GEO, Measurement
When your brand name shows up in an AI answer, one of three things has happened: a real mention, an echo of your own question, or an answer about a different company with a similar name. Only the first is visibility. We classify every answer and score the other two as zero.
Ask any owner who has heard about AI search and the question comes out the same way: does ChatGPT mention my business? It sounds like a yes-or-no question. It is not. When we measure AI brand mentions across ChatGPT, Perplexity, Gemini and Claude, we have to classify every single answer, because a brand can "appear" in an AI answer in three different ways. Only one of them means the engine actually knows you. The other two are noise, and most tools count all three as success.
Here are the three kinds, a realistic example of each, and the reason we score two of them as zero.
We measure visibility by asking engines the questions your buyers actually ask, then reading the answers (our methodology explains the full process). When your brand name turns up in a reply, one of three things has happened.
The engine talks about your business. Correctly, in context, with facts attached.
Illustrative example: a buyer asks "who does small business bookkeeping in Sheffield?" and the answer includes "Harber & Co is a Sheffield firm that specialises in bookkeeping for trades businesses, and reviewers mention fast responses."
That is visibility. The engine holds real information about you and surfaced it when a buyer question made it relevant. Whether it goes on to recommend you is a separate question, and we have written about the difference between a mention and a recommendation before. But a real mention is the entry ticket. Nothing else counts until you have it.
The engine parrots your name back because your question contained it.
Illustrative example: you ask "what do people say about Harber & Co?" and get "I could not find much reliable information about Harber & Co. Could you clarify what the company does or where it is based?"
Your brand name is right there in the answer. And the engine has just told you, politely, that it has never heard of you. An echo is not a mention. It is a mirror.
The engine answers about a different company with a similar name.
Illustrative example: you ask about Harber & Co, the bookkeeping firm, and get "Harber is an open-source bug tracking tool used by software teams to manage issues and releases."
Fluent, confident, and entirely about someone else. This is worse than an echo. A buyer who reads an echo learns nothing. A buyer who reads a wrong-entity answer learns something false, and walks away believing it.
Most visibility tools check one thing: does the brand string appear in the answer text? On that test, an echo is a hit and a wrong-entity answer is a hit too.
Here is the problem. If your question set contains your brand name (and branded questions are a normal part of any serious set, including ours), the name will come back in nearly every reply by construction. Ask an engine about a company it has never heard of and it will still use the name while saying so. A tool that counts string matches will report a healthy mention rate for a business that is, in reality, invisible.
That is why we classify instead of count. A real mention scores. An echo scores zero. There is no partial credit for an engine repeating your question back to you.
Echoes are easy to spot once you know the tells: hedging language ("I could not find much"), a question returned instead of an answer, or a reply that only restates what your question already said.
Name collisions are more common than you would think, and no engine warns you when one happens.
Some numbers from one verified 100-answer run of ours. 15 of the 100 answers were about a different company entirely: a bug tracker that happened to share a similar name with the brand we were measuring. On the same run, two of the four engines answered some questions by asking what the brand name meant. Gemini had zero confusion. The other three produced all of it.
The usual caveat, because we do not dress single runs up as universal truths: AI answers vary between runs, and those figures describe that run, that brand, that day. The pattern behind them is durable, though. When an engine is unsure which thing a name refers to, it resolves to the better-documented one. And the better-documented one is usually not the small business.
To an engine, your brand is an entity: a named thing with a category and facts attached (that is all the word means). If two entities share a name and one of them has years of documentation, forum threads and press coverage, the name resolves to that one. Your buyers then get a confident answer about a company that is not you. We score wrong-entity hits as zero, and we would argue they are worth less than zero.
You cannot rename the other company. You can make it easy for engines to tell you apart. The work is entity consistency: same name, same category, same description, everywhere the engines read.
Three concrete steps:
None of this guarantees a mention. Nothing does, and you should be wary of anyone who says otherwise. What it does is remove the ambiguity that turns your mentions into someone else's.
A raw mention count tells you very little. What matters is what sits underneath it: real mentions, echoes, or another company wearing your name. Our scan asks real buyer questions across ChatGPT, Perplexity, Gemini and Claude, classifies every answer into these three kinds, and scores only the real ones. Run a free scan at whocanfindme.com. No signup. Takes about ten seconds.
Google's search ignores llms.txt. Independent data shows 97% of the files are never fetched. We still score it 5 points out of 100. Here is the working behind that number, and the three things that move AI visibility more.
Read moreChatGPT mentions us in 58% of answers and recommends us in 0%. The gap between being seen and being chosen is where AI search is won, and almost nobody measures it.
Read more