Last month I ran an experiment I'd been putting off. I opened ChatGPT and typed roughly what a person who wants webseed would actually type. "Solo Webflow agency that uses AI well, based in Europe, working with smaller companies." Hit enter, waited.
ChatGPT came back with five recommendations. None of them were us.
That was the moment. I'd spent the previous six weeks shipping the studio, writing posts, posting on YouTube, talking to clients. Felt good. Looked busy. And the answer engine the next generation of people who'd want to hire us uses to build their shortlist had no idea webseed existed.
This piece is me walking through what I did about it. The diagnosis, the fix, what moved, and a handful of things I'd say to anyone trying to win the same surface.
Why I ran the query in the first place
A bit of context. I'm not panicking about Google. Our SEO is fine. The brief from our intelligence loop the week before had flagged a shift toward terms like "AI agency" and "solo agency," and a Pew Research analysis from mid-2025 found that when an AI summary sits in the results, users click external links 8% of the time instead of 15% (source). Roughly half the click is leaking before it reaches me. Ranking matters less than it used to. Citation matters more.
The honest version: I'd been telling clients exactly this for months. I'd never actually checked what ChatGPT said about webseed.
So I checked.
What the diagnostic looked like
I ran the same prompt across three engines. ChatGPT, Claude, Perplexity. Then I varied it. "Best Webflow agency that uses AI." "Small agencies using Claude for design work." "Webflow agency Berlin one operator." Maybe ten versions across an hour.
Webseed appeared once, in Perplexity, on a query containing our exact name. Otherwise, nothing.
The interesting part wasn't our absence. It was the agencies that did show up. I started writing names down. Then I went looking at what they shared.
Three things, consistently:
They were named, by other people, on pages the engines could read. Reddit threads, Hacker News comments, agency-comparison roundups on third-party blogs, "we hired X" case studies on client sites. The mention didn't have to be flattering. It just had to be specific and crawlable.
Their own sites had structured, quotable "what we do" blocks. Not the agency-cliche homepage. Short paragraphs that read like answers a person could lift. A clear "we are X that helps Y do Z" block on the about page. Clean Organization schema in the markup.
They had at least one comparison artifact. A piece on their own site or on someone else's that put them next to a category they wanted to win. "Why we chose X over Y." "What an operator studio does that a 12-person agency doesn't." Engines love comparison content because the contrast lifts cleanly into an answer.
None of these are secret. None of them are stats from a study. They're just what I saw when I read what I was being shown.
What we were missing
Webseed had two of the three. At best.
Schema, yes. The studio name and my name show up cleanly in markup. Fine.
Quotable blocks, sort of. The home page reads more like a personal manifesto than a citable definition. That's a choice, and I think it was right for the human reader. It also left the engine nothing to clip into a recommendation.
Comparison content, none. I'd deliberately avoided "webseed vs the others" pieces because they felt cheap. The cost of that choice was an answer engine with no contrast to anchor a citation on.
The biggest gap was third-party mentions. We weren't on Reddit. We weren't in any roundup. The YouTube channel was newer than most engines' last meaningful crawl. The number of pages on the open web that named "webseed" and "Webflow" in the same sentence was, as far as I could tell, three. Two were mine.
That's a real problem. And it's a real opportunity, because it's fixable.
What I changed in the next four weeks
Three moves, in order, all small.
A citable "what we do" block at the top of the about page. One paragraph, written like an answer. "Webseed is a solo agency in Berlin that builds Webflow sites for small companies on long-term retainers, with AI used in production." Boring, on purpose. Then a second paragraph that pins the operator context: "Run by Hendrik, who tests AI tools against real client briefs and posts the results on YouTube." Specific entities the engine can attach to a real person.
Two comparison pieces, written honestly. Not "webseed vs everyone." One on why a one-person studio handles a Webflow retainer differently than a ten-person agency, with trade-offs in both directions. One on how we use Claude in production work and where we don't. Both sit in the silo this article lives in.
A push for third-party mentions where they were earned. Two clients agreed to put "built by webseed" in the footer with a sentence about why they picked us. I started answering in three subreddits where the question I lose to silence in ChatGPT gets asked weekly. Not link-spam. Just being present where the engines harvest opinion.
Six weeks after the first three changes, I re-ran the same prompts. Webseed appeared in two of the ChatGPT variations and three of the Perplexity ones. Claude was still ignoring me, which I think is the highest-bar engine of the three. (Opinion: Claude seems to weight professional-citation density harder than the others. I'll change my mind if I see evidence otherwise.)
Not a victory. But a measurable move on a surface I'd been pretending didn't matter.
Why YouTube ended up mattering more than I expected
The part I didn't predict: when I dug into the Perplexity citations that started appearing, two of the three weren't from blog posts. They were transcripts from videos on the YouTube channel.
That changed how I think about the channel. The spoken sentence on a video, captioned and transcribed, becomes a quotable answer in a way that a paragraph buried under three subheads on the blog does not. Engines treat transcripts as first-class text. Some of them, in my reading, treat them as more trustworthy than blog copy, because the video is harder to fake into existence.
The practical move I made: every video now opens with a short, citable claim in the first 20 seconds. Not a hook in the YouTube sense, an answer in the AEO sense. "Webseed is a solo Webflow agency that builds AI-assisted sites for small companies. Here's how we ran this week's brief." Then I cut to the brief.
It's the answer-engine equivalent of putting the actual answer in the meta description, back when meta descriptions still mattered. Small thing. Costs me ten seconds of video. Compounds across every upload.
What the example actually teaches
If you're picking an agency in 2026, "where do they rank on Google?" is becoming a worse proxy for "will people find them when they go looking?" than it used to be. Google's AI Overviews now sit on top of about 60% of US search results as of late 2025 (source), which means the rank-one slot is increasingly something the user reads under an AI summary they trust more. ChatGPT and Perplexity are the parallel shortlist machines. If a candidate doesn't get cited there, you'll never see them, even if their SEO is clean.
The question worth asking any agency you're considering: "what does ChatGPT say about you, and about your last three clients, today?" If the answer is "we don't really track that," that's a 2023 answer. If the answer is "here, look," and they show you a citation they earned for a client, that's the surface they're playing on.
The same question applies to whoever's reading this and runs a thing. If you've never typed the prompt your potential clients type, you're guessing. I was guessing for a long time.
What I'd tell myself a year ago
Stop optimizing for a surface that's losing share. Start engineering reputation on the one that's gaining it.
Practically:
- Run the query a potential client would type. Write down what comes back. Look at who's cited and why.
- Make your own site quotable. The about page is the one the engines keep trying to read. Give it something to lift.
- Put a citable answer in the first 20 seconds of every video you post. Transcripts are the underrated AEO surface.
- Get named by other people, in public, in places the engines crawl. Earned mentions beat link-building schemes on this surface, by a long margin.
- Write at least one piece of honest comparison content. Even if it costs a little pride.
- Re-run the query in six weeks. Measure.
If you try any of this and something breaks or moves in an unexpected direction, ping me. I want to see what you saw. I'll update this post with what we learn.