S2E3: "Building for Machines, Not Humans" | Kevin White

Alright, welcome back everyone to the GTM Engineer Podcast Season 2, from hype to operating system. We all know that GTM engineering is here to stay. So again, we're done with evangelism and we're here for some sensemaking. So today we're talking about something that's fundamentally changing how GTM teams think about, and that's discoverability. And we have nothing but Kevin White, the head of marketing at Scrunch.ai. And if you don't know him, he was head of marketing at Commonroom, Retool, and Segment, and he's advised companies like Ashby and Deepnote. So the guy has seen the GTM stack evolve across multiple eras. And now he's working on something that again feels like the next frontier, or at least for people like myself, you know, especially in the last few months, which is how your brand shows up inside AI answers and LLMs. So not just Google, not traditional SEO, but the actual responses from ChatGPT, Perplexity, Gemini, and other LLMs. Which is what is ultimately giving your buyers, you know, like the actual advice and actual answers when they ask about your category. So we're talking about GEO, AEO, or, you know, God knows which acronym the industry is hovering around. We'll probably talk about that also. And yeah, like one of the takes of Kevin that is super sharp that we're gonna deep dive into is that bots crawling your site are almost actually your new VIP visitor. So, which means that the playbook for getting found is being rewritten almost in real time and optimizing for machines more than humans maybe. So we're super amped to have Kevin on the show and deep dive into all of this. Thanks for being here, Kev.

Yeah, so thanks so much for having me. I'm a fan of the GTM Engineering School. I'm happy to be part of it and help out here. So, I really appreciate it. Let's get into it. Awesome.

So, I mean, like, let's start with some sensemaking. So we know that GTM engineering is not new anymore and is hopefully here to stay. I mean, like, I think we actually know that by now, but in your take, Kevin, like being in the trenches, like how is been changing, like this discipline, this new discipline has been changing in the last year or so, especially?

Yeah, I think some of the stuff that I've seen from, you know, when we were, I mean, it's hard to even say like how old go-to-market engineering is, you know, maybe, you know, 2 years max, maybe a little bit longer than that. And then I know a few people kind of thinking of it as RevOps, but I definitely think it's like a really different discipline. Right now, and I think what we've, we've kind of seen evolve is, you know, the typical kind of signals or enrichments that you would use for go-to-market engineering, like using something like a Clearbit or using something like a Clay or using something like a Commonroom to like, you know, get all of the criteria that your sales team would need to make an informed prioritized decision around which account to reach out to, or, you know, someone coming inbound and they use a Gmail and like you wanna enrich all that with like the firmographic criteria. Like I feel like that is kind of like not the old school way, but just like the, where, where kind of this market kind of evolved and, and started is like, oh, there's like so many things like that you can use for both enrichment and then like finding different signals to enrich, enrich a lead or an account with to like, um, do an informed approach to go-to-market. I feel like that's kind of like where we started and now Where I'm seeing it evolve to, and this is like really exciting ground for me too, is that, you know, when I get asked, you know, what is the, especially coming from Common Room, you know, everyone's like, you know, what are the best signals? What signals should I use? And I always like to think, okay, let's reverse engineer this to see like what signal would put your buyer in as a, like, they have commercial intent. They're like, it's a hot signal. It's the right person, right message, right time that you can send them something. So what What is that trigger or what is that sequences of triggers that, you know, puts them into a high-intent category? And oftentimes like that is not something that comes out of the box. It's not like, you know, a website visit. I mean, that's probably like a lagging indicator of one of these triggers, but it's something that's more like, oh, this, they did this activity within your product or they did this activity within a GitHub repo or, you know, some law passed and that is really going to impact like this person's business. Like those are the things that like just don't really come out of the box. And so if you can really get to what is the— I think, I think this is where go-to-market engineering is going to be most effective is like if you can really identify those things and then, uh, identify what that trigger is and then like find that trigger at scale, like that is where, you know, you'll have the, the most outsized like alpha for lack of a better word, of, of getting actual meetings and conversions across the line. And some of the things that are really cool that are able to, some of the tools and products and technology that we're seeing that's able to enable that are things like Claude Code. And, you know, of course, like Clay and Common Room also have some of these functionalities, but I think, you know, there's a lot of teams and individuals who can just do this on their own without having to purchase software because of all the AI tools that they have at their fingertips for like, you know, $20 a month or something like that. So I've seen a lot of those, like the AI tools and LLMs evolved to a place where they can get pretty sophisticated where you're like, hey, find me all the real estate companies, houses that are being sold in this area that have like, are going to be impacted by, you know, this real estate law or whatever. And like you can carve out this super niche and reach out to those people with something that is like so compelling that they can't say no to. And like, that is like what AI is really good at going and sussing out. And I think that's like where the evolution of go-to-market engineering is gonna be. It's like almost everything tailored to the specific, a really small cohort of people who you're reaching out to. I love it.

So again, the good old signal stacking is getting more and more accessible and cheaper, right?

And bespoke, I would say. Yeah.

And bespoke. We're seeing this also with, you know, having things like Clay and in ChatGPT or Anthropic or even like Excel on, you know, like Cloud Code MCP is like super, super easy to access and you don't need to build these tables and stuff like that, right? Anymore, almost.

Yeah. Yeah. It's, it's, uh, the, the barrier to entry there is getting a little bit more accessible. Um, I mean, I would always encourage people to, you know, not listen to us or not like watch YouTube videos and just like go try it on your own. And that's like, what's going to get you like the experience you need to like, you know, toy with these platforms and like figure it out and like, you know, spend some time there and you'll learn it really quick.

Yeah, big time. And if GTM engineering disappeared tomorrow, what would you miss the most?

What would I miss the most if go-to-market engineering disappeared tomorrow? I mean, I feel like it would just kind of go back to, I really love the like nerdy, sophisticated stuff of like, hey, like we don't know that much about this account, but like there's a lot of information out there. Like we can go find that information in like this really sophisticated, like, you know, we're gonna, build all these workflows and like go find all that stuff. I get the joy out of actually like building and hypothesizing and running those experiments. So like, if it was just like not possible to do that stuff, that would be a huge bummer for me. I do think it would be kind of nice though, in a way where, you know, maybe just go pure, pure back to like brand marketing. This is the way to, this is the way to reach people. It's like Mad Men style, Madison Avenue, like that kind of pure advertising, like David Ogilvy style. Like that's kind of fun too. So, so I feel like I could work in both worlds, but I certainly like the go-to-market engineering world. I would say it's more favorable. I prefer it a little bit, you know.

If you're a really curious person, you can find out anything now. And that's powerful. And then you could figure out the right time to reach out to people and how to tailor your message. And yeah, if that were to go away, yeah, we'd all have to buy billboards again. But what failure modes have you seen with, with go-to-market engineering, or what would you say some people early on maybe expected that you can do with it that it really doesn't solve on its own?

Yeah, I feel like there, it's maybe more of an emphasis on the software because the software engineering or software engineering, the software companies are the ones out there communicating the message to the market of like, you need this. And so a lot of that is, what I mean by that is, this, the software for go-to-market engineering is a lot of like, enrich this contact, like find this signal, like build this workflow, which, if you think about like the two parts of the equation of, you know, how you're going to actually like reach someone with an offer that is compelling, you're just finding the audience to reach in that sense, and you're not crafting the right message to them. And so a lot of times, like, that part of the workflow goes missing. And so you tee up all these really amazing groups of accounts and target ICP, and then like you hand it over to your outbound team or someone and they send them a message that just goes like, falls completely flat because that part of like coaching and understanding of like what the message should be is, it goes completely missing. Now, I know there's like, there's tools like, you know, whatever the SDR tools and all that kind of stuff you can like feed into. And so that kind of like gets your message, you know, a little bit more like cookie cutter or like acceptable, passable if like a marketing or go-to-market engineering person's writing it and like putting in all the dynamic insertion and all that kind of stuff. But I still think that, you know, especially for like the really like enterprise style accounts, like you almost need to like get all the information upfront to a rep and then have them have some autonomy over like what that message should be. And if you are just handing that over to the reps without like training them on like, here's what this means, here's a compelling reason to reach out, here's, you know, why this is painful for them, all this kind of stuff. And they just send them like a kind of like milquetoast email or LinkedIn message or whatever, like that is not going to convert. And I feel like that is something I saw a lot, especially when I was at Common Room, which is like that part of the handoff was not addressed enough.

And why is that, do you think?

Um, you know, I, maybe there's not enough, I mean, I guess there's like the outreaches and Artisans and 11Xs and whatever over the world. I feel like they would do some of that education of like, here's how to write a solid email. But I think it's probably, to me, I think I would diagnose it as maybe more like neglect on the importance of an, um, SDR/BDR team. It's like you're hiring these junior people, they, you know, have a smaller OTE, like they're sometimes viewed as like dispensable, which I think is inaccurate. I think that team has, that's on the front lines, that's generating pipeline and generating discovery calls and all this kind of stuff is so critical and important to a business that I think a lot of people overlook it and they're like, oh, I can just have AI go and do this. But really, really, if you invest in the right people, if you hire the right kind of talent and you hire the right people, people with the right kind of mentality and then know how to use these systems. To their advantage, like that is like just as important, maybe even more so important than the other side of the equation, which is like teeing up the right audience that has the right signal. And so I think it's maybe just like that part of the organization was viewed as like in the predictable revenue kind of model, which is like, we're just gonna hire a bunch of people where it's gonna be sink or swim. The ones that make it, like we're gonna keep and we're gonna cut the other ones. And like, this is not a, and a department that we want to invest in. And that's kind of like carried over, unfortunately. But I do think, you know, hopefully the tide shifts a little bit there and people realize that, you know, this is a really important part of the, of the org that's like critical, a critical driver of the business. And like, I certainly see that. We certainly saw that at Commonroom and then also, you know, see that at Scrunched too. And so I think that's kind of like the one thing that I've learned in being in the space is that like invest in those people, hire the right kind of talent, and like you get really, really great outcomes.

And do you, do you see the most successful companies continuing to invest in the SCR team or say, hey, we're gonna automate everything and, and, and, and we're gonna cut that team in half?

Yeah, I think the, the, the success, successes that I've seen are almost like you hire the, the people to do the work. Like it, like the Paul Graham quote, like do, do the things that don't scale first. And so you hire these really smart people that are able to, you know, really understand the audience, come up with a compelling message, and eventually patterns emerge and you can repeat those patterns over time. And so once you get to that like pattern recognition stage, like that's when you can pull in technology to kind of like do that work. It's like, hey, you know, Claude Code or whatever, you know, product, once you're shadowing me, I'm doing this over and over again, like just figure out and go do that work for me. And then that gives them another frontier to go after. And so you're kind of like stacking these different plays on top of each other. Is what I've seen work. And so oftentimes even like, yeah, you wouldn't hire like an army of that team. You wouldn't hire like, you know, maybe instead of dozens of people, you hire 3 or 4 really, really great ones to start. And then they can kind of like start replicating these plays. But like, it's kind of funny that, you know, once that starts happening, you know, you get the addiction to the pipeline creation and the revenue. And so the, or the efficiencies there. And so, It's like you don't, you won't hire more of those people. Like there's, you want to get as much pipeline as possible. And so like if something's working, it's like, oh, just hire more and more of those people. And so I think that's kind of an analogy for AI in general. It's like, it's going to open up lots and lots of different opportunities that we just don't even know about. And I have seen that with, you know, hiring the right kind of SDRs. It's like, oh, we only need 3 for now, but then, you know, you hire more sales, like you need more pipeline and then it's like hire and repeat and do the same. So, while the, maybe the ratios to, from SDR or BDR or go-to-market engineer to a rep have changed where it's maybe like 1 to 10 or 1 to 5 now instead of 1 to 2 or 3, I still feel like once, once that engine's going, it's like you, you want to, um, hire it or sorry, um, scale it until, until as long as you can, you know.

Is there, from your experience, uh, maybe especially at Common Room, were there certain kinds of companies or certain segments that were really able to explore this and have more success with go-to-market engineering? Were there some segments that were more lagging behind? Like, what, where did you see the pockets of success?

It's hard to say really. I mean, we worked with a lot of dev tooling companies at Common Room. Like Common Room is like very strong in the dev tooling space because of the, how much rich data and signal data is in GitHub repos. And so that was a signal-rich and like even data-rich environment, our company that we saw like have a lot of, a lot of success. I would say that, you know, maybe some of the like the laggards type of companies that are maybe in like manufacturing or, you know, something really niche, like they were behind, but like, I don't think they were behind because there's not data there or there's not like go-to-market engineering to happen there. I think that they just, you know, didn't, don't have the, you know, they're not at the forefront of. The market, but I think like the ones that do figure that out, there's almost like greenfield, crazy untapped opportunity there to like look, it's crazy how many like APIs are public out there that you wouldn't even like know or think of, or just like data repositories. And so I think that those types of industries that you wouldn't think of as being like, you know, really ripe for go-to-market engineering are actually ripe. And you know, if I was a, if I was a, you know, joining a company like that, I would be like, oh man, I can, I can join this company and like really just crush it. And like none of our competition is even going to figure this out for another like 2 or 3 years. So yeah, I would say it's probably just like the normal crossing the chasm kind of like learning curve, but, um, there's opportunity to cross all, all industries in my view.

You've mentioned Claude Code a few times and, you know, there is this hype cycle, right? And we, everyone was talking about Clay, and everyone's talking about N8N. Now everyone's talking about cloud. Is it justified? What have you seen people move away from and pick up cloud code and build systems internally? And does this, does this actually, is this actually part of the go-to-market engineer's role? Is this across the company? How do you, how do you think you sort of manage this in an organization when everybody's sort of building their own applications?

Yeah, I mean, next month I'm going to be talking about something different for sure than Cloud Code. Like, this is just what I've been using right now. So this is like date this episode or whatever. Uh, I am, I'm finding it's kind of, it's, it's cool to use because it's, it plugs into like different things. Like it plugs into different APIs. You can like load in like spreadsheets and all this kind of stuff. And like the, while you can do that on some of the other, you know, I was a heavy ChatGPT user too, and I feel like Cloud Code is more sophisticated in the sense of like connecting the dots of like all these different things, like your API to your spreadsheet to whatever system and like coming up with like really interesting like data enrichments, like that kind of stuff. And so, yeah, I mean, shout out to, um, I, I'm sure you guys know Jordan Crawford. Um, he kind of like got me into it and I've been like tinkering with it ever since. So, um, that, that's the, that's the, um, model or the product or whatever du jour. Yeah, something's gonna, something's gonna displace it in a couple weeks, I'm sure. I'm sure.

Would you mind sharing like a specific use case that, that you've had success with?

Oh, sure. So this is something I worked with Jordan on actually, which is we wanted to, so, so at Scrunch, like we, the product we have at Scrunch is we provide AI visibility for brands. So like, what are people prompting and is your brand showing up in those prompts? And so One of the, some of the proxies for like what the signals for like if company would be impacted or have pain points around that would be if they have a really strong organic search footprint that is declining, like that traffic is going away, or if they have a big ad spend, they're spending a lot on organic keywords. They probably want to make sure that those keywords are, sorry, paid keywords. They want to make sure that those paid keywords are also being represented in LLM platforms because users are migrating to those. To, um, you know, get answers. And so we, uh, went through this like whole exercise of using Claude Code to, um, craft a compelling message, go out and ping like, uh, these different— forget what, I don't know which API it was. It was like, it wasn't SimilarWeb, but it was something that like does that kind of, um, like will find paid search spend for different companies. And, um, we use that as a list to like merge. Here's our target account list. Here's the accounts that seem like they are experiencing the most amount of pain and then like draw the line somewhere and then that becomes like our target account list for reaching out with this really strong value proposition. And so that was like a use case that we came up with. And I think the point there though is that, you know, you're not bound by creativity. You can just kind of be, think about like whatever the best possible outcome data source is possible and then work backwards from there. And then a tool like Cloudcode is pretty good at like going out and finding something that'll work. You know, whether it does that by brute force, by just like crawling a bunch of different pages or using some API that's out there or something like that, like the AI platforms are getting sophisticated enough where, you know, you don't, you almost don't even need to buy software sometimes to, yeah, I mean, I guess that, that, that's technically buying software, but it's like a personal use case. You don't need to buy SaaS software to like get this, you know, really rich signal that's bespoke and niche and like, You're the only one that really cares about it. So yeah.

And then now who, I'd be curious here now, who's going to maintain it going forward?

Oh, I don't know. I don't know. That's a good question. Probably not me. Uh, no, I mean, I'm hiring. I don't, I don't have someone who I'm just going to like hand this off to.

Um, so it's like no one comes to mind, but no, I asked this question because I found myself creating lots of bespoke apps, I'd coding, uh, Cloud Code and, and, and other other platforms, and now I have this like whole army of, uh, of these, these applications. And now like, I, I feel like I need a team to like, okay, go change this. Go change that. Yeah.

I mean, I do think, um, I forget what the name of the title of this like role would be. It's like an AI architect or AI czar or something like that, that is just kind of like overseeing all these things. We actually have, we, we started doing a lot of like, um, vibe coding tools that, um, with, with Cursor, a guy on my team, his name's Andrew. He's phenomenal. Um, He, uh, is building things like a prompt generator. We're building like a site audit tool, all this kind of stuff that actually looks, it like takes part of our products using the API. Um, it's, it's also taking like prompt workflows and stuff like that. And it's like creating this like really, really compelling tool that is like used as an offer. And so like, who's going to maintain and manage all that stuff? Like he's the one that's doing that for now. But I think at some point you bring in someone who's just like, hey, here's all the like AI things that people are using. Like let's, wrangle all this stuff together and make sure that we have some redundancy here. So if someone leaves, like it doesn't fall over, you know?

Sounds such like a fun role.

It's super fun. Yeah. Yeah. Yeah. I love it. Although I wish I was, I wish I had more time to do it, to do all that, that kind of work. I mean, I carve out a few hours here and there, but I feel, I feel jealous of these.

It allows you to go like 100 times as fast, but then, then that just opens up 100 more ideas. And so it never ends.

Yeah, I think my nightmare would be creating an army of content creators and then like through AI, and then I'd have to just review and edit all of their work. Like, that sounds like my nightmare. So, but theoretically, it gives you like a lot more leverage, right? Yeah.

I mean, like, just on this, I remember I recently heard Elena Vernau on the latest Lenny's podcast and she was super hyped about getting, having hired a vibe coder just for herself. So there is like a dedicated person at Lovable that just vibe code apps. And as part of, you know, like their growth endeavors and, you know, again, would be such like a dream role for like a prosumer builder, you know, like, especially like a young folk, you know, like that just wants to get their own hands on dirty and It could like just be even for like junior talent, you know, like such a, like a significant step change in how you actually ramp up into a role or ramp up into a company and blend different skill set together, right?

Yeah. I mean, if I had the choice between hiring a VibeCoder to build really cool tools and someone on the content side of things who's going to write a bunch of like white papers, I'm definitely going to hire the VibeCoding person 100 times out of 100. So. I think that's kind of like the new way of doing things. And that makes it sound bad, like content is not an important thing. I think content is extremely important, especially to get visibility with LLM within the AI platforms themselves. But I think as far as like offers go and for demand gen, like the tooling is like much more compelling of an offer of like, you know, you type in your domain and you get like some really cool, like, you know, results or something that is surprising. You didn't know like that. Is better than really downloading some PDF that you're never going to read, you know. Bring it on.

Yeah. Optimizing here for creating unique, valuable offers that are interactive, that are target-specific versus sort of, you know, generic content. Yeah. That's what, that's what's possible now. Amazing.

They also become really great tools for outbound actually for, for GTM engineering. It's like we're, we're creating some tools where it's like, Hey, if you have target accounts, like run, run all of your target accounts, domains through this tool, this will produce like a report card for you of like, you know, how that brand is performing on different AI platforms. It's going to surface things, surface gaps, content gaps, citation gaps, you know, websites that are parts of their website that are blocking AI from, from crawling it, all this kind of stuff. You hand that over to a CMO and the CMO is like, oh, I hope I don't hope no one else above me sees this. Like, like let my, uh, hand this over to the demand gen or SEO or whoever person, and then like task them with fixing it. Like that's a pretty compelling offer to, to share, to, to get a meeting. Right. So, so I think that like the, these tools both work for marketing offers and acquisition, but they can also work on the outbound side too.

Yeah. Scruncher is enabling a lot of those, right?

You guys are both talking at the same time.

Sorry. No. And I think it's like super fascinating how, you know, like these exactly vibe coding specifically is blending goals like you, like use cases and outcomes, right? Because like, if you think about it, as you said, Kevin, you know, like if you build like a website grader, you can programmatically embed that into outbound campaigns. So a Claytable can run a list of accounts through these website grader logic and generates, you know, like mini sites that then get attached to outbound emails as offers, but then they also become programmatic SEO plays because you have, you know, like they index, they become enablement assets because reps can use them in combos. They become brand assets and even exactly again, SEO assets or product marketing assets. So like, it's fascinating how, you know, like these use cases are also like converging with, with vibe coding.

Yeah. Yeah. If you can, if you can think about it, from that like systems level and how it all kind of trickles down or impacts like all different facets of go-to-market, like that's the best case scenario. But I think, you know, once you get the itch, like once you create something, I found like, oh, this is a cool thing that we created. It'll work here. It'll work here. It'll work in like all these different places. And so like creating is the first, first part. And then after doing that, like all the other stuff kind of clicks. And I think the like North Star is just like, is this tool that we're creating actually going to be helpful for our core audience? And does it have some sort of like tie-in to our product? And if you like check those two boxes, then essentially like it'll work as an offer across like lots of different things and have lots of different surface area that you can cover with it. And so that's kind of, you know, we did similar stuff at Common Room. We did like a playbook generator. We have a prompt generator at, um, at Scrunch. We have this, uh, site audit tool. We have, you know, something that'll show you how AI is going to crawl your site and like kind of like Give it a, give AI a task and it'll show it, it'll show like AI trying to like get through that task on your site. Like these are all things that are like fascinating for the person on the other end who you're trying to sell your product to, to see. And I think it just, you can surface it in like so many different areas. And so yeah, once, once you get that kind of like initial kind of like, oh, this is really helpful for the people who are trying to reach, it just kind of like, you know, permeates through like all your different ways of thinking. I was like, I almost thinking, I almost thinking like what, How can I, what tool can I create or how can this like, parlay into something else? Because you're using like the same kind of like foundations of like data from like one tool to the other too. So yeah, it all kind of, it's interesting how that all works. Yeah.

I've gotta start playing with some of this and I gotta put go-to-market engineer school into Scrunch AI, see what happens.

Yeah. Yeah. That's, do it.

But first I gotta ask all these acronyms, AEO, GEO, obviously SEO, AIO, LLM. Oh, I, there's too many of them. I can't even, well, do you have a favorite? What do you call it?

Yeah, we, um, have kind of been agnostic about it. I mean, whatever the market chooses, the market chooses. And like, you know, hopefully 2 years down the line, we aren't, we aren't having this discussion, but we just kind of chalk it all up to AI search, which is maybe, uh, not so great for our actual AI search results because people are searching for AEO or GEO or something like that. But pretty much I've, in anything I've been, been writing, I say like, this is AI search. And then in parentheses. AEO/Geo. So like, it's my way of writing that in parentheses is just to like kind of chalk up like how silly it is that like we're trying to debate around this like 3-letter acronym that hopefully is not the acronym that gets chosen over time. I don't know.

No, that's funny. That's funny. Cause that's actually what we did because we have a module that focuses largely on it in the course. And we did the same thing. We called it AI Search.

Yeah. Some people have like a philosophy around like they're 2 different things. And I'm just like, I mean, you might have that philosophy, but But like the, the, the broader market wins in the long run. And so like everyone just kind of views it as the same, like a synonymous thing. So like AEO and DR are pretty much interchangeable. It's kind of like demand gen and growth. Even like, yeah, you could argue like demand gen is a different, like more ABM hands-on type of thing where growth is more like engineering heavy and centric, but like oftentimes the two things are interchanged and job recs or whatever, or title people are hired and they pretty much do the same stuff. So. Yeah, it's, it's kind of one of those things I would say.

What are people getting wrong or is there a misconception? Are there some, some bad ideas out there around how to get sort of ranked or how to, how to get put into the answers in LLMs?

Yeah. I mean, the thing is we're still in the very early, very early stages of everything. And there's not a ton of really verifiable. I mean, and I mean verifiable by the LLM platforms themselves of like, this is what we view and incorporate into our algorithm. Like they're very protective over that. They're not going to share it. And so the, there are a lot of like heuristics of like, oh, let's do FAQs. Let's do chunkable content. Let's have fresh content. Let's have credible authors on our, on our posts, all this kind of stuff. Like let's write in the Axios kind of style because that's what AI likes. Like those are kind of the, you know, memes that are going around, but I would take all of anyone's advice with a grain of salt. And say, because no one actually knows. The people that actually know are the people that work at these, at OpenAI and Anthropic and Google, and they're not gonna share it publicly. And so, and I think, you know, we kind of saw this with SEO too. It was like, you know, there was not, you know, not a really well-defined, in the early, early stages of SEO, there's not like a really, really well-defined, like, this is, you know, backlinks are the best thing and like, all this kind of stuff, but Google slowly, you know, kind of opened up the curtain and then it kind of shared that. And I think that'll happen, especially now with ChatGPT ads, actually, it's like they need to show you some sort of volume otherwise, like, why would an advertiser pay you? And so I think we're going to see the same kind of evolution where it's like pretty protective now and we'll start to get more sense of like, oh, these are actually the things that, that work. And then also like what works now is not going to work in a year. So like the algorithm has changed, you know, the, the, um, uh, the, the, the platforms will adjust to like what's best for the user. And so what's working now, not, isn't necessarily what's gonna be working in a year from now. And so yeah, I would say instead of listening to quote unquote experts, it's like, it's been around for like 2 years now, max. Like who, there's not really an expert in the space, I would say. But instead of listening to me or others, you know, go out there, create controlled experiments, see if those experiments yield the right kind of results that you're looking for, and, you know, learn and do it on your own versus like listening to what other people say and just taking their word for it.

Yeah.

And that's what Scrunch is for too. It's like, you know, once you do that, you can like use our platform or whatever to like see if you did get more citations, you did get more mentions, all this kind of stuff. So I like being in the like kind of neutral third party space where it's like, hey, we're just here to help you out, help you measure, help you understand what's going on. We're not going to say like, do this. We're not going to authoritatively say like, do this and you're going to get this exact result. It's like, we're here to help you like learn, measure, and like find out on your own.

Yeah, no, that makes sense. In some of my little endeavors trying to figure this out, Reddit keeps coming up as a huge—

I have a take on Reddit, which I don't know if Yeah, I feel like, okay, with Reddit, with YouTube, with Wikipedia, all these citations of like, oh, this is where you need to have your brand mentions in order to show up on LLMs. That is probably the case for, you know, a lot of like the volume is going to the, to like LLMs are using those credible sources as data points for grabbing answers and retrieving them, turning them, returning them back to the users. But I would say that if you think about the prompts that people are putting in there, a lot of the times they're like across the whole entire buyer's journey. And those types of like referential, tell me about like what this thing even is types of prompts are where those citations, uh, for Reddit and Wikipedia or whatever show up. And so you're getting a lot of top of funnel, not super high commercial intent citations or, or, um, mentions from those sources, but the, it's arguable that the mentions or the prompts that you wanna show up that you care about are actually covered in more like niche publications. And so we find that if you have like lower funnel stage, like comparison evaluation type of stuff, like these are oftentimes like written articles that are a super, super long tail of content where the LLMs are referencing versus like something like a Reddit or Wikipedia. And again, I think the, don't take my word for it. I think the thing is like, go out there, identify the prompts or the set of prompts that you want your brand to show up for, you know, start measuring them and then understand like what citations the LLMs are using as a reference point to pull back their answers from. And then use that as your guiding light versus just taking it, taking everyone's word for it that like Reddit is the channel that you need to be on. You know, maybe it is Reddit, maybe it's something else. You just need to like get the data to find out, you know.

And then at Scrunch, who do you typically interface with? I mean, is it, uh, VP marketing, is it the SEO person, a GTM engineer, content engineer?

Yeah, I would say all of the above. Sometimes it's oftentimes in the marketing department where we've started, which is like the thing that the marketing department is, you know, recently are tasked with of the moment is like the CEO, the board, someone's like, hey, all of our users are going to this platform to get answers and we're not showing up in those answers. Like, what are we doing about it? And so the CMO or the head of marketing, whatever, is like, hey, we need to have a strategy or visibility into this stuff. They might kick it down to an SEO manager or a growth manager or someone who's going to like actually own that. And that tends to be our champion. Sometimes it's on the content and PR side of things even, but oftentimes I would say it's more on like the growth SEO side of things. And it's probably indicative of like the size of company. So like a large enterprise is probably going to have a specific SEO person who is now tasked with like doing this. AI search type of stuff too, but maybe like a smaller, you know, 500-person startup is going to have a growth person. This growth person owns all that. And so that's the person who we interface with. So it's really just like the person who is, you know, tasked with like, hey, we need to figure this out and we need to have ownership in reporting and strategy around it, which tends to be growth, SEO, and sometimes content, I would say. Now, another interesting persona that is coming up is on the engineering side of things. So it's like the CTO type level. Um, and that is because AI bots are now crawling your sites and the person who owns the website is often on the engineering side of things, especially with the case if the, your website is your product, like an e-commerce product. And so the CIO or CTO level person is, is now faced with, oh, we're getting this different type of traffic. We are used to like blocking or ignoring that kind of traffic. We need to have some sort of like data governance. Around what these types of bots can crawl, how they interact with our product, all this kind of stuff. And so they are now sometimes tasked with like, okay, we need to figure out how these AI bots are crawling our site. How do we make it so it's safe, secure for us to like open up different types of information to that? But then also, how can we optimize that experience so that we have a better outcome for our business? And so it's kind of, um, interesting time where that is the case where people are noticing that like AI bots are a super important new visitor to your site because those AI bots are doing work on behalf of a human that's prompting it.

Is there an emerging standard around this in terms of like a, like a parallel to like robots.txt?

Yeah. Um, so a lot of people have published articles around, you know, lm.txt and actually like those. At least right now, those, those lm.txt files don't do anything other than they're just like a regular webpage on your site. So like, yeah, the LM is going to crawl it, but they're not going to like take it as direction like they would a robots.txt. So maybe it's a good hedge or something to include for the future, but I don't think it's having much noticeable impact than just like any other page would. I think there's a lot of best practices around like, okay, an AI retrieval agent, retrieval is like, um, someone's searching, it's like a RAG, they're going to, Go search your website and a bunch of other different websites, and they're going to pull back information to the user as an answer. Those retrieval bots are, you know, they want an experience that is essentially like no JavaScript, no heavy media, not a lot of code. They want to just breeze through all your content super fast, get all the information that they need and return them back to the user. So what that, what they, the language that they prefer is Markdown and JSON right now. And so if you can optimize your sites or include even like a, you know, separate version of your site that is serving Markdown and JSON to those AI bots that are crawling it, then, you know, you have a, you know, it's all around like a token economy. So like the more tokens you have on your site, the more JavaScript, whatever, the harder the, the, the harder it is for the AI to crawl it and then retrieve stuff. And so the lighter token, um, fewer tokens, uh, required to crawl your site, the better typically. And so, you know, if you, if you take a heavy JavaScript page and turn it into like a Markdown page, it's going to have an order of magnitude plus lighter amount of tokens. And so arguably the AI is going to crawl more pages and it's going to get more information and it's going to return, be more likely for your site to be shown in the answer. So that's kind of like the, you know, the optimization that we are helping companies out with Scrunch. And it's kind of like a fascinating new world that people haven't really thought about. It's like, oh, I've created my site that's optimized for humans and now I maybe need to create a site that's optimized for AI. And like, no one's really done that or done that in the right way so far. So it's a very emerging market, I would say.

Super interesting.

Awesome. And just like, again, for the non-technical listeners or the least, less technical ones, can you give us like a more of a definition of these AI bots, what they were doing in the pre- LLM era, why they were blocked, and why they're becoming important, which kind of like tasks are they actually doing that is worth for us, you know, like rethinking our website architecture in order to welcome them instead of blocking them?

It's a good question. So there's, we group those bots in 3 different categories. And so there's one that is like the traditional Google search index. That would be, you know, your typical, like, you know, it's going to crawl your whole site. It's going to index it. It's going to show up in the search results. Like that is still a bot that's crawling your site and it's still important. That's what's going to read robots.txt and like going to, you know, do all the traditional like SEO type stuff. There is then the next evolution of that became the training bots, which are what the models are trained off of. And so these are kind of like the bots that'll crawl the whole entire internet and then use that as, um, it's a large language model. It's like they're looking for language. So they're crawling all this language and they are using that to build and train their, their actual model, like the model that the, you know, the LLM is using as like a foundational model. Then like the most interesting ones to me are the retrieval bots. And so retrieval bot is, it's like retrieval augmented generation type of bot, which is I'm going to search for a query. The training data, you know, is maybe old or stale or something. So I, as an AI bot or LLM, I'm going to go fetch the most recent data. That's out there. I'm going to run a few different queries and fan out queries from this prompt to get as much information as I think would be accurate based on the intent of what I'm inferring from this prompt. I'm going to go to different websites. I'm going to go to different sources that are public on the internet. I'm going to pull all that data in and I'm going to serve it back up to the user and I'm going to do so in, you know, milliseconds worth of time. Like those are the types of bots that, you know, are going and you're seeing like citations from in the LLMs. They're like like finding information and returning it back to the user. They do combine, the answer is combined with that and the training data, but the retrieval bots are the bots where you want to optimize your site for. And also are the bots where you're going to see, um, crawling different pages on your site and like, and understand like what they're trying, what information they're trying to gather and retrieve and return back to the user. So they're really important to pay attention to right now. In this new world of like showing up in LLM models. Very cool.

I think we'll need another episode just on that, Kevin.

We have analytics at Crunch. I'm not the plug Crunch too much, but we have analytics to like track all these bots on your sites. It's really cool to see for us. I mean, we're an AI company, so we're pretty like outside of the norm, but it's, we have more bot traffic now than we have, than we have human visits to our site. And so, you know, the switch is starting to happen. I'm sure if you look at your own analytics or whoever's out there listening to this, if you look at your own analytics, you've probably seen something similar where there's like a lot of AI traffic. You know, we have like a real-time feed and, you know, every time I pull up the real-time feed in a demo, it's like, oh, in the last 30 seconds, you know, we had like, you know, dozens and dozens of hits from ChatGPT, Perplexity, Claude, Facebook, like Meta has a bunch of, um, it's crawling us too for whatever reason. So yeah, it's kind of fascinating to see how all these bots are consuming your site and It's something that as marketers you would typically dismiss in the past where you're like, oh, bot traffic, like this is not going to give me any information on like the user experience. But now it's like, oh, I want that bot traffic because there's a person with intent behind that bot. And so it's just kind of a different like shift in thinking that, that, um, you know, marketers and technical website folk have to understand and adapt to.

And to be clear, like how do you actually track whether that's a bot or is a person? Like you can do that on GA?

This gets a little bit outside of my like domain expertise, but there's a, the, the visitor has like a, um, in the, in the visit, there's like the, the head of the, the visit or whatever, like the, the tag that's coming through, like we'll tell you if it's like a retrieval bot. We have a whole article on it, but it's essentially, it's like they, they self-identify as bot traffic and there's like unique identifiers, um, that you can capture. We use. Like whether you're using Cloudflare or Akamai or something like that, like if they, if you have like a CDN integration or Vercel or something like that, it's pretty easy to just like, we'll just do it all automatically for you. But it, you know, there's unique identifiers where you can just do this on your own.

So they're being pretty transparent in the user agents. They're not trying to—

Oh yeah, yeah, yeah, yeah. They're not trying to fake that they're not a bot or anything like that. I mean, they do show up in traditional GA4 and probably like PostHog too. So like, Yeah, yeah, yeah. You can identify it there too.

Zooming out, this is even more fascinating because I think like we're basically witnessing a shift, not just in AI search, but you know, like we're basically optimizing for building for machines versus humans, right? You know, again, I think like we are at the very beginning of this, but you know, like we're even the expectations of our industry, you know, like up and to the right. More content, more pipeline and all of that. Like, it's kind of like we're just setting an expectation for every human to perform in a straight up growth trajectory and which it resembles more of how a machine thinks, right? You know, like, and rather than a human that is more of like a cycle-based type of human, we need, you know, like we're We have growth peaks, we have rest moments, you know, like we go more in cycles, but I think like especially tech and especially the AI, you know, like industry is going more into, you know, like it's expecting everyone to behave by machine, like as machines, right? You know, like you always need to be doing LinkedIn, you need to be pulled, you need to be doing everything, all channels at all times, more pipeline and so on and so forth. So I'm seeing this more and more and I'd love to hear what you guys think about this.

Yeah, I mean, I think of, you know, when machines are talking to machines to do a job, you know, I think of like even the next evolution even of where things are going where it's like, oh, maybe, you know, your website is instead of just being optimized fully for, you know, with Markdown and all this kind of stuff, you just have it connected to your whole entire like internal documentation and all this kind of stuff that, you know, you safeguard, but you say like, oh, this is okay for AI to crawl. It's going to give you like pricing information. It's going to give you competitive intel. It's going to give you all this kind of stuff. And, um, and then maybe like the interface of, of AI talking to AI is like not even Markdown. It's not even human language. It's like some other crazy language that the AI just, you know, prefers to like, uh, talk to itself. And so it's like you create like a, qualified or whatever kind of like chatbot on your sites is like speaking AI for AI. So I don't know, this is like getting a little bit like overly conceptual, but like I think of like, you know, things could devolve into that world and maybe we just like sit back and like give it a job and like it'll just all go and build and figure stuff out on its own. But until then, you know, we still, we have to use what we have, I guess. There's still humans in the mix right now, at least. For sure.

And if we switch gears and talk about a little bit of like teams and hiring, right? You know, like we definitely like with AI, yes, job market is also having, you know, like a crisis or undergoing a crisis. How have you structured or restructured also on the back of your time at Commonroom, you know, like your GTM team or org at Crunch and where does kind of like you see GTM engineering sit?

Yeah, I mean, we're still like very early, I would say in, I mean, we've only been had a, I think our team went from 4 people in April to like 50 plus now. So it's like fast growth, but like not a huge team. The way I'm thinking about it is kind of similar to, you know, what we were talking about before of, instead of a, a 1, 1 SDR to like 2 or 3 reps ratio, it's probably gonna be more like a, a 1 GTM engineer growth-oriented type of person who's, you know, gonna do whatever it takes to get a meeting and use the tools at their disposal mapped to like, you know, 10, 10 or 5 enterprise reps. So that's like the, one of the bigger impacts I do. We have someone who's doing like vibe coding and all that kind of stuff. Maybe I'll carve out like Elena did like a, a full-time role for doing that. I mean, I would love to have that as a, as a, as a hire. But still like the basics, you still need the basic infrastructure of marketing. I think you still need, you still need content. AI content is not great enough yet. And I feel like, you know, if you're creating and publishing content from AI-generated content, the models know that like that is going to like devolve into revert to the mean, I guess. It's gonna be like the snake eating its tail. And so they're not gonna reward that content. So I think, you know, I, to me, all of marketing, most of marketing is content writing. Ads, emails, blog posts, you know, everything you do in marketing is writing. And so like having someone with a strong perspective and copywriter and content writer, and that's like the language that AI goes in, you know, that's what AI goes and crawls and pulls back to the user right now. So I feel like, you know, content is like a very key thing along with product marketing to support the sales team. So I think a lot of the traditional stuff in marketing still applies. Like I'm still hiring for those pillar-based roles. The thing that I think evolves next though is like maybe that team doesn't have to hire as much of a team for themselves. Like they have these AI agents or other people who are, you know, using AI to, you know, not have to hire like an army of people within like the product marketing org or the content org. So essentially like the way I'm building the team is, you know, hire people with really subject matter expertise, like know their domain really well, have a lot of like great IC output. And then they can use AI tools to leverage themselves and like maybe they won't hire like a bigger, a huge team. So I don't, I don't see our marketing team exploding to like 50 people in the next year. It's probably going to be like the core set of like, you know, the core pillars and then like a few people like maybe that are supporting those pillars or contractors or whatever, or AI agents for those pillars. Love it.

Yeah. I mean, um, shall we like, I think we're coming to a close of the episode.

Yep. I gotta, I gotta run to my next meeting actually.

All right. Well, we'll wrap this up. This is really awesome. Very informative, Kevin. I think Kevin's core point throughout this conversation was that great go-to-market engineering, it's not just about fancy tools and it's more about creatively stacking the right signals, triggers so you can spot real intent. You can show up with something that's relevant, a really strong offer. And I think you also said, Kevin, that the best teams, they really start with a few sharp, curious people who can do a whole bunch of, you know, unscalable experiments and then bake those winnings into something that is into systems that can scale. And then as far as the emerging edge of AI search, companies are designing content and offers and their whole websites for these for these, for the new way that we're finding information. And, uh, and that's a space that we all have to learn about. So thank you for sharing all your, your intel here, Kevin. This is awesome.

Yeah, 100%. Thanks for having me, guys. Uh, it's been great.