ChatGPT won't recommend me, and here's what I found out

I run a small Attio implementation business. A few weeks ago I did the thing every founder eventually does at midnight. I opened ChatGPT and asked it to recommend an expert in my own category.

It named four. None of them was me.

I want to write down what happened next, because the answer was not "your business is worse." The answer was a pipeline I could not see, a lever that turned out to be locked, and a strategy change I would not have made without getting stung first.

The query that stung

The prompt was simple. Some version of "I need to hire an Attio expert, who can you recommend?" I ran it in English and in Russian to be sure.

Both times ChatGPT gave me a clean, confident list of four names. All four are real competitors. All four are good. And the way it described them was specific, down to the tier, the rating, and the review count.

That detail was the clue. ChatGPT was not vaguely summarizing the internet. It was quoting something. The phrasing was too exact to be anything else.

Where the answer actually comes from

So I traced it. The four names, in that order, with that exact phrasing, all live on one page: the official Attio experts directory.

ChatGPT had not crawled the open web and reasoned about who is good at Attio. It had pulled from the canonical directory and repeated the top of the list. The directory ranks by tier first, then by review count. ChatGPT took the top of each tier, listed them, and stopped.

My business sits in a lower tier on that directory. Not because of reviews. I had more reviews than one of the names it recommended. I sit lower because of the tier itself, and the cutoff landed above me.

That was the first real lesson. For a recommendation query in a category with a dominant directory, the AI is not judging quality. It is reading a list. If you are below the fold of that list, you do not exist in the answer, and being better does not save you.

The lever I thought I had, and didn't

The obvious fix was to climb a tier. So I read exactly what that takes.

It turned out the next tier up is gated on bringing net-new customers to Attio. New logos that the platform attributes to you. Not implementation volume. Not happy clients. Net-new accounts that would not have chosen the tool without you.

Here is the problem. My business mostly serves teams that have already chosen Attio and need help setting it up well. I am the person you hire after the decision, not before it. The metric the tier gate measures is one my business model structurally does not move.

So the single highest-leverage action, the one any normal SEO playbook would tell me to take, was closed. Not hard. Closed. I could pour a year into it and not qualify, because I am the wrong shape of business for that particular gate.

That stung more than the original query. It is one thing to be behind. It is another to learn the obvious path forward is a wall.

The reframe that actually helped

Once I accepted the directory lever was gone, the problem got clearer, not worse.

If I cannot win the query that routes through the directory, I have to win every other query. The ones where the directory is not the source. Things like how to set up an AI-native CRM, how to migrate from HubSpot to Attio, what an Attio implementation costs, how to build an Attio AI agent. Nobody answers those by quoting a partner directory. Those answers get assembled from content, and content is something I control.

I checked where I stood on a handful of those non-directory queries. I surfaced on one out of eight. That one out of eight was the real scoreboard. Not the directory I could never climb, but the open field I had barely started playing on.

What I learned about getting cited

Digging into how AI systems actually pick sources changed what I thought the work was.

Traditional SEO gets you ranked. The newer game, generative engine optimization, gets you cited inside the answer itself. They overlap, but the second one rewards different things. It rewards content structured into short, self-contained chunks an AI can lift cleanly. It rewards specific numbers with sources over adjectives. And it rewards presence on the places AI trusts that are not your own website.

That last part was the one I had underweighted. Per Ahrefs data from December 2025, brand mentions correlate with AI visibility about three times more strongly than backlinks. Not links. Mentions. Your name showing up in a YouTube video, a Reddit thread, a third-party comparison article. The web talking about you matters more to an AI than you talking about yourself.

I went looking for my own footprint on those channels. On the brand-mention side I was close to invisible. No YouTube presence to speak of. No Reddit mentions in the communities where people ask which CRM tool to use. Not named in a single one of the 2026 comparison articles that recommend Attio partners. I had been writing on my own site and assuming that was the whole job. It was maybe a third of it.

What I'm doing about it

Nothing flashy. Mostly I stopped chasing the locked door and started building the thing I should have built first.

I am writing for the non-directory queries, the ones where the answer is assembled rather than quoted, and structuring each piece so an AI can lift a clean paragraph out of it. I am building a footprint off my own site, because a mention I do not own counts for more than a page I do. That means actually showing up: video, the threads where people ask the question, the comparison articles that list everyone but me. And I am pushing my happy clients to leave reviews where it helps, because even below the cutoff, a stronger profile gets retrieved more often in the adjacent answers.

The thing I am not doing anymore is treating the directory ranking as the game. It is one query. I lost it for a structural reason I cannot change. So I am going to win the other hundred.

The part worth remembering

The midnight version of this story is "the robot snubbed me." The useful version is different.

AI recommendations are not a verdict on whether you are good. They are a readout of a pipeline. Find the pipeline, and you learn whether the lever in front of you is one you can actually pull. Sometimes it is not, and the real move is to stop pulling and go build leverage somewhere the door still opens.

ChatGPT will probably keep recommending those four names for that one query for a while. That is fine. I am no longer optimizing for the answer I cannot win. I am optimizing for the ones I can, and writing this is one of them.