Bots Have a Preferred Content Type
Dan Kurtz of Miss Pepper AI on why the crawlers outnumbering your humans are just another customer segment — and why you already have the playbook
I've said "write for humans first" on stage, in decks, and in about a hundred client calls. I still think it's right. But Dan Kurtz said one sentence to me this week that made it feel like a half-sentence, and I haven't been able to put it back down.
Dan runs Miss Pepper AI. He is not an AI evangelist and he is not a doomer — he's an agency owner who reads his own server logs and pattern-matches against twenty years of watching SEO get gamed. That combination made this one of the more useful conversations I've had this year.
Spy vs. Spy
Worth knowing where he's coming from. In 2006 he was in college for graphic design, watched India and the Philippines come online, and went: no. He picked up Seth Godin's Linchpin, asked himself what he could jump to that made the most sense from where he was standing, and landed on marketing. Then he stumbled into where he says most of us actually got started — Black Hat World.
Rank and rent first: build the site, charge the business for the leads. Then a deliberate one-eighty into the hardest affiliate verticals he could find — gambling, and the three Ps. His description of that period is the best one I've heard: it was Spy vs. Spy. White coat, black coat, fighting each other. Learn all the black hat stuff on one side, all the white hat stuff on the other, then find the middle ground with judicious judgment about what you can use where.
"How big is this pond if I decide to go play in it?" he said. "It turns out it's pretty big. Also very small community."
There was a software he built in 2012 or 2013 that was doing well until a system-wide footprint got it hit with a Google penalty. He mothballed the project. In 2020, with nothing to do, he took it back out and rolled it into a full-service agency. That's the résumé: got punished for scale once, came back and built the thing around scale done properly.
The number underneath all of this
Somewhere in the middle of June, Cloudflare put out a finding that bot traffic on the internet had officially surpassed human traffic. Dan's reaction was the correct one — it's a big one to sit with. There are now more automations and hooks out on the internet than there are people currently browsing.
His own logs say the same thing from a different angle. Traditional Googlebot crawling is down around 60% across his sites. AI bots doing crawling and retrieval are up around 80%. He thinks the destination is obvious: we're going to be looking at something that equates to the ChatGPT screen to go find things, inside Google, pretty shortly. Same search bar you've always known. It just generates the AI page now.
Which is also, incidentally, why he shrugs at the domain extension question. Dot com, dot ai, dot agency — it depends, and it counts today for expired domains, EMDs, authority stacking, all the fun little tricks everybody has. But once the data is being ingested by the AIs rather than looked at by a person, he thinks that goes out the window.
Slop is old. Scale is new.
I asked whether AI slop is a feature of the moment or a permanent headache, and Dan's answer was refreshingly unbothered, because he's seen the shape of it before.
Keyword stuffing was slop. The famous Lorem Ipsum test — where somebody took placeholder Latin, swapped in "rhinoplasty Plano" and the LSIs, and ranked number one on it — was slop. "Slop has always been around," he said. "It's just we can create more of it at scale."
So his prediction is the boring, likely-correct one. Google has been around long enough and been gamed enough times that it will do what it always does: see everything across the board, push an algorithm update, and label it scaled content abuse. They have the most skin in the game so they'll probably go first, and the other three or four — GPT, Claude, Grok, Perplexity — will have to follow suit, including putting stop triggers on certain websites based on how they're retrieving data from them.
The oldest play in the book, pointed at a new reader
Here's where it clicked. I'd brought up something I saw on LinkedIn: Google says don't use markdown, but a full HTML page takes roughly three times longer for a bot to load than the tokens and time it takes to process markdown — and markdown is organized the way these things like. I also mentioned my interview with Mike Montague, who pointed out there's already an economy of millions of dollars being spent by AI bots with AI bots, no human directly involved. Somebody gave a directive, handed over a wallet, and the site was able to interface with it.
So I asked Dan the uncomfortable version: are we looking at a layer of optimization that runs against "write for humans first"? Not all your visitors are human anymore. Even the human ones are arriving through what I've started calling a bot sandwich.
His answer didn't treat it as a new discipline at all. As SEOs and content marketers, he said, we've been doing this for decades — we call it multi-purposing. Take the blog post, turn it into an infographic, turn it into a YouTube video, make a reel, make a short, make a press release. We do it because people learn in different ways and each of them has a content type they'll go after.
And then the line:
"Bots have a preferred content type. Technically, they're just an additional customer."
That's the whole reframe, and it costs you nothing to adopt. It's not a new philosophy. It's the same practice you've run since 2010, with one more audience added to the distribution list. His risk math is clean too: spend the time and invest in it, because the worst thing that happens is they fly right over it — and then it's just a matter of directing them to it next time and saying, hey, this file exists.
The version of this he's actually building for clients goes further. He's toying with connectors an LLM could just plug into, so the site becomes a real-time query box. If a bot comes and scrapes, can you feed it the appropriate data it's looking for instead of making it go hunting? Auto-queued, he said. A high-five baton pass on the way out, and they just have the data.
The table saw
Then he gave up a secret-squirrel one, his words. If you go to your hosting provider and reissue a new SSL certificate, it forces all of the AI bots to re-crawl your stuff all over again.
He is honest that he doesn't know the cadence or the frequency, and fully expects somebody to spam it to death and find out their website doesn't take bots anymore. His own warning is the one to quote back at yourself: "Like anything else, this is a table saw. Watch your fingers and make sure you can still count to ten."
My version: salt is good, but cover the whole plate in it and you'll get sick. SEO has always had that tendency — the money is on the table, and from the black hat side, three-quarters of the motivation is the thrill and the bragging rights anyway.
Why he won't sell you deliverables
I asked about the productized agency — four blog posts a month and some link building, call it a day. Dan has never liked that model, and he said why in seven words I've been repeating since:
"Deliverables in SEO means that you're not paying attention."
You're focused on hitting the quota. X blog posts, Y guest posts, Z of whatever — and it feels static, in a market where everything is pivoting fast and companies themselves are now able to turn faster.
What he's building instead is an AI enablement layer first, with the human working on top of it. He's pointed about the distinction, and about the companies getting it backwards by firing ten thousand people and calling it a strategy — a six-hundred-million-dollar lesson. His goal is that nobody experiences scope creep: you sign up for the thing, you do the thing, everybody's happy. He'd rather an employee spend an hour or two on the clock learning something and come back with a test idea than absorb one more "hey, can you just."
For context on who this is for: his ICP sits around the five-million-dollar mark. Smaller than that probably isn't ready to play, and larger than about twenty-five million is out of scope.
The 7 a.m. boot sequence
The part I'm stealing outright. Dan built himself an AI-powered operating system inside Claude. Every morning he boots the laptop, opens it up, and via MCP connectors it goes into all his accounts and scrapes everything — client work from the project management software, calls that came in after hours on Friday and over the weekend — and hands him a list: here's where we're at, here's the needle movers, here's what you should be focused on. Then a 3 p.m. update on all of it.
But the supercharger is what he fed it. He took the personality tests — DISC assessment, Myers-Briggs, and if you're into the ethereal, he says, hand it your star chart too — so it knows his patterns of behavior and, more importantly, knows where he falls short. So when he says he'll get to something later, it pushes back: six out of ten times you've tried this in the past, that hasn't happened, so this is going to the top of today's list.
The most common complaint about these tools is that they're built as yes-men. This is the cheapest available fix, and it isn't really coding — it's just telling it who you are.
One AI prompting another
Then the one that actually made me stop the conversation. Dan takes the SEO Gets report from the MCP, takes the Microsoft Clarity report for a given landing page or opt-in page, gives Claude access to the live page, and lets it iterate the next version based on the user data and the engagement metrics. He's using it to run his A/B testing.
I told him that's a moment, and I meant it. That's a week or two of specialty expertise, meetings, and getting a human to execute the concept. I've had a MozCon talk stuck in my head for years about a page ranked number three where the heat map showed people hitting one image and bouncing; the speaker swapped the image — same weight — and got more full-page scrolls, more conversions, and better rankings. The missing piece was always the execution loop, not the insight.
Dan closed the loop. It's wired to his CRM — GoHighLevel, which he describes as simultaneously everybody's favorite and the bane of their existence, but with a genuinely robust MCP. Combine that with Desktop Commander and the API docs, tell it where the landing page lives and how to set up an A/B test, and it duplicates the page. Claude drafts the prompt that goes to GoHighLevel's AI, which builds the variant page in two or three minutes.
"So now I've got my AI talking to other AIs that are deploying websites."
One correction worth noting, because it surprised me: he's doing all of this in Claude Cowork, not Claude Code.
Fifty dollars a week of market research
All of these systems talk to each other by design. Everything he does for SEO has a dashboard with an API. The exec thought-leadership platform has an API. The appointment-setting platform has an API. So when an ad campaign performs, it passes that to the intelligence agent inside the thought-leadership platform, which proposes a new landing page based on what Clarity and SEO Gets are showing, asks the exec to approve, and on a yes kicks off the rest of the software automatically. Sales talks to SEO talks to marketing talks to appointment setting. "It's now a little ecosystem flywheel machine."
For fuel he's using EXA plus a news collector, and the social intelligence piece is the one I'd copy tomorrow. He scrapes the top 100 accounts on each of about ten social platforms and studies what the style was, what the hook was, everything going into their system. The trick that makes it survivable: hook up Gemini and let it transcribe the video, or watch the video frame-by-frame to see how the hook lands. A hundred accounts times ten platforms gets you a pretty good moving average, refreshed weekly.
Cost: "it sounds stupid, but we're spending like probably fifty bucks a week in AI credits just doing that." As he put it, doing this years ago would have been a team of a hundred people looking at a hundred accounts and writing down what they thought.
So here's my question
Dan's rule for himself, after admitting the thing all of us do — "as SEOs we love tinkering with stuff but never getting anything off the ground" — is blunt: if you're going to start this project, you damn well better finish it. Full scope mapped, projected launch, close the loop. He expects to publish real data on what an agentic flywheel does to a company around Q1 2027, from a large corporate coaching client they're wired all the way into.
So: do you actually know what your bot-to-human traffic split looks like right now? Not your rankings — your logs. And if bots are genuinely a customer segment on your site, what have you ever shipped that was built for them?
Tell me. I read every reply.
Find Dan: LinkedIn · Miss Pepper AI · Data for LLMs · AI Optimization
Full episode: https://www.youtube.com/watch?v=XqByDTeIl_M
Recap article: https://unscriptedseo.com/dan-kurtz-bots-as-a-customer-segment-markdown-agentic-flywheel/
New here? The Unscripted SEO Podcast is 150+ unrehearsed conversations with people actually doing the work. Subscribe and you'll get each one as it lands.