I watched an AI sell me to a client. And where it lied
A prospect vetted me through a neural net. The AI invented one of my numbers and half-talked the client out of me. I break it down on myself - and why it's about you.
A prospect recently messaged me - a large supplier expanding its geography. And he sent the chat with a neural net: before reaching out, he asked the AI what kind of marketer I am, what my downsides are, and whether I can be trusted. I read the whole dialogue. It turned out to be a better audit of my own GEO than any report - because I saw myself through the eyes of the machine that stands between me and the client.
If you think this is about me - no. It’s about you. AI already describes your clients the same way; you just don’t see it.
What the AI knew about me
The good part first. The neural net knows me in detail and pulls data from my own site: the path from team lead to head of marketing, ten years in cleaning, the niches, cleaning and retail cases with numbers, publications in three languages. It described my reputation as clean - “no direct negative reviews.”
That’s GEO in action: I spent months filling the site with facts, and now the machine retells me from them. Whoever the AI sees well in the sources exists for it. Whoever it doesn’t see isn’t there.
But then it got interesting.
And then it invented a number
The neural net told the client that in residential construction I grew a company’s revenue “1.8x in 18 months.”
I never delivered that result. The real number, which sits on my site, is 1.5x in 12 months. Where did “1.8” and “18” come from? The machine glued them from neighboring figures in my own texts: my Yandex Direct CPL is “1,800-2,200 rubles,” and timeframes elsewhere are “12-18 months.” The AI added up the scraps and produced a false result that sounds confident.
I write about exactly this all the time: a neural net doesn’t lie out of malice, it builds a picture from what sits poorly. And here it happened to me - in front of a client. Lucky the number came out more modest than reality. It could have been the reverse: it would have credited me with something I never did, and I’d be the last to know.
How the AI nearly talked the client out of me
Then the client asked directly: is a specialist like this a fit for a large supplier entering new regions? And the neural net started talking me down. Politely, but talking me down: “a retail solo marketer may not handle the scale,” “make sure there’s B2B experience, not just services,” “one person won’t manage, you’ll need a team.”
Formally - you can’t argue. But here’s the catch: the machine judges me by what it sees. And it sees cleaning, local and retail - that is, “small and niche.” For the query “large B2B supplier, regional expansion” it has, supposedly, nothing strong to cite about me, so by default it slaps on the “won’t handle it” label. Even though the case exists: wholesale with coverage across 5 regions (lumber, CPL 405 rubles). It’s just not visible in my AI footprint - that same Gemini didn’t even mention it. The experience exists, but it’s not in the AI’s answer.
And separately, about “you’ll need a team, a solo can’t cope” - that’s just wrong. I’m a manager: before marketing I ran shifts of up to 100 people in cleaning, then headed a marketing department. Assembling and running a team of doers and contractors across regions is exactly what a head of marketing does, not something I “can’t do.” It’s just that my public footprint is dominated by the “I do it all myself” story - my own post about being a solo head of marketing - and the AI reads me as a solo executor, not a manager. The machine isn’t lying maliciously: it just takes what there’s more of in my footprint. I showed the management side little - so it isn’t visible. That’s the gap you close not with excuses, but with content.
It’s not about being right. It’s about the footprint. AI recommends the legible, not the best. And if your footprint skews one way, you get downranked for other queries automatically - even if you genuinely know how to do those tasks.
What I do about it - and what you should
Three things, exactly the ones I advise clients to do.
-
Give the machine a clean fact. I rewrote the construction case so the result reads as one unambiguous sentence: “as head of marketing, revenue 1.5x in 12 months,” and separately spelled out that 1,800-2,200 is CPL, not revenue. So there’s nothing for the AI to glue.
-
Close the topic gap. The relevant case existed (wholesale, 5 regions), but there was no page for the query “B2B supplier + regional expansion” itself - so the AI had nothing to take as a targeted chunk. I built that page, and it leans on this case. No page for the query, no you in the answer.
-
Check what AI says about you. Regularly, not once. Ask ChatGPT, Perplexity, Copilot and your local search AI about your company - and read who you look like from the machine’s side. It often sobers you up more than a traffic report.
Bottom line
The neural net is now your first meeting with a client. It happens without you: you’re not in the room, you don’t hear the questions, and you can’t correct it when the machine misquotes your number or calls you “too small.” You can influence only one thing - what sits in the sources it assembles you from.
That’s GEO. Not a trendy channel, but hygiene: watch what the machine tells clients about you, and feed it the right facts. I broke this down across three pieces - why a search engine is now liable for what its AI invents, how the neural net started shortlisting contractors for the client, and how to get into its answers, by the numbers.
Have you checked what AI answers when someone asks about your business? If you want, I’ll check for you: message me on Telegram @dimik90 or order an audit.