Ask a place-blind question, get a nobody answer

Can ChatGPT find my local business? It is the question we hear most from owners, and for a while our own tool answered it badly. Not because ChatGPT, Perplexity, Gemini or Claude were misbehaving. Because the questions we were asking them were wrong. This post is about the experiment that exposed the problem, the two numbers that came out of it, and what they mean for local business AI search in general. If you run a salon, a restaurant, a clinic or a trade, the short version is this: your AI visibility is decided on questions that name your town, so that is where to compete.

The experiment: 0 out of 20, then 53 out of 80

We measure visibility by asking engines real buyer questions and reading the answers. A question set (the list of buyer questions we put to each engine) is the core of the method; the full detail is on our methodology page.

Our early local question sets included questions like "what is the best hair salon?" No town. No neighbourhood. Place-blind.

Across 20 answers to place-blind questions, the local businesses we were measuring scored 0 real mentions. Not a handful. None.

The same businesses, measured on branded questions (questions that contain the business name), scored 53 real mentions in 80 answers.

Two caveats before anyone quotes those numbers back at us. They are our own product's numbers from our own early runs, not an industry study. And 20 answers is a small sample; AI answers vary between runs, and we would not hang a confidence interval on it. But 0 out of 20 next to 53 out of 80 is not subtle. The gap was the finding.

A real mention, for clarity, means the engine actually talked about the business. Not an echo of a name the question contained, and not a similarly named company somewhere else. We classify every answer this way; the reasoning is in three ways a brand appears in an AI answer.

Why a place-blind question is unwinnable

Ask "what is the best hair salon?" and the engine has to answer for everywhere at once. So it reaches for whatever is globally famous: celebrity stylists, glossy chains, names that turn up in international listicles. A good salon in Halifax is not losing that contest. It was never entered.

We think about visibility in two halves. Readiness asks: is the door unlocked, can the engines read your site at all? Visibility asks: is anyone walking through it, does your name actually come up in answers? A place-blind question fails the second half in a way no amount of readiness can rescue. Your door can be wide open. Nobody is walking down your street, because the question never mentioned your street.

This also explains a pattern we see in AI visibility reporting. Score a local business against place-blind questions and it looks invisible. The truthful reading is narrower: it is invisible on questions no local business anywhere can win. That is a fact about the question, not the business.

What a winnable question looks like

"Best Italian restaurant in Leeds" is winnable. It is geo-qualified: it names a place the way a buyer would say it. When a question carries a place, the engines change their sourcing. They lean on pages that know that place: directories, review sites, local press, the "best of" round-ups written for that town. A local business can be present on those pages. That is the whole game.

So we rewrote our entire local question set around geo-qualification. Where a buyer would say the neighbourhood, we use the neighbourhood. Where they would say the city, we use the city. One honest note: this makes scores harder to flatter, because a geo-qualified question has real competitors in it. We prefer it that way. A number you can improve by weakening the question is not a number worth tracking.

Notice that this is also how buyers behave. Nobody planning dinner asks "what is the best restaurant?" and books a table on another continent. They ask about Friday night in their own town. Our mistake was measuring a conversation no buyer actually has.

Three practical steps for a local business

You cannot change how the engines answer. You can change what they find when a geo-qualified question sends them looking.

1. Name your place the way a buyer would say it

Put your town or neighbourhood in the words on your pages, not just in the footer address. "Wedding photographer in York" beats "photography services for the wider region". Page titles, headings, service pages, the about page. Engines work from extractable text; if your place is not in the text, the link between you and your town exists only in your head. What the engines can and cannot read from a page is its own subject, covered in what AI search engines actually read on your site.

2. Keep name, address and category identical everywhere

Entity consistency (plain English: every site that lists your business describes it the same way) sounds bureaucratic and is quietly decisive. Same spelling of your name, same address format, same category label: your site, your Google Business Profile, every directory that carries you. Engines assemble a picture of a business from many pages. When the pieces disagree, the picture blurs, and a blurred business is easy to skip and easy to confuse with someone else.

3. Be present where engines source local answers

For geo-qualified questions, answers draw heavily on directories, review platforms and local press. The work is unglamorous. Claim and complete your directory listings. Gather reviews steadily on the platforms your buyers already use. Take any honest chance to be named in local press or a local round-up. One genuine mention on a page the engines already trust for your town is worth more than an elaborate page on a site nothing links to.

So, can ChatGPT find your local business?

It depends on the question, and the questions that matter are the ones that name your place. On those, visibility is winnable, measurable and worth working on. Weekly, if you are serious; these answers move.

See where you stand on the questions your actual buyers ask. Run a free scan at whocanfindme.com. No signup. Takes about ten seconds.