All episodes

Episode 001 · 1 hr 13 min · 17 November 2026

What Comes After Prompt Tracking?

With Josh Blyskal, Profound

This episode is brought to you by Notion.

Josh Blyskal landed in New York from Singapore the night before this recording. A day later he sat down with Tom Critchlow and Jeremy Cabral for 73 minutes on what Profound sees across its customers, and where it thinks AI search goes once agents start doing the shopping.

He knew AI was real on 30 November 2022, the day ChatGPT launched. In January, on his first day back on HubSpot's SEO team, he was told to track it. That turned into prompt engineering workshops, an internal AI team and a prototype AI search grader that reached the same idea as Profound independently. He joined Profound as its second employee.

Prompt tracking turns into journey tracking

Tom asks the uncomfortable question first. Profound built a business on prompt and citation tracking, so what happens when an agent books the podcast studio and nobody sees a link? Josh thinks the measurement follows the user: journeys, preferences and decisions inside personal agents like Instinct and Muse. Walmart's MCP points the same way, and Josh wonders aloud whether ChatGPT ends up showing a results page of MCPs.

Freshness, chunking and YouTube still work

Josh's team A/B tests pages, comparing optimised pages against a control on bot visits and leads. Freshness still moves the needle when the refresh rate matches what the query fan-outs ask for. A bank refreshing rate pages monthly found ChatGPT asking about three-day windows. Self-contained answer chunks work. Product stats belong in readable HTML. Off-site, YouTube matters more than follower counts: good descriptions, chapters, transcripts and several languages.

He shares Profound's figures on timing too. A new piece of content takes about six days on average to earn its first citation, and half of the most-cited content is 13 weeks old or younger.

AI slop is a skill issue

On AI-written content, Josh is blunt. Slop comes from having no perspective. His picture is beads on a string: the beads are your ideas, your IP and your aha moments, and the model is only the string. Context does the rest, which is why Profound runs its own marketing through its Context Manager. Tom pushes it further. The next argument is about whether the ideas are original, and who wrote the words matters less.

The marketing engineer ties it together. Josh describes one accountable person who understands the analytics, speaks each team's language and builds agents so that PR, email and the blog are each findable by agents. He thinks SEOs are best placed to do it, because they've spent 20 years working between people and complicated machines.

Josh Blyskal

Profound

Josh Blyskal was Profound's second employee and leads its research into how AI assistants search, cite and recommend brands. Before Profound he was on HubSpot's SEO team, where he was handed ChatGPT to track in January 2023, ran prompt engineering workshops and co-founded an internal team bringing AI to the rest of the company.

Key takeaways

  1. AI slop is a skill issue

    Slop comes from a lack of perspective. Feed the model your non-negotiable ideas and your own context, and let it do the joining up.

  2. AI shows up before purchases, unseen

    Profound's study of a few thousand real browsing sessions found an AI conversation somewhere before up to 30% of purchases within a month. Very little of that reaches analytics.

  3. Let the fan-outs set your refresh rate

    A bank refreshing its rate pages monthly saw ChatGPT asking about three-day date ranges. Matching that cadence sent bot visits up sharply.

  4. Write answers that stand alone

    Self-contained chunks get extracted. Put product stats in plain HTML. ChatGPT is forming its own view of products and leaning less on comparison listicles.

  5. Treat YouTube videos like blog posts

    The videos being cited aren't viral. They have clear descriptions, chapters, clean transcripts and several languages, and they answer the question fully.

  6. Expect a first citation in about a week

    Profound's data puts the average at about six days: around four for ChatGPT and 7.6 for Claude. If nothing is cited after 30-odd days, something is wrong.

  7. Track 250 to 1,000 prompts

    Josh says that covers about 85% of use cases. At 10,000 prompts, a single visibility score moves 0.1 a week and tells you almost nothing.

  8. Give it one owner

    Profound tried spreading AI search across teams and nobody owned the best insights. Josh wants one marketing engineer accountable for it.

Chapters

  1. 0:00Cold open: AI slop is a skill issue
  2. 0:48This episode's sponsor, and welcome
  3. 2:00The day ChatGPT launched, and tracking it at HubSpot
  4. 5:42From customer success to stages of 10,000 people
  5. 7:19How AI search reached the boardroom
  6. 12:56Is GEO a distraction? Measurement and the McNamara trap
  7. 15:29Profound's browser-session data on AI before purchases
  8. 16:36Tom's ChatGPT and Google usage, month by month
  9. 17:46Personal agents, and what comes after prompt tracking
  10. 22:46Walmart's MCP and a results page of MCPs
  11. 23:28A 1995 paper on cybermediaries, and the new middlemen
  12. 25:03Unified checkout and data partnerships
  13. 27:11What marketers can do now: A/B testing pages
  14. 29:08Freshness set by the query fan-outs
  15. 29:58Chunking, schema and product stats
  16. 31:33The marketing engineer
  17. 33:54Brave, Exa, Parallel and ChatGPT's own index
  18. 40:00Optimising the index layer apart from the answer layer
  19. 42:39Does training data still matter?
  20. 46:04Cheap compute, the open web and agent-ready pages
  21. 50:43Mapping your site's task gaps
  22. 51:52Citation bots as a proxy for human attention, and YouTube
  23. 55:16What it takes to dislodge the cited brand
  24. 56:45Time to first citation, and citation decay
  25. 57:30How many prompts to track
  26. 59:09Should companies use AI to write content?
  27. 1:02:32Context as the antidote to slop
  28. 1:03:51The next argument is about ideas
  29. 1:05:20Three things to do now, and who should lead it
  30. 1:06:08SEO heritage and digital shamanism
  31. 1:08:06Who wins: Google or Elon?
  32. 1:09:31Rapid fire

Rapid fire

AI stack
Profound, Wispr Flow, Claude, Cursor, Instinct, Muse, Grok and ChatGPT.
One number for the CEO every Monday
Visits from citation bots, as distinct from crawlers in general.
One website change for agents
An accessibility tree for a site that lacked one. Too early to see much effect.
Job title and interview question
Marketing engineer. What do you know about agents?
Most overrated
Listicles.
Most underrated
Documentation.
Worst advice
Scaling content around the clock without taste or an opinion.
Most-shared reading
Anything by Paul Graham.
Changed his mind on
That browsing would be the only thing that mattered. Interest in training data keeps coming back.

Mentioned in this episode

Questions from this episode

What comes after prompt tracking?

Josh Blyskal expects it to become journey tracking. As personal agents book and buy for people, the useful signals are the decisions and preferences inside the agent's session, and citations matter less.

Is GEO a distraction?

Josh's view is that people already use ChatGPT, Claude and Google's AI Mode to find products, whatever attribution shows. Profound's browsing data found an AI conversation before up to 30% of purchases within a month, and brands that act only on what's measurable will fall behind.

How long does it take for new content to be cited by AI assistants?

About six days on average in Profound's data: around four days for ChatGPT and 7.6 for Claude. At the 90th percentile it's 30-odd days, so content that isn't cited after a month has a problem.

How many prompts should a brand track?

Josh suggests 250 to 1,000 for most brands, which he says covers about 85% of use cases. Very large sets need breaking down, because a single visibility score across 10,000 prompts barely moves.

What is a marketing engineer?

One person who owns AI search and agents across marketing. They understand the analytics, work with each team on its goals and build agents that make that team's work findable by other agents. Josh thinks SEOs are best placed to take the role.

Should companies use AI to write content?

Yes, with a point of view. Josh says slop comes from a lack of perspective. Give the model your ideas and your context, and treat it as the string that joins them up.

Which search engines do AI assistants use?

Profound's research found Claude using Brave Search. Josh thinks ChatGPT runs a hungry index of its own and blends sources. The personal agent Instinct uses Exa, and he expects index providers such as Exa and Parallel to matter as much as the assistants themselves.

Read the full transcript

0:00Josh Blyskal: We're evolving from answers to evaluation. And then from evaluation, we're going to move to action. And that's where the paradigm is going to move. Should companies be using AI to generate content? I think AI slop is a skill issue, honestly. I think that at the end of the day, everyone has experienced slop in some form, but slop is the result of a lack of perspective. It's the result of a lack of opinionation. It's actually probably the result of a lack of a marketing engineer on the team to understand how to combat it. I mean, AI is trained on The whole of everything in the world, right? And it is basically the opposite of marketing in that way. That marketing is about differentiation. AI is about distillation. And so you need to work actively. You need to combat intensely this notion of slop.

0:48Tom Critchlow: This episode is brought to you by Notion, the collaborative AI workspace. For leaders investing in AI, the challenge is turning powerful models into work that makes a difference right across the business. Notion brings your company's knowledge, projects, and AI agents together so teams can finances, automate recurring work, and make decisions with the right context. From fast-growing startups to global enterprises, Notion helps teams put AI to work together. We actually use Notion to run AI Search Leaders. It has all our context. Our agents use it all the time, and we love it. Learn more at Notion.com. Hi, welcome to the AI Search Leaders podcast, where we talk to experts and industry leaders about what AI Search is and how the world is changing. Josh Blyskal from Profound is here. Thank you so much for joining, Josh. A little background intro for you. You came via Australia and Singapore, Founded in NYC last night. Last night? Smashed some caffeine tablets maybe. We're not going to ask too many questions but excited to have you here. You are on the founding team of Profound. Profound as I think needs no introduction has been on a rocket ship as part of the AI Search Journey, but I want to talk about your journey. You were at HubSpot for a while. Talk to me about, like, you're at HubSpot. You ended up on the founding team at Profound. When did you know AI was going to be a big deal? How did that transition go? Talk to me.

2:00Josh Blyskal: I mean, November 30th, 2022 was the day that I knew that AI was going to be real. I mean, it was the day ChatGPT released. I was finishing up university. It was the last month of my university degree. And another student, a colleague of mine, basically sent it to me. And I had seen DaVinci. I was kind of a nerd like that. I kind of liked the Transformer models. I used them for jokes. Basically, you could be like, write me an Elon Musk rap album. I don't know if anyone's ever seen Talk to Transformer, but across the world, when people ask me, where did you get started with It was these like joke, like hugging face spaces where you could use GPT-2 and it would be basically gobbledygook, you know, nonsense. So I believed in it from the beginning. I was like, this technology is fundamentally going to be really interesting. And the fact that ChatGPT was a consumer site and that it was coherent, I was like, oh, this is going to change the world. This is certainly going to be used by basically anyone to generate Essays, questions, text of all varieties.

2:50Tom Critchlow: Okay, so you see the world's gonna change?

2:52Josh Blyskal: Text is about to change, yeah. And then what do you do with that? And then nothing, right? It's incredible. I'm like, all right, so I have this knowledge. You have the first key knowledge. College is always the first key in being able to make something really interesting, but you've got to have other things happen. And so I was fortunate enough to be joining. I already had a signed offer with HubSpot. I did an internship with them in the summer. So I was going back to the SEO team. And of course, the determinant of whether I could really do something is what they want me to do in that internship. Luckily enough, I get there January, I'm really good with dates, January 17th, I think it was, I get there as my first day and they say, what is this ChatGPT thing? We want someone to track it. You're new blood, you know, you don't have a huge remit yet. Why don't you just be the guy Whatever this little ChatGPT technology becomes, you're just tracking it. So I'm going from being like the little SEO guy who's like in charge of doing a ChatGPT to eventually starting to give like these talks that are sponsored. If anyone knows Dharmesh Shah, Very popular. One of the best marketers in the world, even from like a CTO. He's a CTO. He'd never claimed to be a marketer, but he's a great marketer.

3:54Tom Critchlow: He's still hands-on. He's awesome.

3:56Josh Blyskal: He's still hands-on. So I'm giving these like prompt engineering workshops under his kind of tutelage. And then it starts to become, who is this Josh guy? He's on the SEO team. He used to do a little bit of SEO, but he's really transitioned into like doing AI. And his remit doesn't really make any sense. I co-founded a team at HubSpot to basically bring AI to all the other teams, building out automated emails, yada, yada. So after about a year of doing that, the SEO team reaches back out to me and says, you know, SEO, we want to build an AI search grader. We want you to build it and prototype what that would look like. And so I built Profound's, like basically the V zero of what Profound ended up being. And it wasn't inspiring Profound, but it was like the same, we reached the same conclusion independently about six months beforehand. So when they came out of stealth, I always wanted to join a startup. No other startup's gonna take like a former search kid with like some automation experience. Like that's a hard sell. I'm a 23 year old, you want me to be your general operator, but at a company that's specializing in AI search, I'm a kid who already I built once for HubSpot, which you're seeking to build, and I have an appreciation for search, and I want to talk about it, and I could talk to your customers. It was a perfect f***ing sell. It was great. So I ended up joining in as the second employee.

5:08Tom Critchlow: That's amazing. You know, it's funny. I got my career started, and you got the gray hair. When SEO was the same thing. I was at a web development agency many, many years ago, and they were like, I think Google search is a thing. Why don't you go figure it out? But it's like you get that marginalized. You're off to the side, but you have free license to become known in this fledgling industry. That's amazing. I want to picture the scene. You've been at HubSpot for a while. You've kind of owned this kind of niche of like AI search and AI. You're giving presentations internally. When did that kind of fame inside HubSpot first translate into the broader world?

5:42Josh Blyskal: It was the AI search creator, but that wasn't even really Right, frame externally. I mean, I talked to, you know, Marty Swant from Digiday, maybe, maybe not. Marty Swant interviewed me for that. That was like the very first time that someone outside of HubSpot would ever know my name. And it wasn't many people. I think probably like five people read that article, but it was important, right? It was a watershed moment. We're talking about AI Search. And it's public, but really in joining Profound as the customer success engineer, I was the second employee. You start answering questions at a very micro scale. So you're like, hey, this is where you can find your visibility data. This is where you can find your citation data. And you start scaling that up. And so you start talking to like these SEO operators, then that scales up to talking to their team leads. Then that scales up to talking to a C-suite that needs answers. Then that scales up to talking about a parent co or a holding group. And maybe you're at their investor event, or maybe you're at their all company onset and you're doing a panel because they really love the answers you're giving in the framework. They're thinking about AI visibility, which scales up to massive events and influencing and LinkedIn posts. And the funny thing is you're still answering If you zoom all the way back down, it's just the scale of the questions you're answering. Because the same patterns that I would find that I would talk to, you know, Carly from XYZ company would be the same things I would shout about on a stage of 10,000 people. It's just pattern recognition and the ability to extrapolate and communicate effectively. But I always joke that basically being an influencer is like being a CS person at scale.

6:59Josh Blyskal: It's true. And there's a very clear line you can make. So, I mean, being in the ground though, by the way. Best training data you'll ever get on doing SEO stuff or anything GEO. You can't really sit on high in the clouds. I still talk to customers to this day because it keeps me grounded on the patterns and the data.

6:59Jeremy Cabral: Information gain city right there. That's how you get the golds put into your content.

7:19Tom Critchlow: That's my favorite thing with Jeff Bezos is just being close to the customer. Channel the customer needs. Yeah, that's key. You mentioned something I want to pick up on and kind of dive deep on actually. And I think it's one of the most important pieces of AI search, which is, This is an executive level of concern. SEO, I rewind to like 2020, 2021. I think there's a general sentiment that kind of SEO hadn't really changed for a while. It was like before the helpful content updates. It was before the rise of Reddit. And SEO had kind of been a stable thing for a little while. And because of that, I think SEO was kind of becoming marginalized. CEOs were less interested in it. It was less on people's roadmap. It was much more commonly understood how to do it, etc. And then AI Search comes along and everything changes. And, you know, I've sat in meetings with sales teams, marketing teams, content teams, PR teams, engineering teams, and also like, you know, VPs, directors, C-suite, board levels, like investors, like everyone cares about AI Search. What do you see as different about the conversation, those different kind of stratosphere levels? How are investors and board advisors talking about it differently than how VPs and directors and so on are talking about it?

8:33Josh Blyskal: The interesting thing is that it really gives SEOs to the ability to Practice how they sell up and down the org. I don't think we've had a lot of practice having the latitude to be able to talk to investors and kind of extrapolate these things. The old adage, it depends, doesn't work when you're brought in in front of a group of CMOs or CEOs and they just need some directional information. But one thing that we do very well, which is making decisions with incomplete information, That's where we shine. We can give you the paths and we can tell you from our experience, these could be some of the outcomes that you're looking at. And the cool thing I think is like being able to shop those things around in a not strictly technical way and really relate those concepts to someone who really just wants to, how do we solve AEO? And you know, if you hear a CEO say, how do I solve SEO? You know what kind of conversation you're about to enter into where you really have to keep those concepts very high level. You have to understand the understanding of, you know, How this continues to be a game with market dynamics. People are going to use some strategies. Those strategies are going to become lame after a while. They're going to end up getting adopted by the whole industry. We're going to optimize on top of those optimizations, and it's going to kind of layer and layer. Some things are going to be spammy and bad, and we're going to not do those things. Some things are going to be really good, and they're going to help people and bots find our content. We're going to definitely do those things. But at the end of the day, either it's somewhat similar, it's somewhat different,

9:48Josh Blyskal: depending on how you talk to your CMO about it. I think the thing that changed with the CMOs overnight though that they noticed themselves their search patterns were changing. The reason that this whole thing kind of blew up is that you can't tell a room of CEOs and CMOs, no, don't worry, this is the same as SEO if all of them are feeling that these are not the same search behaviors. This is not how I'm finding companies. And so it was a very lived experience from I think the boardroom that really dragged us up into it, especially.

10:15Jeremy Cabral: I think what's interesting is like, Everyone wants to put SEO into a box. Neatly, this is what we do. It's always on. And I feel like there's the same need and want to do that with GEO or AI Search, like we call it. But in reality, what you got here is you need to surf this thing because it's not what it appears to be. It's actually entirely more than industry changing. It's business changing the way we need to operate. There's functions that don't exist to be able to handle this. So it does this executive buy-in and Just realizing that it's R&D, really where we're at right now is trying to figure out not only what's working today, but directionally, what does a company that can actually sustain through the future look like? And how do we, I guess, realize the role of AI Search? And even just calling it AI Search is limiting in some respects, right? It's really this entry point to ultimately having your business be a chosen business where a company transacts with you. Or a consumer decides to use your platform for whatever they're seeking to complete. So yeah, it's just something that's like to distill it down to a set of metrics so I can simplify it isn't the answer. It's actually more about a journey that needs to be always on as an agenda item until we finally get there and figure it out in the next few years.

11:26Josh Blyskal: What kills me is so many companies out there want the answer right now. Like they don't realize that this is all the era of ROI across the whole board. I'm doing ROI, I don't know yet. They're doing ROI, you're doing ROI, you're doing ROI, or sorry, R&D. You're doing R&D, R&D, everyone wants ROI. It's really hard, even in the most advanced situations to give you that clear picture. But the point is that with any of these new channels, you're looking at social, you're looking at TikTok, for instance, the people that are first on these channels, the people that master the channels are in a better position than basically anybody else But you have to be willing to accept kind of the vibey nature, the like incomplete nature of the channel as a marketer to be able to go in there, you know, take no prisoners, try some things, be a mercenary and then go make some wins, take some losses and try to see what works so that when the ROI That's why a picture does get clear when we do have the attribution, when things are making sense, when the audiences do arrive. By the way, if you asked me in 2024, is AI Search, should it be a priority for every CMO? Even before CMOs were like, I think of 2025 really as like the year that this really exploded. Even in 2024, when Search, we were doing this at HubSpot, they were like, yeah, we like this Search greater, but Search, You think like AI Search is gonna be like two, 3% of all search. It's like you're drastically undercounting Google AI mode, AI overviews. It's a joke, but that could be a thesis.

12:41Josh Blyskal: And it was a thesis for a few months in the beginning of that year. And slowly, that thesis got a lot quieter before people could really start to take action in a meaningful way before all these companies existed. So to your point, I think you gotta be vibey. You gotta be willing to take a jump.

12:56Tom Critchlow: Right. I don't want to dismiss that argument. You said it was a joke, right? Just this last week, before we recorded the podcast, there was a lot of discussion on LinkedIn about whether GEO is a distraction. The measurable traffic from ChatGPT showing up in people's Google Analytics is tiny still. There's a bunch of, I think, somewhat flawed research out there showing that AI usage is still a tiny fraction of Google usage, etc. What do you say to those people? Like people that I think are having a hard time, have spent their entire careers based around measurability, right? So it's like, you talk about like, We're cooperating with incomplete information, which is something that SEOs have done very well in some ways. At the same time, the entire SEO industry has been click -based, measurement -based. We've always been more measurable than everyone else. I remember back in the early days of SEO we made fun of PR teams because they were unmeasurable. But SEO was like the golden child where you can measure it, there's clicks, directs. Now a lot of people Have spent their career around measurability and are now stepping into a world where measurability is hard. A lot of things are not showing up in attribution in all kinds of ways. What would you tell somebody who's in that position? Is hearing these things on LinkedIn from other influencers, et cetera,

14:12Tom Critchlow: about the fact that maybe this thing isn't real, that isn't a big deal, right? Like I think the three of us kind of believe it's a big deal. What would you say to somebody who's like, you know, having that hard time?

14:20Josh Blyskal: I mean, honestly, regardless, I have a few different points. One is that Regardless of how the AI race shakes out, people are going to use ChatGPT, people are going to use Claude, people are going to use AI mode, AI overviews to find brand products and services. The user behavior, if you survey users, is there. People ask these systems questions, they're some of the most used web The attributed traffic is the difficult part, right? Is the click through coming? Can we actually draw the line? The issue is that this is kind of Darwinistic, right? Like the brands that are able to adopt this paradigm of thinking are gonna be the brands that end up surviving in the longterm, It's so hard. Like you can tell a brand all day long, you might not be able to get measurement on this, but you still need to do it. And some brands might just say, you know what? I don't want to do it. And that puts you in a, I think, Darwinistically tough position as a brand to be able to have to say, we only really want to do what's hyper measurable Have you heard of the McNamara paradox? Yeah. Like you only want to pursue things that are hyper measurable. You do that at the expense of everything that's unmeasurable or intangible. But what I'll say is that we were in Sydney and we were in Singapore and we were in London and we have some really interesting data that's relatively exclusive. Obviously we talked about it there, so it's not dead exclusive, but this is the first podcast I've been on where I talked about the data where we took real browser sessions from Chrome, from other browsers. This is a few thousand real browsing sessions.

15:35Josh Blyskal: And we looked at when people talked in ChatGPT and when they finally made purchases. And what we found is that for someone that actually had an AI conversation and ended up purchasing something, Basically up to 30% of conversations within a month, sorry, 30% of purchases within a month had an AI conversation beneath them at some level. 30% of those purchases. So one in three, someone's talking to AI in a relevant way about that. If you narrow it only down to-

15:35Tom Critchlow: And this isn't just like AI power users.

16:01Josh Blyskal: No, if you narrow it to AI power users, it's 51% and then 25% within a week. Can you believe that? So I don't necessarily need to even belabor the point to say, oh, you know, get your feelings right. You know, you just know that people are using it. I mean, that's true. It's truthy, but we have the data and we're looking at it and we're seeing, you know, 30% of people across the board and that's 16% every week. So it's tremendous, tremendous signal that, you know, up to it, I would say up to a third of your, you know, attributed traffic might just be sitting there quietly, attributed influence at least.

16:36Tom Critchlow: I mean, I found this podcast studio on ChatGPT. Right? I mean, again, back to that, I think, you know, in times of change, I actually think anchoring to your own behavior and your own usage is one of the clearest signals that you can anchor to. It's like, what do you truly believe to be true? We can't see so much of what's out there in the world, but you can see what's right in front of you. I actually just downloaded my ChatGPT usage And my Google Search usage archive. In July of 2026, my ChatGPT usage surpassed my Google usage for anything that wasn't navigational. So not to be just typing in Airbnb or whatever. September, 2026, my ChatGPT uses to dispatch all Google usage, including directly typing in domains. So I am using ChatGPT way more now than Google personally. And so that's a wake-up call. And that curve, even for me, who I think is kind of an AI power user, like I use AI for everything all day long. ChatGPT was still like a third of my Google usage up until about June, July this year. So the speed and scale of how fast this space is changing It's really hard to get your head around.

17:46Josh Blyskal: Have you seen these personal agents now? Instinct, Muse. I have. What do you guys think? What's the perspective there?

17:52Jeremy Cabral: Well, I think that's where it all ends up, right? And we're focused very much on discoverability, but ultimately task completion is the main game. And the more that people hand off tasks to these agents, and they're not only just triggering when we ask them to, but they're proactively, like they're saying, I'm just imagining a world where there's not even any engagement with these platforms at all and it's just happening in the background. And the question then becomes is about whether I'm a mentioned brand Or am I just the chosen and pre-selected option ongoing? How do you dislodge that? What is that game going to look like? And I think that's what this podcast is about, bringing people on that can, I guess, not only talk about where we're at as an industry and what's happening, what the norm is and the standard, but like where's it going? And how do we prepare for that world early because I think there's some really great brands that need to know what to do, how to exist. And I think some awesome agencies that have been helping these brands for an extremely long time that hopefully can continue to be valuable to them. I'm very much a believer in the human still being relevant and extremely important and guiding through these brands and then what happens next. So yeah, definitely my hope of what we can achieve through this pod.

19:04Tom Critchlow: I always go back to like... Wait, can I ask a question? Yeah. I want to get a little real talk. You know, Profound, I think, is many things, but has kind of built a business on the back of citation tracking, prompt tracking.

19:04Josh Blyskal: It's true.

19:04Tom Critchlow: But we're just talking about the fact that, like, maybe kind of prompts and citations are kind of going away, right? It's like somebody's going to ask to book a podcast studio. They're not going to see any links or citations. It's just going to book the podcast studio for you.

19:04Josh Blyskal: I think it turns into journey tracking.

19:04Tom Critchlow: Okay, so how does that work?

19:33Josh Blyskal: How does Profound do it? I think it is a big question right now. What's the paradigm? You know, what we're always thinking is that the evolution of citation and visibility tracking kind of evolved from our users. When we first started, we didn't know that it was going to be citations are this and visibility is this. And share voice is this. I think that we're gonna have to try a few different modalities of testing with these different models to figure out what actually gives us signal. Of course, we can go through full user conversations with Muse and see what happens. We could spin up new sessions, try different, the interesting thing is the mono chat versus side chat thing. If you don't have context on that, it's like, there's these like instinct, for instance, you text it. There is no start a new chat. It's just one long bit of context. For Muse, you can do the mono chat, the one text chat, or you can do side chats. It's always an interesting question, what we do with that, how we manage that, how we scrape that at scale, how we use that at scale. I think at the end of the day, something new is gonna have to be invented to deal with personal agents in terms of the tech stack for scraping. But I think you're dead right. There's nothing precious about prompts or citations to the end that marketers want to track. If marketers want to track prompts and citations, so be it. But I think we both agree for the personal agents, it's going to be a lot of execution. It's going to be a lot of preferentiality. It's going to be a lot of decision-making.

19:33Jeremy Cabral: And it's interesting, right?

20:48Jeremy Cabral: Because like we're in a phase that software providers can actually keep up with how things are changing. I remember, you know, being a customer of like a, I don't know, Moz or Ahrefs or whatever back in the day, you'd wait months before they We're not able to react because code was difficult. It would take time to prioritize that thing. Whereas you can have a feature release and immediately have tracking of that thing. I think the pace that you guys are releasing at is honestly very impressive. There's some companies out there that have done some things when they're moving quickly like Cloudflare and others. But I think the fact that you are so fast in releasing features is impressive. And yes, it's going to be interesting to see where it goes from here and so many surface areas to track. Like honestly, like we call it almost like the source terrain, essentially, where there's just like this landscape of like so many things to look at. And it's like, so like what's going on out there? Like which sources should we pay attention to? And how do we start tracking each of these? Because it's so divergent. It's true. Divergence like it's, and it's hyper-personalized. And there's one thing to track the initial prompt, but what about refinement prompts and all these things as well? Super interesting. I always go back to when Facebook first hit, everyone dismissed it. It was like, whatever, it's not a thing. It's just a fan page, whatever. And then it's not the fact that It's small. Right now, it's the growth rate.

22:03Jeremy Cabral: It's exponential. The week over week of AI assistant impact is massive. I think this Black Friday that's coming up is going to be enormous in terms of really painting the picture for everybody out there and brand marketers and understanding It's time. You need to be agentic. Get your platform showing up and also make it agentically used because that's literally how shopping works today. Ecommerce is the thing that always used is a daily use product. I think we're going to be thankful for Ecom. Bringing brands forward because it makes it critically important. And then it drags everybody else along with that. So yes, it's so interesting seeing how it plays out. Like we have that kind of stop start from ChatGPT. It's like we're in shopping, but we're not.

22:46Tom Critchlow: Well, just in the last few weeks, Walmart announced their MCP, right? Walmart and MCP. Oh my God, it's just insane. But that's the thing. It's like the space is moving rapidly. That's what I think the future holds for shopping, by the way.

22:58Josh Blyskal: I agree. These MCPs. I agree. Why are we trying to scrape? I've been yelling at us for months. Why am I scraping Walmart? Why am I trying to scrape Macy's? Why am I trying to scrape, you know, all these different stores? Let's just MCP directly into them. You know what the funny thing though is, when I ask a question to ChatGPT, is the world, is ChatGPT's backend just gonna look at a list of blue link kind of MCPs to click on and go into? Check out the Walmart MCP, check out this MCP, check out the Tesco MCP. Which MCP do you want? And ChatGPT is gonna basically, it's gonna be the SERP of MCPs yet again. Is that not interesting? I think it's quite recursive.

23:28Tom Critchlow: Is it going to be the SERP of MCPs, which is a very provocative phrase, or are we going to have new middlemen? So there's this great, there's an academic paper from 1995 about, it says, I think the title is Intermediaries or Cybermediaries, right? And with the rise of the internet, right? 1995, right? 1995, they're like, are intermediaries just going to go away, right? Is everyone going to go straight to our brand? And this academic paper is great because it's just like, it's full of like spicy talk, right? Just dunking on all the other academics at the time who are like, Intermediaries are going to go away. And they're like, don't be idiots. I'm pro-intermediary.

24:02Josh Blyskal: Pro-intermediary, but new kinds of

24:04Tom Critchlow: intermediaries which is a critical piece. And so I think we're at the same point again, where it's like, if you have a shopping comparison website without any differentiation, that's lived on keyword traffic, et cetera, et cetera, et cetera, you're going to have a hard time. Your future is not rosy. But the future picture for intermediaries writ large. I think beautiful. Whole new industries and websites

24:28Josh Blyskal: and businesses to be built. Whole new users to be translated into. Think about these agents as users. It's a huge user group. And they're supposed to take over the entire internet. Right, right. I mean, I think about OpenRouter.

24:37Tom Critchlow: You think in a media way is going to go away?

24:40Josh Blyskal: Look at OpenRouter!

24:41Tom Critchlow: So The OpenRouter for MCPs in every industry is going to be the thing. It's your point. It's like, wait, is ChatGPT going to have access to 25 e-commerce stores, MCPs, and it's going to go through one by one? No, that's clearly not how it's going to work. And yet, We don't quite know how it's going to work, right? So it's clearly not quite that, but nobody can quite pay the picture of what it is.

25:03Jeremy Cabral: What I think is interesting, the e-com thing is right now, I think it's inefficient that there's multiple checkouts. Like what about a unified checkout? Like Stripe, if you're out there, build it. Like literally, it's like you buy slots and it's like cool. And then Stripe's got 3PL and it's like just coming from everywhere. But like one checkout, who owns the customer in the end? Stripe. So that's the end game play. That's the defining question. Who owns a customer? Well, that's actually interesting, right? So maybe if you exchange, like, okay, you can have this slot in this cart in exchange for the customer's data. But yeah, I just think that's where it all changes. Because ultimately, it's about the same primitives. I think that things that matter, 10, 20 years ago in search are going to matter right Exactly. Search. Like the speed of the thing being completed, like the comprehensiveness of the task. It's sort of like Google always calls out like the number of milliseconds and how many pages searched. And I think these things are going be true now into the future. So the same things that ultimately and a task completion, you know, thing. What do we call that? But really what plays out from here is the partnership model is going to become massive. And I don't think many brands are great at partnerships.

26:18Jeremy Cabral: And so what does that new platform and marketplace look like where it's like, all right, my traffic's gone. If you want my stuff, here's the deal. And Cloudflare had a crack. And that's interesting to see what that plays out as a protocol. But I do think that data partnerships become the new game where it's like, it's inefficient to scrape the web. When it's trained on all the world's data, why do you have to keep doing that? And wasting all this compute. When it's like, we know everything already. Like if you've got a thing, the customer wants a thing, just give it to them. And that's a partnership. And like, how do you take the clip on that? Like, yeah, I just think there's entire industries that haven't been built yet that are like waiting to spring up over the next couple of years and ultimately I think the game, the way we talk about it, needs some level of frequency like this pod to really rise up and bring that to attention. And yeah, it's exciting to have companies like Profound that are willing to put a lot of engineering behind helping marketers solve this problem. So, yeah.

27:11Tom Critchlow: But wait, I love, by the way, how glibly we're throwing around the idea that traffic's going to zero. Yeah, yeah, yeah. Just what a moment of time to be living through. Okay. But there's a tension here, right? Because we're all kind of head nodding and talking about massive industries being disrupted, the landscape radically changing, maybe prompt tracking isn't even a thing in six months. As a marketer, not necessarily as an executive or a CEO, they can hold that complexity better. They can look at that roadmap and vision better. They can look across all of their resource allocation. But for a marketing leader or a marketing professional, what do you do? You can't just throw your hands in the air and be like, let's sit on our hands for six months and wait for the future. You have to be doing stuff. But when the landscape is changing so fast, measurement is hard. What are some advice that you have? What are some things that you've seen from some of the smartest marketing teams that are actually still deploying resources, making actions, doing strategies amongst this uncertainty?

28:10Josh Blyskal: Things that we're doing right now, we do tons of A-B testing right now. So we'll take a group of test pages, we'll optimize them for AEO. Basically taking that, comparing to other sites on the website, looking at the delta and traffic growth from bots or from humans, trying to look at different leads generated from both of those. What we want to do is we want to isolate the effect of AEO one way or another so that someone can have some number that they can point to, even if it is, like you said, traffic. I don't really love traffic as the end all be all. I really like decision making or execution or user journey completions. But at the end of the day, not everyone has that kind of data. And it's really hard to do at a brand level. Basically, I'd say impossible right now.

28:48Tom Critchlow: But at the end of the day. Wait sorry that A -B testing just to get into the weeds like let's talk about it. What kind of things are you testing? You've got a bunch of pieces of content. Because Again, there's a lot of bad ideas about there, about chunking and all kinds of things. What does that mean for you when you talk about actually making actions on a page to improve it? What does that look like?

29:08Josh Blyskal: I can't tell you how much freshness helps. Fundamentals like freshness, it's fundamentals like freshness.

29:13Tom Critchlow: It's fundamentals - Wait are we in 2022 right now? Yeah. I can't tell you how

29:19Josh Blyskal: much It helps specifically on the month-to-month to week-to-week level. If you look at the fan-outs, it's really interesting. The query fan-outs tell you what content cadence you need to have for a specific piece. I was working at the bank. They were like, oh, we refresh our interest rate pieces once a month. But if you look at the query fan-outs, it's literally saying Like a three-day date range. Well, you need to work on these in a three-day date range. It used to be January 1st to January 9th, whatever. January 1st to January 3rd. The ChatGPT fanouts are demanding that we do it at this cadence. And therefore, it's going to give us way more bot visits. If we did that, we did that. It skyrocketed because ChatGPT was literally verbatim asking at those cadences what exact dates and what exact placements we wanted.

29:56Tom Critchlow: Yeah freshness everyone knows freshness

29:58Josh Blyskal: I mean chunking to my perspective actually does work relatively well honestly tell me about it I mean literally building the answer in a self -contained chunk so if you're asking about a running shoe I work with sites that have Absolutely abysmal content layout. I mean, they're basically avoid making a clear point. It's very editorial. If you're going to write an editorial piece, if you want to do the comparative listicle thing, which is already declining right now in ChatGPT across the board, but if you want to do it, you have to do it in a way that ChatGPT can extract the insight from the piece of content. We still find that that's still relevant to be important. The schema matters only to the point of showing up in Google right now, right? You show up in Google a little better, you'll show up in Exa, you'll show up in Serp API, you'll show up in all the tools that the AI models use as a derivative. So SEO is still tremendously, tremendously important, has great staying power within the space. But there's still times when I'm working on sites, just moving the JavaScript over into HTML readable content or bringing product stats directly in. We're seeing ChatGPT loves, loves, loves. It is almost basically wholesale rejecting a lot of this comparison And stuff specifically to try to be its own evaluator. It's doing more query fanouts. It's doing deeper research. And it's saying, forget what the rest of the web says. I'm going to form my own first principles opinions about the product through real content, through the product stats, through the millimeters of foam in the shoe,

31:14Josh Blyskal: whether or not it's running for outside or trails or whatever the heck it is. We're literally going to just figure it out for the user and execute on the user's behalf, which is, I think, where we're going. I think about this curve. We're evolving from answers to evaluation. And then from evaluation, we're going to move to action. And that's where the paradigm is going to move.

31:33Jeremy Cabral: So I think just back on your point, something that I know the profound team has spoken about is the marketing engineer. I think that in my view, I agree with that conceptually. And I think Rand way back said something like every marketing team is an engineer, essentially, in this blog post famously. The reality is it's about putting a chessboard in front of me is like just having a piece on the board that like is responsible for this thing whilst we're trying to figure it out. You know, what does that mean? Like a marketing engineer, it's like someone's like T-shaped but can like build anything and support like wherever there's a lack of infrastructure. They can build it on the fly. And that really is the case right now. Do you wanna explain that concept of the marketing engineer? What does that mean for people?

32:17Josh Blyskal: Yeah, I mean, the marketing engineer is literally the person who makes it possible to work with agents, makes it possible to work with AI, It makes it possible not only to find agents and be found by agents, but to use them to show up. So you need someone who understands the analytics, someone who understands how to talk to marketing KPIs. I need to be able to go to talk to a team like a PR team. I need to understand their goals. I need to be able to understand how they do their work. I need them to understand how agents could possibly assist and support that work, but I also need to make sure that I'm doing it in such a way that that work is now discoverable as well by agents as well as humans in a tasteful way. And I think the challenge here is thinking in terms of dynamic systems. You know, These agents are constantly evolving. So the state of the art is always shifting, always moving. And you just really want a DRI who's in there, who can keep plugging and playing and evolving those systems and self-reinforcing, but also kind of trickling that knowledge down into the rest of the org as well. But you really want that like consolidated person. Like we've tried with a bunch of different ways to kind of build this rollout and think about it. And we've tried kind of distributing it, but then kind of no one feels a sense of ownership over owning kind of any of the exceptional information and any like the top 1% stuff. So really kind of concentrating it and pointing at someone and saying, you are gonna figure out how PR is gonna be impacted by agents. You're gonna build agents to that end.

33:32Josh Blyskal: And then you're gonna make us discoverable from a PR lens in agentic world. And then next year or next month The next six months, you're going to think about the same lens, but from our email marketing campaigns or from the same lens for our blog or for our podcast. It's all kind of the same motion, but obviously for different movements, you're going to work with specialist marketers who do all those things. It's exciting stuff.

33:54Tom Critchlow: Yeah, yeah. I want to come back to something that you touched on very briefly, and you've written a bunch of research and publications about this. You mentioned Exa. I personally love Firecrawl. You also published some really interesting research about how Claude is basically just using Brave Search under the hood. Talk to me firstly a little bit about the current landscape of what we know about the plumbing today that the various systems use. But then talk to me a little bit like how do you think that changes in the next six months? Oh my gosh.

34:27Josh Blyskal: I can see you shifting in your seat too. Claude is brave. ChatGPT, unknown. I think ChatGPT is doing their own thing. They've always kind of been doing their own thing with their own index, but there's still like- We kind of think like SERP API, at least for some of it, right? There was SERP API for some of it, but the problem is I can't quite disambiguate Exactly where it's coming from. There's a lot of like, I have a lot of conspiracy theories, which are too informal to present publicly. But for a long time, I mean, obviously we knew that there was like Bing playing a role. And there was like, then, you know, Aleyda Solis put out a great piece about it basically switching over to Google, many descriptions, Google titles. And then it kind of went backwards. And like ChatGPT has basically been blending sources from a bunch of different places. Maybe they have their own providers. Maybe they blend different providers. I don't think they would do that, though, because of latency. So it makes me think that they have their own exceptionally weird kind of index that's just very hyper. It's extremely hungry. It's voracious. As far as any of the other tools, you know, Instinct is using Exa, which is great. Very interesting to see. Parallel and Exa are gonna play a tremendous role in the future for this stuff. I think at the end of the day, the stack for where you get those answers, The index stack is maybe as important as the preferences of the model itself up till the point that I would give you an exception for a browser use.

35:42Josh Blyskal: That is the place where it breaks down too. So like instinct has got its own index, right? But then that all breaks down when you tell it to just navigate to Americanairlines.com or Delta and start doing stuff. And then it can't do something on American Airlines that it can do on Delta. And then all of a sudden it books you a Delta ticket, right? Like index schmindex doesn't really matter then. But you have to find that break point and you have to figure out where it is. I think browser uses 20 channels.

36:09Tom Critchlow: I think that is happening today. Oh, yeah, it is. Like that example you use of people like, you know, trying to use these tools to do real world actions. Oh, yeah, it's real. And then encountering things where the agents like the, you know, for whatever reason can't use Delta, but it can use United or vice versa. That doesn't, no, I think people are reading it. I think, It's hard for people to understand the pace and scale of how this world is changing. And I just want to really clarify that that is happening today.

36:33Josh Blyskal: That is happening right now.

36:35Tom Critchlow: Yeah, it's crazy. So you've got like these blended searches, you've got Exa under the hood for some of them, you've got Brave under the hood, obviously Google, Google's doing Google. But even that is going to change, right? Switch to the hypothetical for a moment. How do you think that evolves in the next six, four months? For all of them. I think the plumbing and the infrastructure to me is so fascinating. Just to wax lyrical for a second. Please. I think Exa is doing some of the most interesting work in this space. Yes. And specifically what I think Exa is doing that's interesting is... We've treated the infrastructure pipe as dumb pipes. And the future of all dumb pipes is that they will no longer be dumb. The price of AI is cratering, right? And so everything that was dumb will become smarter over time. And so you'll still have ChatGPT or the Claude or the Instinct or the Muse, whatever kind of interface and visible layer. But under the hood, it will be calling these things that are as fast as a regular API, right? Tens of milliseconds. But they're doing agentic AI things under the hood. And the example that I have here is Exa Highlights, which I think is one of the most fascinating things, just to get real nerdy for a second. So Exa Highlights is basically And then you're doing what Google has been doing for a long time, which is you get the raw list of search results.

37:51Tom Critchlow: And then you basically figure out from the raw list which pieces are relevant to the query. And then you only return that bit. It's amazing. So that's almost like Google's full index, re-ranking, and sorting, and filtering. The whole process. The whole process inside a single API call. 10, 15, 25 times for one query that is then filtering and sorting and searching. But Exa is so fascinating in there because there's a ton of AI things that go into that sorting and filtering and dynamic highlights, right? And so it kind of, I think, points to a little bit of where Google is even going of... The era of 10 blue links. I mean, not only is currently dead, but like it's been dead for a while, right? And every AI platform is gonna have their own version of this under the hood. I don't know, what do you think that, how do you think that changes? How does the landscape evolve?

38:41Josh Blyskal: I think there's two different worlds where I see, I see like, I think, I see the world where Exa and all these intelligent tools end up being this really intelligent, basically these AI systems don't, ChatGPT, OpenAI, doesn't wanna spend all their time trying to figure out how to deal with spammers and you know, You know, optimizers, folks like us. They don't want people who do all this crazy stuff to be able to toy with OpenAI and ChatGPT's results. And they're gonna offload a lot of that risk and a lot of that consideration over to Exa. And Exa's gonna handle a lot of that risk and consideration. And maybe Exa's gonna offload that There's someone else who helps them build their pipes. And I think the pipes, as you go further down the stack, do get dumber. The things that build excess pipes are a little bit dumber than the hole there. And then the things that build those pipes are dumber than that. And the things that build that is probably like some phone farm somewhere Some scrape of the internet that fell off the back of a truck. At the end of the day It's all, you know, everyone's seen that meme of like how the modern internet is all held together by one domino from one thankless dev back in 1998 who's got this unpaid project that's hosted for free somewhere. It's still, I think, the same, but now you've added 10, 50-ton elephants on top of that massive stack. When they made that joke in 2010, it's massively increased in scale. I think there is a world, though, where You know, marketers are gonna be optimizing.

40:00Josh Blyskal: They're gonna talk about optimizing for Exa. They're gonna talk about optimizing for Parallel, right? Is the same, almost at like the same level of importance or maybe even higher level of importance. Like Exa and Parallel are like, you know, the gateway. And then ChatGPT is the decision. So we model the decision in ChatGPT, and we handle the inclusion in Exa in parallel, almost like you've almost decoupled indexing from appearing at all in search. Does that make sense?

40:23Tom Critchlow: Well, I think the parallel here, back to the kind of classic SEO world, is like, Smart SEOs, when you do stuff at scale in the old world of a couple years ago, you would separate these concerns. You'd be like, okay, is there a problem with indexation? Of course you would. Is there a problem with crawling? And then separately you would look at ranking and so on.

40:39Josh Blyskal: But it's still the same company. It's still the same system end-to-end thing.

40:44Tom Critchlow: Sure, but the conceptual idea of separating these things, right? To build that kind of mental model of like, where is the problem in our stack, right? Where is the problem and how do we optimize each piece of it is right. But to your point, it's maybe just multiple systems and companies now.

40:57Josh Blyskal: Take an example like Reddit though. Imagine Reddit's got a great partnership with Exa, but doesn't have a partnership with OpenAI. Whoa. So you can now be indexed in one system, but you're capped in another system or you're doing one thing and another thing differently. I don't know. It's tricky to think, you know, you'd like for It invades new abilities for things to be much more complicated. But yes, I agree with you. Exactly. You separate those concerns. Go ahead.

41:21Jeremy Cabral: It's interesting, right? Because I often think about in these companies battling with each other, it's the terms of use that creates a wall. So we can't do it because at times you said that. So we have to partner and have somebody reach over the wall for us to get it. And then we have access through somebody else that decided to do that. But often when I'm thinking through each of these kind of barriers, even when I'm building with code, How can I use something to reach over the wall in terms of use of OpenAI or Claude or whatever this restriction that is included? Yeah, and I think that as we start to realize there's walled gardens that all have value, it's like how do we appear inside them? How do we more than appear inside them, like optimized for each of them because we don't know the relative importance because ultimately they're doing these data partnerships between each other and we're not privy to that, it's all behind closed doors. And I think that yeah, we've really got to start mapping this ecosystem in great detail To understand the layers and by which they can or possibly impact And again, the measurability piece? Good luck. It's probably like conversations like privately. It's like, are you guys doing a deal with them? Yeah, we are. That's good to tell. Good alpha here. Let's roll with that for a bit. Yeah, I don't know. This is a totally different ballgame. I've got a question for you, right? Often what we're talking about is like the open web.

42:39Jeremy Cabral: Are you tracking training data and training data movements? Like when people have managed to dislodge or change what's said about a brand? Like how does it take before that, you know, finally three or four months into like, you know, three or four times into a model release? It's like, you know, the latest one finally changes training data. Is that something that you guys have?

42:39Josh Blyskal: I'm very uninterested in training data. Really very uninterested. Yeah, right. Tell me more. Only because it feels... It feels pedantic, at least to my perspective. I would much rather show a marketer seven URLs that they could change in the digest of an LLM and producing an answer rather than seven quadrillion terabytes of data that they have to change. Interesting. Seven quadrillion things that they have to change in the model ways to change the training data. But what I will say is that interest in training data is coming back. I would say that it certainly dipped off. If you ask someone in 2023 how to influence an LLM, they'd say, Training data. And now there was this like big browsing thing and everyone's browsing is the gold standards. But I mean, I was in SEO Shinjen. I've been, you know, across the world and the conversation I keep having with people and might change my opinion soon, might change my opinion soon, is what are we doing about training data? What are we doing about like, you know, there's... There's so much that you can test within OpenAI's API even, right? You can test the, oh God, I'm forgetting it, like Common Crawl, the way that it's pulling from Common Crawl and things like that. And there are people who are influencing that. I don't know how good or bad I feel about it, but I know that that stuff is also as real as asking for an answer when it does browse. The question that we had always come up to is, well, does it browse for the kind of questions that our customers wanna track?

44:15Josh Blyskal: You know, the highly commercial, highly relevant, what products right now for me to buy, that behavior of the LLM used to give us a lot of credence that, yeah, for training data, you know, for like, how do I tie my shoes, really, yeah, whoa, let's try to win it. But for, you know, what's the best shoe right now in 2026, it's more important to browse. But I'm curious how that opinion shifted.

44:39Tom Critchlow: I guess the distinction for me is like, I agree, I think, for kind of most situations and most day-to-day work, you don't really care about training data, right? And it's like, you can't influence it very much. So, you know, focus on the bit that you can influence, the bit that is visible. But the flip side is, you know, I was talking with a company that rebranded, right? And caused a lot of trouble for their kind of like, you know, understanding of who they are. Sure. Even at a just simple name disambiguation level, right? Sure. Again, I showed my gray hair. I started an SEO a long time ago, back when Google didn't crawl the web all the time. And when you made a change, you'd have to wait four or six weeks for Google to recrawl the web and update its rankings and all this kind of stuff. It does feel a little bit like that era, again, for some of this stuff where when a big change happens like that, it takes them a while to actually catch up. And so I do think, I am interested in training data, I think, as that kind of increasingly also We're going to not have this distinction between training data just be informing the ability for it to speak language and do its basics. But more so, it doesn't want to have to fetch the web every time. So there is this point of it wants to know a lot of stuff. But how could it? I mean, they're getting better at it. I mean, I don't know.

45:55Tom Critchlow: I think it's, we're going to see a lot of innovation in like the base. I mean, you know, Google just released back-to-back Gemini models two weeks apart.

45:55Josh Blyskal: I actually, I totally disagree, actually.

45:55Tom Critchlow: Oh, okay, go ahead.

46:04Josh Blyskal: I think computers is going to continue to get cheaper. I think these API calls And these searches are going to continue to get cheaper to the point where basically it's throwaway, basically. That knowing every possible product, knowing every possible brand, the stats about it, the risks for these agents, the betrayal of sending you the wrong product, the betrayal of telling you the wrong time or sending you to the wrong place or getting you the wrong podcast studio. You come here and you're like, whoa. You said it would be here. Oh my God, you're so right. It's New York, Pennsylvania. Oh my God. Good luck. Here's a train ticket. And then it's the wrong train. I think that that Is worth, at least at this stage, maybe we're in Uber in 2011 when rides were $7 and there was a $20 rebate on everything. Maybe it's just free and cheap time, which it is, by the way. It's easy time for money. But my perspective is that the compute is just so cheap. You can't have it both ways where a computer is cheap, but they don't

46:59Tom Critchlow: want to search. So you think the implication there is that every time you ask for something, it will be doing a laundry list of searches under the hood. Find a personal assistant who can find you anything without searching.

47:10Jeremy Cabral: I want this in.

47:11Josh Blyskal: I think it's intent specific, firstly.

47:15Jeremy Cabral: And if it is trained on all the world's information, which apparently by 2028 that's the case, isn't it inefficient in the raw sense of it to have traffic between two servers to try and get an answer? Versus straight from, here's the answer. So I think that speed is the ultimate variable. And that for a lot of queries, they're just going to skip the open web altogether. And through data partnerships, they're going to pre-calculate all the permutations of what it could possibly be and that the open web is inefficient to actually go and crawl to get an answer. The winner is the one that gives you the answer the quickest and it's accurate. Assuming it has those two things and it's comprehensive and all that, then I can't see how the open web is necessary in every single task we completed. For plenty, totally agree with you. It needs to go like, what's the best thing to do today? And it doesn't have data partnerships with all the options. It needs to go to the open web and fetch a whole bunch of stuff. But for things that are fairly baked in with high confidence, high confidence being, I guess, We're talking about 2022 through to, let's say, 2028. They've got a lot of user sessions to train on by then. And they'll know by then what is high confidence. We met the task. The users were happy. And I think for those where they have high confidence that they can complete

48:31Jeremy Cabral: the The task is skipping the open web altogether. That's just expensive at scale. Can you imagine for now and to ever always having to pay for those crawls? I see it's going to taper off. And yeah, the partnerships drive the next piece.

48:46Josh Blyskal: We can hold two things in common though, right?

48:48Jeremy Cabral: Right, yeah.

48:48Josh Blyskal: I think you're dead right, but at the same time, it's funny to see that right now the personal agents, instead of trying to figure out the JavaScript thing of like, let's render JavaScript into VM, they're just like, yeah, you know what? Let's just browse. Let's just set up a browser. The way that they've gotten over that whole problem is just the dumbest, slowest, like I probably could have vibe-coded that, not at the scale, by the way. I could have vibe-coded a small-scale version of that in 2024, and I would have thought to myself, well, OpenAI would never.

49:15Tom Critchlow: There's this theory of disruption, right? Which is like, just because new things come along that are more efficient and enabled by new technology, there's such kind of hangover of the old world that maybe the old world just sticks around and everyone just figures out how to- For a time. For a time, right? But actually maybe longer than we think, right? Like under the hood, it's actually just like firing up a browser, clicking on the legs, because it can, right? And so it just decides like, you know what? That's how I'm going to Solve this user need.

49:37Jeremy Cabral: I actually think that inefficiency is by design. Well, that's the other thing. Where they're actually just training up on how crappy these websites are or are not. And it's just some sort of agent experience score. So, all right, cool. These are the We have a lot of websites that we can work with and now we know we're not going to go hit the browser. We are just going straight to completing this task.

49:57Tom Critchlow: I think we have this today, not to my knowledge, but the idea that right back to your kind of Delta versus United question of like, ChadGBT can go and open up United in a browser and can't for Delta or vice versa, that knowledge should live somewhere. And I don't think it does today. And so if you think about crawling the web, maybe ChatGPT's crawl of the web isn't just the URLs, but it's actually some notion of, is this page agent accessible? It's a heat map.

50:25Jeremy Cabral: It's like, can I use this page to complete a task?

50:29Tom Critchlow: And if not, what is the alternative? Go back to our SERP of MCPs, et cetera. Okay, but I want to kind of bring us back to Earth. The web's dead, clicks are dead. It's all going to be Brad Tom Critchlow, tomcritchlow. Potter chips.

50:29Josh Blyskal: Before you go back on Earth. It is interesting. I think someone should make, it would be a killer LinkedIn post of like the x-axis is like super informational type. The y-axis is super, you know, execution-based tasks. The depth is, you know, how complicated the task. And then the length is something defined by the company, which is like, maybe it's like dimensional, like book a ticket, find out information, whatever. You could almost make a heat map of your site's gaps. Like I could literally see United having like a bunch of red about like finding what time flights are, but then like a huge gap about like finding an exact price for a first class ticket today at a certain time. I think there's something to be done there that's quite interesting.

51:23Jeremy Cabral: Talking about this. There's Claude Code over there in the corner.

51:28Tom Critchlow: There's a lot of ownable frameworks here. But I want to bring this all the way back, because I asked you a question earlier. You didn't dodge it exactly. I said, you know, with all this stuff about the web is dead, as a marketer, you start to make action. And you said, okay, we're A-B testing a bunch of content. A-B testing a bunch of content. Is that the only thing?

51:42Josh Blyskal: No, it's not.

51:43Tom Critchlow: Because I want to get back to this point. It's one thing for us to be sat here talking about the scale and speed of this disruption, but what do you do? What do you do today?

51:52Josh Blyskal: Maybe text a bunch of content, number one. Two, I think that citation bot visits and bot visits in general are an analog for human attention. There's humans behind those bots. You can use that bot attention to basically extrapolate what level of human attention you're getting globally. You can understand that. Some very forward-looking marketing teams are very happy to track that as well as the corroborated leads from that piece of content and understand whether or not those lines are moving in parallel, whether or not they contradict. Oftentimes, more bot attention from citation bots generally means more traffic from users, generally trends positively in traffic. Does it always trend the exact same slope? No. There are tons of times when You know, top 15 inspirational quotes gets oodles of bot traffic in many leads, right? But that depends on the actual CTR. That's all diagnostic, right? So what do you do with that information? What do you do with the information? I mean, you're literally looking at what changes you're making on the content.

51:52Tom Critchlow: But is it all on-site? What about off-site?

51:52Josh Blyskal: Off-site is so important. Are you kidding me? Oh my God.

51:52Tom Critchlow: Okay, so tell me. What do you do?

51:52Josh Blyskal: Oh my gosh. I mean, YouTube is huge. YouTube is going to be a massive new play. I don't like to get into the whole channel thing because it matters industry to industry. Everybody is doing this thing where they're posting Reddit's up 12x and it goes from like 0.0 One to like 0.14. And they ignore it.

53:01Jeremy Cabral: It's like, oh, Reddit's down three percent but we're done with Reddit. Clearly it's not an important source on the internet. It doesn't matter anymore.

53:07Josh Blyskal: But it's like telling someone that it's currently 47.3 degrees on average, Fahrenheit, by the way, by the way, it's like, you know, what, 10 Celsius on average in the world, go get dressed in the morning. It's like, I don't know, I could be in Cabo, I could be in Antarctica, I could be, you know, in the middle of Arabia. That's not a very good number. So me saying Reddit is 6% for the world, It doesn't help anyone. You gotta look at what those offsite pieces of content are, the publishers, the channels. YouTube is usually a pretty big one. And you have to strike some partnerships. You have to start getting in those pieces of content. But it really is quite direct, at least specifically with YouTube. I think literally looking at those YouTube videos has put into perspective for me How direct it should be to join that media ecosystem. It's not millions of followers. It's not hundreds of thousands of likes. Have you seen it? It is normal YouTube vids with really good meta descriptions, really good descriptions, really good chapters, really clear transcripts, multiple languages. They're using all of YouTube's metadata features really well. And they're delivering comprehensive answers about the question. It's like a freaking blog post in video form, but it does a great job.

54:13Jeremy Cabral: You know what's interesting is like the point on like, you know, the 6% down to 4% or whatever. It doesn't serve anyone. I actually think it serves everybody in a way that they need a referenceable snippet that they can take to their meetings that day to show that they're on top of it. And it's so fascinating because the amount of people that I spoke to across so many different industries and roles, like, oh, yeah, Reddit's not important anymore for ChatGPT. They just needed that thing to say that day to be like, oh, I have no idea what's happening, but I have the stat. It makes me look that I'm right on it. That's why it plays on LinkedIn.

54:44Josh Blyskal: Yeah. But the better thing to do is to have your own industry's data.

54:47Jeremy Cabral: You got to have your company, you have to get

54:49Josh Blyskal: HubSpot's data.

54:50Jeremy Cabral: For us, it's this. So here's a question, right? Let's assume you've done all that. Like back in the old game, SEO, people used to look like, you need this many referring domains and this many whatever, and you can dislodge your position one. Sure. Like how do we... Is there any referenceable data on what it takes to dislodge the top mentioned, the cited brand? What are the variables that are in play? Ready for a classic SEO answer.

55:16Josh Blyskal: Depends. The depth of the question, I mean, it's like, it's the same thing. It's like, how long tail is it? How difficult is that prompt slash keyword? How entrenched is the incumbent? Why is the incumbent entrenched? Is the incumbent entrenched because you're asking about Nike shoes and the source is Nike? I mean, at the end of the day, it is some of the same fundamentals with some very small changes that are extremely important to understand so that you can take action. But at the end of the day, I think the effort is It's weird because it's a little more open to change in that the information ecosystem is more utilitarian based. Like there's no concept of like entrenched legacy kind of authority yet, I don't think, in my opinion, in AI Search. So like to the best answer, the win goes. To the winner, you get whatever you're gonna do because you're gonna create the best content. You're gonna get the most citations because you're gonna have the most comprehensive answer about trail running shoes that are available in North America in Q3 2026. You're going to have that. If you're doing better consistently than Nike at delivering that and delivering it in aggregate, there isn't a situation right now. And you can ask Exa, by the way, you can look at how they eval their searches. It was really interesting to learn from their team that it's very much based on the...

56:35Josh Blyskal: Arbitrary answer of whether or not the answer is contained within the piece of content rather than almost anything else. They're just seeking answers. It's very utilitarian compared to what we've seen before.

56:45Jeremy Cabral: So just two quick ones. So assuming we do all of this, how long does it take to get a result? How long does it take to see something change?

56:55Josh Blyskal: It's up to six days on average to get the first citation from a piece of content, but I can't tell you how often it takes to beat someone. So Claude, ChatGPT, both aggregate around six days. ChatGPT is a little shorter, four days. Claude's a little longer, 7.6 days. But it was really interesting to see. We did a bunch of different results. And like the 90th percentile, I think it's like 30-something days. So if you're not cited within 30-something days, it really is a problem. Which goes back to the freshness piece. It's very so important. Yeah well I mean literally 50% of top-sided content is 13 weeks old or less. Citation decay is a real thing. Wait, say that again. 50% of top-sided content is 13 weeks or less old.

57:28Jeremy Cabral: Huge. Wow.

57:29Josh Blyskal: That's insane.

57:30Jeremy Cabral: Yeah. Okay, so last question related to this, because I think it's fascinating. It goes all the way back to the beginning. We're saying prompt tracking is not really the game. But to actually measure if it changed, it is the game, right? And like how many prompts do we actually have to track? Because like, say like, you know, you started with like 100, it's like, how are we gonna know if we made the impact that we needed, right?

57:49Josh Blyskal: It's super challenging question. I think it's like, to me, it's like putting buoys out in the ocean. You can get away with 10, But you're not going to know whether a tsunami is going to come. You're going to miss a lot of things. You can get away with a thousand. You're going to have a real granularity. You might even have more granularity than you even bargained for in terms of optimizing. I think somewhere between a high hundreds to a few thousands, depending on how expansive your business, how many business lines you It's pretty good for most people. If you wanted me to like give, throw you a number, I'd say between 250 to a thousand would satisfy 85% of use cases. Now that said, I work with companies that have hundreds of thousands of prompts, individual prompts being tracked. And I work with companies that have literally 15 prompts that they track and they're small little one person shops. What's the best photo retoucher in Sioux City, Iowa? I don't think you need 2,000 prompts for that. If you wanna know and you wanna do it at a low cost, do it at a low cost, it's chill. But you gotta know that the things that you're indexing on are tremendously important to your numbers in the aggregate. That one out of 15 of your results that day is gonna be dictated You can't go ballistically by that situation. Whereas when you're looking at a thousand, it might be really hard to see changes that you're making in the aggregate, which is why you have to drill down even harder. So we work with teams that have 10,000 prompts. I'm like, you cannot possibly try to report. You can, but you really shouldn't be trying to report an overall visibility score.

59:04Josh Blyskal: It's basically gonna move 0.1 every week. It's not gonna change that much.

59:09Tom Critchlow: Anyway, that's what I think. All right, I've got a million dollar question. Should companies be using AI to generate content?

59:18Josh Blyskal: I think AI slop is a skill issue, honestly. I think that at the end of the day, everyone has experienced slop in some form, but slop is the result of a lack of perspective. It's the result of a lack of opinionation. It's actually probably the result of a lack of a marketing engineer on the team to understand how to combat it. I mean, AI is trained on The whole of everything in the world, right? And it is basically the opposite of marketing in that way. That marketing is about differentiation. AI is about distillation. And so you need to work actively. You need to combat intensely this notion of slop. Because the marketing that ends up working is not often working. And we're talking about the larger marketing world for the same kind of continuous reasons. Good marketing kind of scratches our neurons in different ways and excites people in different ways. There's not a science of marketing where I could say, you show me an ad and I'll tell you, oh, that is gonna get us 73% more leads because of this ad or because of this exact blog post. There is an art and a science of it. I think of it like stringing beads onto a string, that the beads are your perspective. The beads are the things that matter. They're the intellectual property. And the string is the LLM. The string is the AI that just simply makes it cohesive and makes it coherent.

1:00:30Tom Critchlow: Tell me a bit, bring that to life. Because I agree, by the way, the slot being a skill issue piece is a great soundbite. How do you do it personally, right? So like your job at Profound, or at least part of your job at Profound is doing research, And then communicating that research out into the broader world, presentation, a blog post, a white paper, et cetera. I'm guessing that AI helps in a bunch of those pieces. Tons of ways. What are some specific ways that you have, you know, the skill issue piece, like what is the skill that you have used What have you found that allows you to use AI at various points, but reigns it in in a way that maintains your perspective and expertise?

1:01:04Josh Blyskal: I'm not different than most other people, I don't think. I think that I have a very high standard for what I want to do, but I think one of the things that's... One of the things that's tremendously helped me is being very honest about the parts of something confusing that made it make sense to me and never losing that thread. Because that's the thread that you can relate back to someone else. If there's like that aha moment, you capture that. You don't let that get distilled. And so you give ChatGPT all these ahas. You give it all these mental images you wanna create that are non-negotiables. You have to talk about it in this way. We're gonna phrase it in this way. And that's why the beads thing works. It's literally, it's meta because it's how I thought about it. And I'm of course relating And of course I'm doing it. And so that's literally how I would discuss it if I was making a LinkedIn post. It's like, we're going to make a LinkedIn post about how these beads are the ideas. And ChatGPT, of course, because there's 438 different things swimming around in my head right now, it's going to be able to distill that kind of sentiment and emotion into something that's, it's well-reasoned. It discloses information at the right time. I'm whisper flowing constantly. So I'm always going, let's do this, let's do this, let's try this, let's try this. And it's never in the right order for the post, but the ideas are all there. They're all non-negotiable. I think as well as that, you gotta have good context. I think context engineering does a lot to help your brand or business. Whereas a lot of this stuff,

1:02:19Josh Blyskal: this slop stuff just has to kind of Exist as like good assumptions about what your brand, product or service does. But being able to pull on your context, your past marketing work, assertions that you said, you have these tremendously powerful crutches.

1:02:32Tom Critchlow: And where do you store that? Is that stored in Obsidian, Notion?

1:02:32Josh Blyskal: No, it's stored in Profound. We have Profound Context Manager. We store all our brands. Shout out to Profound for being a context manager. There's the plug, Profound Context Manager. That's where we do it. We do all of our marketing work through Profound as well.

1:02:36Tom Critchlow: Eating your own dog food, I love it.

1:02:47Josh Blyskal: We have to. We literally have cans of dog food in the office. But context is tremendously helpful. No one really talks about it as like an antidote to slop, but it's absolutely, for anyone who's executing, it's the thing that you're not thinking about. But the way that you're going to reference an old post or the way you're going to continue a narrative, you don't have to pitch it the same way. And it makes your work just sound more cohesive. yeah. So much, so much, so much marketing work recently. It just sounded like it's a new one-off execution every time someone wants to step up to the plate, which I think is disgusting. I want a brand to be cohesive. I want a brand to sound like a brand. And you can do that with context. And that helps everything kind of move along And then you're just plugging things in and you're kind of relating them. So that's my perspective. A lot of people have different perspectives though. Some people use like, I use 10 AI agents that all yell at each other until it's like original. You can do that stuff too. But I think you have, I think that stuff is way more, unless you're executing at the top 1.0% of the whole industry, that stuff is very fraught. You could get yourself feeling confident about a certain strategy, but then realize that any of the users who are actually reading that content can see right through it. You might feel that it feels special, but it's not special.

1:03:51Tom Critchlow: My strong opinion here is that we're going to move from talking about whether you use AI for the words to whether you're using AI for the ideas. Of course. We're not already doing that? Well, no. I think a lot of people are still hung up on the Pangram scores. They're hung up on, is this sentence written by AI? I don't care. I don't care either, but I think a lot of people do care. But what's fascinating is that even once you get past that, and I've had this in some of my own blog posts, you can use AI as a thought partner And it will suggest ideas to you. But you could then write completely by hand and it's still flop. So the great example I find is I've done a lot of thinking and writing about management theory and going back to previous industrial revolutions, technologies to try and figure out how do you frame the current moment. AI will tell you All the time, about the rise of electricity, and about how factories, when electricity came on, and motors came on, and old factories, like change didn't happen straight away, because old factories couldn't adapt. And it's like, it will tell you that. Yes. Over and over and over and over again. It's like a broken record. It's like that one friend of the party that used the same, same story. Like I heard this story a million times, but you could still write that piece by hand, right? And it's still going to have that slot in it, right? And so I think that that's where the discussion is going to go next is like, oh,

1:05:07Tom Critchlow: it's not the smell of the words. It's the smell of the ideas. The idea is strong, right? Is the actual, the idea behind this piece crisp and clear and original, or is it slot?

1:05:17Josh Blyskal: Yes. That's what tickles our brains nowadays, a new perspective.

1:05:20Jeremy Cabral: Yes. Yes. Yes. So a brand marketer's got, like literally, they're in a list with you and like, please, I'm desperate here. Like, help me out. Three things, tell me. What do I need to do to win an AI search?

1:05:20Josh Blyskal: Oh my gosh. You got to have data. One, you have to execute like hell. Two, I You got to feed these agents. These agents are voracious. They're hungry. Three, you got to bring real quality content to the show. You got to be making things that are worth answering questions about your brand. And you got to own those answers in a way that is discoverable and actually helps people consume, buy, understand you. You're not trying to create slop content. You're not trying to cover things with s***ty wallpaper. You are trying to answer. You're trying to do that answering at a very high execution level. And you need data on whether or not that's working. That's my perspective right now.

1:06:03Jeremy Cabral: And so who in my company should lead this? Which function?

1:06:03Josh Blyskal: It's your marketing engineer, man. That's what it is.

1:06:08Jeremy Cabral: So you see it as a marketing leader that's ultimately-

1:06:08Josh Blyskal: I do. I do. I see it as like an agentic ready marketing ICP or low level leader. Exactly. I think that that person is primed to understand how the agents work, how to use the agents. And how to understand the data to get the signal they need to execute on this. If there was one person at a company that was the most exceptionally built to do it. And I think those people by and large, the heritage is SEO heritage. It's an SEO heritage. And I don't see that lightly. You can check any other podcasts I've ever been on. I came from the SEO industry. I do this work. I have not seen a professional class be able to graduate. Their job is to intermediate between these complicated machines. They've been doing it for 20 something years now, the good ones exactly and okay so whether we call SEO or GEO more or less of those jobs in the future? I think that's all right. The funny thing is I was reading this book about shamanism and it's all shamanism. And I think there's going to be more and more of it as we see more intermediaries. Wait, wait, wait, wait, wait, wait, wait, wait.

1:07:11Jeremy Cabral: Back it up.

1:07:11Josh Blyskal: Digital shamanism, Yeah, I want to explain that to me. In Mongolian shamanism, we're going to There are two kinds of shamans. There are the white shamans and the black shamans. I was reading this and there's nothing innocuous yet. I'm like, okay, two kinds of shamans. The white shamans are wishing for good luck and good things to happen. And yeah, keep the rain coming, you know, keep the clouds happening. The sky is so blue, let's keep it blue. The sun is so yellow, let's keep it yellow. The black shaman are the interesting ones because when something goes wrong, someone breaks a bone, someone has a seizure, someone falls sick, the black shaman goes up to the top of the mountain and basically negotiates with these capricious deities, these gods, and kind of begs these gods and tries to like figure out what they want so that the good outcomes can happen. And I was like, God damn it, that's SEO. Please Google, please. Let's rank. What do we have to do to rank? Your first born child. I'll take it.

1:08:06Jeremy Cabral: Well, you prayed to Google. Who do you think wins the game?

1:08:09Josh Blyskal: I think Google still wins the AI game, by the way. If you want my honest opinion, I think that Google has to sort out what it's going to do with its ads business. But once it does that, it's going to, all that Gemini stuff is just going to come right into the serve. It's going to be already generative. I think the only reason you're not seeing a generative UI on Google search is just because it's search ads right now.

1:08:27Jeremy Cabral: I think we should not, sorry, I don't think we should underestimate Elon. No, I don't think we should. There's something about infrastructure, which like, Man, everyone's building on top of infrastructure that's rented land. And ultimately, these satellites going around, electricity, all the things Elon's solving for. I mean, he co-created OpenAI. It's great, right? So I think he's going to come out as a dark horse and win the game.

1:08:50Tom Critchlow: Is Elon the black shaman of the AI world? I feel like Elon is more willing to do a deal and do the dirty stuff that a lot of other people are not willing to do.

1:09:00Jeremy Cabral: I think he's going to buy out one of the key players that allows him to have index of the web. And then like what's left, right? Like if you inside these platforms have like document creation and email and whatever, but that's just commoditized at this point. And I think that you can suddenly like, you might have had the system, which is like a data mode, which if you migrate out of very quickly, Because, you know, DropBot or whatever it is solves I reckon Elon comes out of nowhere and wins the game. With infrastructure being his unfair advantage.

1:09:26Josh Blyskal: It's true, maybe.

1:09:27Jeremy Cabral: Yeah, let's see. Play forwards a year or so from now.

1:09:31Tom Critchlow: All right we should do some rapid fire questions for the wrap up. Do you have your phone? All right, let's do some rapid-fire questions. First one, what's your AI stack right now?

1:09:38Josh Blyskal: Profound. It's got to be Whisper. It's going to be Claude. Love, love Cursor. Huge user of Instinct. Muse, GrokBot as well. Did I say ChatGPT as well? Maybe ChatGPT. Do you use them all the models? Astra 6. Ultra is insane. I yell at it to deploy six sub-agents every single day. It can only deploy six sub-agents per query. So I say, to finish this, deploy six sub-agents. Find me a place to eat dinner. Deploy six sub-agents to complete this task.

1:10:02Tom Critchlow: That caveat is going to get longer and longer. Deploy six subagents. Make no mistakes. Deploy six subagents. Check it out. It works. Anyway. If the CEO could see one AI search number every Monday, which number would it be?

1:10:15Josh Blyskal: Bot citations.

1:10:16Tom Critchlow: Okay. Okay. Wait, like bot crawls?

1:10:18Josh Blyskal: From citation bots. Not from just general bots, but from citation bots. Spicy.

1:10:25Tom Critchlow: I don't agree, but we haven't got time to get into it. When you see your brand started in an AI answer, what do you do next? Do you screenshot it? Do you screenshot it? Like how do you represent, how do you share that around team? Like a screenshot?

1:10:25Josh Blyskal: I always screenshot it, yes.

1:10:25Tom Critchlow: What's one change you made to a website in the last year for AI agents and what happened?

1:10:44Josh Blyskal: I built an accessibility tree for a website that didn't have one. I don't think it actually depended on the, it did that last week actually. And I don't think it actually impacted it very much.

1:10:53Tom Critchlow: Interesting, okay. Josh can't do AI Search, you heard it here. No, I'm kidding, I'm kidding, I'm kidding. If you hire someone to own AI Search tomorrow, what's the job title and what's the one interview question you'd ask?

1:10:53Josh Blyskal: Marketing engineer and what do you know about agents? Okay.

1:11:11Tom Critchlow: Most overrated thing in AI Search right now?

1:11:11Josh Blyskal: Listicles.

1:11:11Tom Critchlow: Most underrated?

1:11:11Josh Blyskal: Documentation.

1:11:11Tom Critchlow: What's the worst advice you hear in AI Search?

1:11:20Josh Blyskal: Thank you so much.

1:11:25Tom Critchlow: ChatGPT wrote this, by the way.

1:11:27Josh Blyskal: The worst piece of advice in AI Search right now is that you can just, like it's all these people are just like, just f***ing create scaled content 24 seven, just go nuts. You've got to have some taste. You got to be really opinionated and thoughtful about it. I think there are people that are being led right now into chaos. So you've got to be very thoughtful.

1:11:43Tom Critchlow: You know, I always watch out for my fellow SEOs. I love that. I love that. People being led into chaos. I agree. What is the book, essay, or paper you sent to people the most?

1:11:43Josh Blyskal: Anything by Paul Graham. It's quite good.

1:11:43Tom Critchlow: Okay. What did you believe about AI Search 12 months ago that you've changed your mind on?

1:11:43Josh Blyskal: I think one of the things is the browsing thing. I fully believe that browsing was gonna be the only thing. I continue to be confronted with people who wanna talk about training data. So I think by reason of people's interests and my appreciation of the market, I've been wrong about that thus far.

1:12:03Tom Critchlow: Okay, good, I appreciate it. One sentence on a billboard outside Google or OpenAI, what does it say? One sentence on a billboard outside Google or OpenAI? Presumably for profound, but it could be for any other project you're interested in, I suppose.

1:12:23Josh Blyskal: Don't f*** it up, how about that?

1:12:23Tom Critchlow: Josh, thank you so much. What a great podcast. I love it. Thanks for listening to the show. This is the AI Search Leaders podcast. For more episodes, go to searchleaders.ai and like and subscribe to follow along.

Get the newsletter, too.

AI Search Leaders on Substack: Tom Critchlow on AI search and the people working it out, by email. Free.

Subscribe free (opens Substack)

Hear the show. Then join the room.

The podcast is free. Membership is the community, the expert sessions and the tools, with 150+ senior search people working out AI search together.