Ask it twice

Before you read any further, do this. It takes about a minute. (There is a builder further down this page that will write the question for you and do the counting.)

Open ChatGPT, or just Google and read the AI answer at the top, and ask:

Who is the best hair braider in [your city]? Give me names.

Write down every business it names. Then open a brand-new chat — not a follow-up, a new one, because a follow-up remembers what it already told you — and ask the exact same question again.

Compare the two lists.

What you probably just saw

One name held. Maybe two. The rest were replaced by businesses that weren’t there sixty seconds ago.

That isn’t a glitch, and it isn’t personal. It’s how these systems answer, and almost nobody running a service business knows it’s happening.

It has been measured

In September 2026, BrightLocal published the largest study of this I am aware of: 200,875 unbranded local prompts across 1,342 business locations, run over 60 days against ChatGPT, Google AI Mode and Google AI Overviews, producing 1.9 million citations.

One of its findings is the reason I am writing this. When the identical prompt was run again, this is how much of the result set stayed the same:

Twenty-one to thirty-three percent. Two thirds to four fifths of the names change between one ask and the next.

Why it re-rolls

There is no ranked list sitting in a drawer somewhere with your salon at number four. Each answer gets assembled when somebody asks, out of whatever sources the system reaches in that moment. Proximity to the person asking plays a part — the study found a share of picks falling within five kilometres of the searcher. So does which pages it happens to pull.

Which means there is no position to hold. Being named once doesn’t bank you anything. The next customer who asks gets a fresh roll.

And the list got shorter

Here is the part that costs money. The same study looked at how often a tracked business got mentioned at all:

Google Maps mentioned a given business in 66% of searches. The AI surfaces: 32% to 38%.

And when AI does answer, it names very few — an average of 2.5 businesses per answer on AI Overviews, 3.5 on AI Mode, 4.1 on ChatGPT.

Think about what that does to a customer’s behaviour. A Maps result is a list she scrolls, compares, reads reviews on. An AI answer is three or four names in a sentence, and most people take the first one. The same search that used to put twenty braiders in front of her now puts three — and they are a different three every time.

There is a smaller finding worth knowing if you work out of your home: businesses with a storefront were mentioned 33.6% of the time against 29.4% for service-area businesses. It is not a huge gap. The study also notes that a business with a fixed public address is more likely to be recommended. Correlation, not proof — but if you have a real address you are comfortable publishing, publish it.

What it was actually reading

This is the finding that changed how I think about my own work.

The study traced all 1.9 million citations back to where they came from. Of the 116,000 unique domains it pulled from, 108,000 were business websites — 93% of them. Business websites accounted for 42% of every citation made. Google Business Profile accounted for another 28.5%.

So when AI answers “who’s the best braider near me”, it is largely reading business websites and Google Business Profiles, and quoting them back.

Now hold that next to how most braiders, barbers and nail techs actually operate: an Instagram profile, a DM inbox, maybe a link in the bio. All of it inside somebody else’s app.

There is almost nothing there for it to read. And nothing of yours to cite.

The part I won’t sell you

I need to be straight about what that study does and does not say, because this is where the industry is about to start lying to you.

It measured what AI cited. It did not test what makes a business get picked. It ran no model of cause and effect. Its own recommendations — get more reviews, fix your listings, build a clear website — are presented as advice, not as tested findings.

So nobody can make AI recommend you. Not me, not an agency, not whoever starts selling “AI optimisation” packages this quarter. Anyone who hands you a formula is guessing, and charging you for the guess.

What you can do is exist in a form it can actually read. That’s it. That is the whole honest claim.

What that actually means, concretely

Seven things. None of them are a ranking trick; all of them are “is there something here to find”.

1. A website on a domain you own. Not a profile, not a link page. An address with your name on it.
2. Your services written out as text. Each one named on the page, in words — not inside an image, not in a flyer graphic, not only in a story highlight.
3. A fixed, public address or a clearly stated service area.
4. A claimed Google Business Profile. It was behind 28.5% of every citation in the study.
5. Booking that happens on your own site. Not “DM to book”. A page a stranger can complete at 11pm without you.
6. Reviews that are publicly visible — on the profile, and quotable on your own pages.
7. Hours and contact details that match everywhere. Same hours, same phone, same spelling of your name, in every place it appears.

Why I build the way I build

I build complete business websites for service businesses — booking with a deposit, an owner’s dashboard, built for a phone first. The client owns the domain, the hosting, the source code and every account outright, and there is no monthly website subscription.

I’ll be honest about the costs, because you will hear it from me either way: owning it does not make it free. A domain renews. Hosting costs money. Card processing takes its cut. Anything using AI has usage behind it. What owning it gets you is a fixed scope, a predictable cost, and every account in your name.

And it does not make AI recommend you. It means there is something of yours out there to find.

muriellehairbraids.com is live right now — a braiding business taking real bookings and real deposits.

Run it on yourself

Method, and what I am not claiming

Every figure above is from one source: BrightLocal’s Local AI Visibility Study, published 16 September 2026, read in the report itself rather than in coverage of it. One honest wrinkle: the report gives slightly different counts in its results section (200,085 searches, 1,300 locations) and its methodology section (200,875 prompts, 1,342 locations). I have used the methodology figures throughout. Either set supports every claim here; the difference changes no percentage.

What is mine rather than the study’s. The 93% is my own arithmetic from its 108,000-of-116,000 domain counts. The comparison to “twenty names” describes a typical Maps page and is not a reported figure. “Two thirds to four fifths” is my restatement of the 21–33% band. The seven-item list is my judgement about what is worth having, informed by what the study found AI cites — it is not a finding, and the study tested no such list. The verdict wording in the builder above is mine too, and one reader’s single run is an anecdote, not data.

What this article does not claim. That anyone was shown to gain or lose business from any of this. That owning a website makes AI recommend you. That any factor has been proven to cause inclusion in an AI answer — the study explicitly did not model that. I have also left out the “times harder” multiplier that appears in some trade coverage of this study, because it is not in the report.

I have written before about whether the businesses AI names are real at all. This is the other half of it: not whether the name exists, but whether it is the same name twice.

Common questions