right. So there is looking at panel data, which I think is what everybody who's publishing these studies mostly is doing. And I in previous lives worked in ad tech. So to the extent the web is maybe less pleasing of a place to be, like we're all looking for the guy who did this, it was me. And we used just a ton of panel data in that pursuit to do consumer modeling and stuff like that. So I mean, that's sort of the secret of all of these studies of what people are doing on the Internet is they generally come from like clickstream panel data, which is valuable, but it also could be misleading because the type of people who are in panels isn't necessarily representative of the whole population. And specifically it's not necessarily representative of people who are doing like high value things like being, you know, staff software engineers at Stripe or something like that. So the other thing we've done is we've just done sort of qualitative research. Like we've just, you know, like followed people home, watched them use products, asked them, you know, asked them like post survey follow ups where somebody's been like, I found you through ChatGPT. We've been like, what were you doing when you found us? Things of that nature. And so that's where a lot of those B2B software examples come from.
- [36:02]
Swix
Just to double click on the B2B software thing. I'm sure people ask you this, but I don't think you've covered it yet. What if I try a prompt injection inside of my feature list page?
- [36:11]
Robert
Yeah, I mean it often works right now. Right. Like I don't Think this is a secret? Yeah.
- [36:18]
Swix
So everyone should do it.
- [36:20]
Robert
I think it works until it stops working. Right. And I would say, like there's not a lot of stories of people getting banned for like ChatGPT search so far, but it's coming, right. So I would always say, yeah, so there's a window, right?
- [36:35]
Swix
I don't know. Do you get banned? I mean, you're just trying things?
- [36:39]
Robert
I think, you know, again, like talking my own book. What I would say is as things get more sophisticated, like black hat stuff tends to get kind of pushed out and penalized and there's negative consequences to doing it. And then there's some gray hat stuff which like typically, you know, people do and don't talk about. Really what I would say though about most AI search things in terms of like just optimizing how you show up in these answers if you're a business, is like, I think we're actually in one of these rare windows of time where it's positive sum all the way around. And what I mean by that is like when you look at improving how you show up in like chatgpt answers most of the time the problem is not that you were insufficiently glazing yourself in your product description page a lot of the time. The problem is you're just not being very descriptive about your product in general on your product description page. And so if you provide more information, more context, it helps the model do a better job. It's going to have a more accurate comparison table, it's going to guide the user to a better solution. And so assuming you do a good job providing that context, you're going to be happy, the user is going to be happy, and ultimately the platform provider is going to be happy. I think at Google and SEO, things often feel sort of like pretty zero sum. You know, everything's competitive and like everybody's looking for a trick. There's so much low hanging fruit in this space that I would just say like, you know, before you go to that I would focus on how do you actually focus on serving the user, right? Serving the user through the AI platform they're using. And just like clear writing, good structured content, adding lots of helpful examples and facts and FAQs makes a huge difference. But if you want to prompt inject ChatGPT through search, it's definitely achievable. And again, I think these Systems are like 100th of the sophistication they'll eventually be. Again, not even Talking about the LLMs, just talking about sort of like the Glue code right between the search and web pages and AI.
- [38:32]
Alessio
And when you say tweaking the content, do you still mean doing that in the traditional formats? Or what about things like, you know, LLMs txt? Are these like, AI native ways to alternatively serve content?
- [38:45]
Robert
ChatGPT today, right, is using your existing web pages, so it's using HTML you're serving off your web server. I know that there's debate about this, but I'd say the evidence is like, ChatGPT is not indexing and it's not retrieving content from LLMs TXT by default. And I think a lot of people who are a little bit further removed from the original audience, like the fast AI guys for LMtxt, sort of like have mistaken what it's for. It's great for documentation. It's great if you've actually already written a ton of prose and you're like, how do I load this into my context window more efficiently as a discoverability tool? I would say it's not moving the needle for most people yet, but these things can always change, right? Like everything's changing day by day.
- [39:24]
Alessio
Anything else that you think, people think it's good, but it actually doesn't make a difference. What else would you put in that list?
- [39:30]
Robert
Oh, man. Embeddings. You know, I think a lot of people are very focused on, especially in some of like the SEO optimization community, there's a lot of focus on like understanding embeddings and similarity search and things like that, with the idea that again, like, maybe the search technology is changing. And I think obviously embeddings are very, very useful in general. That, I think is a tool to understand how these AI platforms are consuming your content. Embeddings aren't that relevant. You're better off focusing on like just again, having a good structured set of content that makes sense like to a human with like clean HTML and things like that, trying to like super optimize like, you know, topic similarity and things like that, I don't think makes much of a difference and may harm things in some cases.
- [40:14]
Swix
I just want to open up the space to any other practices you see that maybe super effective or super ineffective that people have got into their heads that, oh, we got to do this for our geo, but it doesn't matter.
- [40:27]
Robert
I feel like we've covered quite a bit of it. I would just, again, I would say clear writing matters a ton. And then this is, I guess we haven't said this on the POD yet. So this is one thing definitely people should take away, which is like ChatGPT doesn't execute JavaScript, most of the AI search indexers and retrievers don't execute JavaScript. So if your content is not rendered on the server side, it's typically not going to be available. Right? And we see this trips people up all the time. And I would just say like, I'm perhaps of like an old fashioned like SSG enthusiast, right? But even if you're using like next JS and things like that, we've seen lots of examples with customers where you know, they've got like a use effect in some page somewhere and it breaks server side rendering and then all of a sudden they're like, none of this content's available.
- [41:16]
Swix
Turn JavaScript off, check your page. If it looks good, you're fine. If it's not, fix it. Every developer here knows how to do that. I think that's really helpful. I think the one for me is I have been involved with a company that put a lot of effort into this programmatic SEO, but augmented by AI. So you get into this really terrible, horrible situation where you're generating a whole bunch of Pages for using LLMs in order to be read by LL in order to rank higher, right?
- [41:47]
Robert
The infinite chain, maybe it works.
- [41:50]
Swix
And if it works like, okay, there's a number where it makes sense.
- [41:53]
Robert
I don't know, I think there's good bad versions, like everything, right? So what I would say is first off, like, it's just a fact of life that a lot of content, especially in marketing websites is being produced using AI. Like that's just like it's already so prevalent, you know, when you get to like scaled programmatic SEO, which is a little bit more controversial, I think that it can be super helpful in some cases if it's done well. And really when I say done well, what I mean is like if it's bringing some kind of like insight that actually is particular to you and your company and what you do and making it easier to consume for AI, right? Or for search in general. So if you're just taking kind of like content off the public web that's there already, that's already like super well represented in AI and you're like remixing it and like maybe they'll cite my page instead of this other page. I mean it works sometimes, but I wouldn't say it's a durable strategy. We have customers who are doing things where like they take for example, like support tickets that are coming, you know, they're looking at their support tickets and they're Doing programmatic SEO generation of like how tos from the support tickets and putting that on their website. And number one, it's super helpful for users. It's helpful for the support team because they get fewer support questions. And that content is like the almost the exact ideal content you could give to an LLM, right? Because people are going to chatgpt, like how do I solve this problem? And it's like, here you go. I think it can be done well and it can be really helpful in those cases. I think there's a lot of things that work right now in terms of just like using AI to remix content and get more scale. But like everybody has access to these tools, right? So eventually, you know, it's no longer relevant.
- [43:21]
Alessio
I think we just had a couple more things. So do you have any sense on difference between ChatGPT search versus deep research and how they leverage search read content? Is it just using the same tool or do you see very different results?
- [43:36]
Robert
Yeah, I mean the ingestion pipeline is similar. Right. So you know all the practices I mentioned, like Deep Research also doesn't read JavaScript. Right. It's just quantity. Right. It's doing more searches and it follows up the thing that makes Deep Research really powerful. So like maybe there's, it's good to have a taxonomy. There's like regular search which is like it does one search, it gives you the results, it re ranks them, it summarizes them, you get an answer. There's multi search, which we're starting see more in like regular ChatGPT4 and also in AI mode where it just does a bunch of searches simultaneously and then Deep Research. What's different about it is that it's sequentially following up with a reasoning model to like be like these sources. Didn't answer the question. Let me try something else. And that's really the game changer about like how how deep Deep Search works. I think from an optimization perspective, there's not necessarily a ton of fundamental differences but. But what I would say is like sometimes more content isn't better. And that's true for me as a user using Deep Search. Like I definitely have experiences where I'm like, Deep Research gives me results that are inferior to just using regular ChatGPT search because it's adjusting more content, but the content isn't necessarily like contributing to the understanding I'm looking for. And then as a business who's publishing content that's being consumed by this thing, it's like, how consistent is your content? Right. Do you have outdated things like is it getting information that contributes to the user being helped and having a helpful understanding of what you do or is there outdated and conflicting stuff and classic example and that would be pricing. We see cases all the time where people have tons of pages on their site that mention pricing and some of them are out of date. And the more content ChatGPT is consuming off of your website and more likely it is to get conflicting answers. And sometimes people end up with the wrong prices. So say that's something to be mindful of, but I don't think it changes the game a ton in terms of your strategy as a business.
- [45:34]
Swix
One thing I might think about just trying to think this through. I've never thought about this problem but like if I were trying to optimize, you know, my websites for deep research, I would tell them what to search next. So you really need that like next link or like here's like related links and you obviously want to make it favorable to yourself.
- [45:51]
Robert
I don't know, you can certainly do that and you see that that works to some extent. Like it is obviously because it's serialized, it's influenced by the previous results. But what I would also say in that situation is like why not just include the content in the first place, right? Rather than having it do a follow up.
- [46:06]
Swix
Well, you just can't include everything, right? Like maybe there's just different branches you could take. So the branching factor is high for sure.
- [46:14]
Robert
There's limits on context window and things like that. So I mean I think it is, I think it is reasonable. But what I would also say is like actually that's a perfect case for something like programmatic SEO where you might also benefit from saying like let's have more focused pages that describe like a complete solution rather than, you know, or the complete piece of information for like some particular version of a query or version of a Persona rather than sort of it being choose your own adventure. And I think like that's where having more technical sophistication and how you manage content and like what you show maybe to AI versus like to Google or to regular people can be interesting. And so we have some folks doing that.
- [46:50]
Swix
Some of the sites have infinite scroll. I guess if you're not rendering JavaScript that doesn't matter. But this is this whole thing about LLMs Txt like you just cat everything into one giant file. I mean I can do that. It's just that actually matters.
- [47:05]
Robert
Yeah, I mean there's tons of LLMs full TXTs that are well over the context window of Common models. So you're like, is this helping? I don't know. Again, to me, I'm like, for some of these things, search works pretty well. Like, regular search over your regular website works pretty well and naturally sort of solves that chunk size problem. But you do need to have content on your website that's like, that's helpful and targeted to the questions people are trying to ask.
- [47:30]
Alessio
Awesome. Any case studies on companies that are doing this amazingly well that people should learn from, or are people still keeping it under wraps?
- [47:40]
Robert
We definitely have case studies and I think the ones that are most interesting to me personally, again, are the ones where we're seeing people who are not just getting more traffic, right? Like, traffic is kind of like the first approximation everybody uses in SEO and then also in this AI search phase, but people who are actually getting more conversions, more actual business. And so like two that I would mention, like two really great customers who are both in kind of the dev infrastructure space. One is Clerk, you know, the user authentication company. And they have like, they have great docs in general. Like, they're well set up to succeed in AI because of that. But they've seen a huge lift in AI traffic from sort of like targeted optimizations and like looking at the type of content that AI wants to use and generating more of it. But they've actually seen an even bigger uplift in conversions coming from ChatGPT. So I think the stat is like, they've seen like a 6x growth in ChatGPT traffic, more traffic from native ChatGPT perplexity, but the common set. But they've seen a 9x lifted convergence. And again, I think that goes back to like when people are looking, if people are asking like, you know, how do I implement enterprise? So in my app, for example, right? It's because they actually have that problem. Like they're looking for a solution and they're ready to implement. Right? So the closer you can get them to actually being able to like solve the problem, the higher propensity they have to convert. And that's like the dream, in my opinion.
- [48:52]
Alessio
And are you guys helping with that? Just, you know, for people, you, they sign up, Clerk signs up for scrunch, you guys kind of do look at how it performs right now and then are you helping them generate this content? Like, how much of it do they do on their own?
- [49:09]
Robert
I think where we're at right now is like, we're the feedback and experimentation system. So we, you know, I can't take credit basically, is what I'm saying. The team at Clerk is great and has been really thoughtful about like the types of content they need to create. They know their audience really well, right. They're developers, they have a discord, they're super engaged with the people who actually use Clerk every day. So we're providing I think the supporting role of helping them understand what's working and double downing on it. We're not like a content generation company, right? We don't know their business as well as they do. They can do a better job generating content. We are I think working on helping them put that process more on rails, you know, so creating more structure, being able to run multiple experiences at a time and ultimately helping them try to figure out like how to deliver more technically of that value to the AI platforms. And that's where maybe some of these other things like we didn't get into it but you know, obviously mcts, NL Web, everybody is wondering what the tactical mechanism is going to be for getting this content into these AI platforms if it's not just traditional AI. Search in the future is so starting to do more work there which I'd also put under that sort of like agent experience bucket. But they're the star of the show and I'd say that's true for all of our customers. So you know, we are a solution people can use to solve this problem. We're not an agency that comes in and solves it directly but we do have, I'll give a plug here, we do have tons of really great agency customers. So if you're looking for an agency to do it for you, we can definitely steer you in the right direction.
- [50:36]
Swix
And can you give a range of like just literally okay, like you know, I'm a company, I want to improve my rankings or organization how long and what kind of uplift is typical at just just to give an idea of what's on the table for, for these.
- [50:50]
Robert
Like, you know, I mean I, I think it varies a little bit by vertical. Like devtools is, is, is a very good fit for AI. So I'm not going to promise that everybody's going to see a 6x uplifted traffic but there's a lot of low hanging fruit. So I would say we typically see people get double digit improvements in traffic within a month or two if they, you know, if they are actively working to improve. Right. Updating their website publishing content. Some folks are more in wait and see mode and you know, that makes sense for some businesses but if you're actively working it's achievable those things on.
- [51:20]
Swix
Like higher conversion, but also just, just raw. Higher traffic is not something that I was thinking about or watching, but it's really like it's, it's starting to flip for some people where like it's actually more than normal Google and then like, okay, like your. While your budget has to shift over. Basically.
- [51:36]
Robert
Yeah.
- [51:36]
Swix
Just the last wrapper. Two last wrapper questions, quick. One is, which is, do you have a sense of like market share of let's say ChatGPT versus Google AI overviews? I assume, like Claude is a lot smaller. What is the market share? What are the rankings in your mind of what people care about?
- [51:53]
Robert
ChatGPT, I would say by far in a way is the thing that people think about as being an AI platform that has the strongest consumer presence and the most durable relationship with consumers. Right. So everything else is kind of a distant second now. AI overviews and AI mode, which is brand new, so I don't have great stats on that. Obviously are like they're putting. Being put right in front of your face. So huge, huge traffic. And then meta AI, right. Like Sleeper has published stats on, you know, having 700 million, I'm sure more than that now active users because they're, you know, if you do an Instagram search, you end up in that AI.
- [52:28]
Swix
You haven't mentioned Perplexity, which I think is interesting.
- [52:31]
Robert
Yeah, yeah, Perplexity is pretty big. So I would actually say that like Perplexity is the second biggest AI native search platform after ChatGPT. And it's got, I think more importantly, it's got like people who are really passionate about using it. You know what I mean? I would say that like people who are really into Perplexity are probably more into it than many of the other options, you know, but it is, it is smaller, right? In terms of raw volume. I think what's actually really interesting. And I didn't pull like you guys could pull this up the same as I can. But if you go look at like App Store rankings, for example, I think what's actually more interesting than just looking at like, you know, how is ChatGPT the number one app in the App Store in the category, which mostly has been. Right. What's actually interesting is look at the volatility of it. And if you look like, go look back at Deep Seek, look at when Gro first launched the mobile app, things like that. And you can look at similar, you can look at similar data from like, similar web, for example. Chat has pretty consistently been very high traffic. Obviously the growth has been insane. But in terms of, like, relative market share, it's been consistently one of the best. And in contrast, we've seen more peaks and troughs with things like Gemini and Deep Seek and Rock and things like that. So they may well become firmly established. But nobody has this sort of durable relationship with tons of people like ChatGPT does in the space so far. And Perplexity, though, if you look at it again in absolute terms, it's lower, but if you look at the consistency of people using it over time, it's really, really strong. So I would say, like, especially if you are in a category that has an affinity for perplexity, if you're in tech, if you're in maybe some seconds of finance, things like that, early adopters. Yeah, you need to pay attention to it. And same thing for Claude, right? Like Claude, famously, like, way fewer users, like revenue, actually different story in terms of, like, being compared to OpenAI. But, like, people who use Claude are very, very passionate about it for the most part. That's an interesting thing to think about from a strategy perspective. And I will say I don't think it's. ChatGPT is durable, but not infinitely so. If you look at the reaction to Glazegate, the reason people, I think, stick with it is because they actually just really like the product, which is sometimes under appreciated. I think in tech, we always talk about platform wars and politics and stuff like that, but people just really like it. But when people feel maybe a little bit betrayed, like the product's going the wrong direction, there's a lot of pushback. And so it's going to be an interesting time to be alive over the next couple years as it has been so far, but I think it's here to stay. Awesome.
- [54:58]
Alessio
Robert. This was great. Anything else we missed or any call to action for people? Are you hiring? Are you? Obviously more people should use it. That's kind of obvious.
- [55:10]
Robert
Yeah, I mean, I would say, I think this is the future of the web, right? Like, the future of the web is like, you need to be sort of AI compatible and understand how the things you're publishing shows up in these systems. And so just like, as somebody who's been a web enthusiast for 30 years, because I'm old, don't get caught up in snake oil and investigate what works and, like, do things that make sense and, you know, don't ignore it either. I would also just say, like, for. Especially for the latent space audiences, like, we are definitely hiring and I think a lot of what we're doing right. Is we're doing research into exactly what I just said, which is, like, understanding how AI and the web kind of interoperate in the future and, like, what the future of the web should look like if you're a business who's trying to be in front of customers. And so if you're a type of person who's interested in helping us figure that out, like, we are definitely hiring. We have a lot of open roles, and we would love to. Love to talk to you.
- [56:06]
Alessio
Awesome.
- [56:06]
Robert
Thank you, Robert.
- [56:07]
Swix
Thank you so much.
- [56:08]
Robert
See you guys.