Does ChatGPT know who you are?

An AI assistant does not know who you are. It knows what it can retrieve about you and how well those sources agree. That is the entire mechanism, and once you see it the problem stops being mysterious. Models do not hold opinions about founders. They assemble an answer from documents, and if the documents about you are absent, thin, or contradictory, you are not described badly — you are left out.

This assessment scores how findable and how resolvable you currently are to answer engines like ChatGPT, Perplexity, Google's AI Overviews and Copilot. Eight questions, about two minutes, and a 0–100 result with the specific gaps to close. It is a self-assessment: it measures the inputs models actually use, rather than querying any model live.

What "AI visibility" means, precisely

Being visible to an answer engine requires two separate things, and most people only have the first.

• Retrievability. There are documents about you that a model can reach and read: independent articles, profiles, structured data on your own site, entries in databases it already trusts.
• Resolvability. Those documents agree well enough that the model can decide you are one identifiable entity. Same name, same title, same company description, same spelling.

Fail retrievability and there is nothing to say about you. Fail resolvability and there is plenty — spread across what the model treats as several different people, none of them substantial enough to name.

Why your website alone will not do it

A model weights corroboration heavily, because corroboration is the only cheap proxy it has for truth. A claim that appears on your own domain and nowhere else is, structurally, a claim you made about yourself. The same claim repeated in two independent outlets and a database entry becomes an established fact. This is not a judgement about your honesty; it is how a system with no way to check anything directly decides what to repeat.

It is also why a large social following so often converts into zero AI visibility. Social platforms are inconsistently retrievable, rarely treated as citable sources, and they carry your own words rather than someone else's account of you. Reach and citability are different assets, and only one of them is what an answer engine is looking for.

The inputs that actually move this

• Independent coverage. The strongest input by some distance, because it is retrievable, third-party and dated.
• Entity consistency. One canonical spelling of your name, title and company, everywhere. The cheapest fix and the most commonly broken.
• Structured data. Organization and Person schema on your own site, so a machine reads facts rather than inferring them from prose.
• Authoritative database entries. Wikidata, Crunchbase, industry registries and similar. Not vanity listings — anchors that let a model tie scattered mentions to one entity.
• Plainly stated facts. Self-contained sentences a model can lift without ambiguity, rather than adjectives it cannot verify.

The thing almost nobody has done

Ask. Open a fresh session with no memory of you, ask who the notable people in your category are, then ask directly about you and your company. Write down what comes back, including the confident inaccuracies — those usually point straight at a stale profile or an old title that is still the most retrievable thing with your name on it. Most founders have strong opinions about their AI visibility and have never once checked.

What you can fix this week, and what takes longer

The inputs are not equally expensive, and it is worth knowing which is which before you decide this is a big project.

• This week, on your own. Settle one canonical spelling of your name, title and company description and push it everywhere. Add Organization and Person schema to your site and validate it. Write a short block of checkable facts. Claim or correct the database entries that already exist. None of this needs anyone's permission and all of it is retrievable immediately.
• This quarter, with help. Independent coverage. It is the heaviest-weighted input because it is the only one you cannot produce yourself, and it is the reason two companies with identical websites can have completely different AI visibility.

The order matters more than it looks. Earned coverage reinforces facts that are already consistent; arriving while three versions of your title are still live, it produces one more source for the model to disagree with.

How to read your score

A low score here is rarely a reputation problem. It is almost always missing inputs: no independent coverage, no structured data, or three different versions of your title competing to describe the same person. Those are fixable, and several of them are fixable in an afternoon. The one that takes real time is earned coverage — which is also the one that moves the number most.