Do Customers Actually Use ChatGPT to Find Local Businesses?

The honest answer, with the numbers attached

Eight percent of American local searchers started their most recent hunt for a local business inside an AI tool. Of those who did, just 2% used an AI tool and nothing else. Both numbers come from the same place, BrightLocal's July 2026 channel study of 1,227 US consumers who had searched for a local business in the previous three months.

One housekeeping note before anything else. The test at the bottom of this post is a smoke test. It is directional only. It tells you whether you showed up in one answer, on one machine, in one location, on one day. It is not a measurement, it cannot support a month over month claim, and three answers is not a sample size. Treat it the way you would treat driving past a competitor's yard to count trucks.

Here is the part nobody sells you. In that same study, 23% used an AI tool at some point during their most recent local search, 19% said they will not consider using AI at all, and 39% of the AI users went and looked at social media afterward. The behavior is a loop, not a funnel. AI answers a question, the person goes and checks the answer somewhere else, then comes back.

So when you say your customers are fifty five and they call the number on the truck, you are half right and the half you are right about is shrinking. Pew Research surveyed 5,119 US adults in February 2026 and found 63% of adults under 50 use AI chatbots, against 42% of those aged 50 to 64 and 23% of those 65 and over. Twenty three percent of retirees is not zero. It is one in four of the people calling you about a water heater.

What a real prompt actually looks like

Every AI marketing post shows you a prompt like "act as a discerning homeowner and compare the top five licensed HVAC contractors in Phoenix by warranty terms." Nobody types that.

The best public read on how people actually prompt comes from Stella Rising, whose January 2026 survey of 524 active LLM users was written up by John Morabito at Search Engine Land. Two thirds of respondents reported writing prompts of 15 words or fewer. About 60% phrased the query as a question and only 9% gave a direct command. And the pieces that matter to a local service business show up at measurable rates: 24.5% of prompts include the word best, 28% mention price or budget, 32% include a personal attribute, and 16% are explicitly location based.

That is the single most useful sentence written about AI search this year, and it should change what you test. If a prompt is eight words long, the answer is decided by a handful of retrieved pages, not by a rhetorical duel. It also means the prompts worth testing are the boring ones. Short, misspelled, no punctuation, city name attached.

One extension the sources do not make, so we will. Those four measured attributes (best, price, personal detail, location) are not four separate prompt types. They stack, and they stack in a predictable order as the person gets closer to spending money. "ac repair cost" becomes "best ac repair st george" becomes "licensed ac repair st george under 500 same day." The library below is built on that stacking order. If you run a trade where the emergency question and the price question arrive together, the vertical version of this is in AI marketing for HVAC companies.

The library: prompts sorted by distance from money

A word on where these came from, because it matters. These are not transcripts of anyone's chat logs. No public dataset of local hiring prompts exists. They are constructed to match two measured things: the shape of real prompts from the Stella Rising data above, and the deciding factors BrightLocal measured among 1,227 local searchers, where the top reasons for choosing a business were clear and complete information at 32%, the most or best reviews at 30%, and clear pricing or an offer at 29%.

Swap the bracketed parts for your trade, your city and your business name. Paste them one at a time.

Band one: research. They do not know your name yet.

These sit furthest from the money. Being named here is a bonus, not the goal. The goal is that whoever gets named is not exclusively your competitor.

Band two: shortlist. They want names.

This is the band that decides whether you are in the consideration set. BrightLocal found 72% of consumers look at three or fewer businesses before deciding. Three slots. If you are not in the answer here, you are not in the running.

Band three: hire. Your name is already on the table.

This band is the one owners never test, and it is the one closest to the money. Somebody has your name from a neighbor, a truck, or a yard sign, and they are checking you before they call. A wrong answer here does not cost you a lead, it kills a lead you already earned.

If band three comes back wrong, that is a data problem, not a marketing problem, and it is fixable. We wrote the repair sequence in what to do when ChatGPT has your business wrong.

Use this graphic on your site

Free to republish with a link back to this page. Copy the embed code:

<a href="https://timpsonmarketing.com/blog/do-customers-use-chatgpt-to-find-local-businesses"><img src="https://timpsonmarketing.com/assets/infographics/do-customers-use-chatgpt-to-find-local-businesses.svg" alt="Printable checklist of nine AI prompts grouped into research, shortlist and hire bands, each with a box for what came back and whether the business was named" width="1200" style="max-width:100%;height:auto"></a><p>Graphic by <a href="https://timpsonmarketing.com/blog/do-customers-use-chatgpt-to-find-local-businesses">Timpson Marketing</a></p>

Run the check in twenty minutes without fooling yourself

This nine prompt check is the first thing we run on every ChatGPT marketing engagement, before anyone touches a word of the website, because a wrong phone number in band three is faster and cheaper to fix than a content gap in band one, and it costs more when it is left alone.

Darren Shaw, who runs the largest annual survey of local search practitioners, puts the honesty bar where it belongs for the whole category.

That applies double to AI answers, where the retrieval step happens before you ever see the words.

If you are missing, fix things in this order

Most of what gets sold as AI optimization is decoration. We went through the full diagnosis in why AI does not recommend your business, and the short version is that the expensive parts are the ones that do not work. Google's own generative AI documentation states plainly that structured data is not required for generative AI search and that you do not need to create new machine readable files, AI text files, markup, or Markdown to appear in Google Search. Ahrefs tested that empirically, tracking 1,885 pages that added JSON-LD against 4,000 controls and finding no meaningful citation uplift on any platform. Separately, 97% of published llms.txt files got zero requests in May 2026. That is the same evidence trail we walk in is schema markup worth it and in GEO versus SEO, and it points the same direction both times.

So skip all of that. The order that actually matters:

Step four is not padding. BrightLocal's 2026 review survey of 1,002 US adults found 74% of consumers seek out reviews written in the last three months, so a wall of five year old five stars reads as a closed business to a human and offers a model nothing current to summarize.

Step six is the one owners resist and the evidence keeps pointing at it. Aleyda Solis measured 15 brands and found 84% to 93% of AI citation weight sat on third party sites rather than the brand's own domain. On the local side, Whitespark's 2026 ranking factors survey of 47 experts scored presence on expert-curated best-of lists as the single highest factor in its new AI search visibility column, with a dedicated page for each service second. You cannot write your way into an AI answer from your own website alone.

That also cuts against the crawler-blocking instinct. If you or your host blocked AI bots wholesale, you removed yourself from ChatGPT search answers and Claude search answers to prevent training you were never going to get paid for. OpenAI documents four separate bots with four different jobs, and only one of them is the visibility gate.

The reasonable read across both datasets is that AI answers compress the winner list far harder than the local pack does, which raises the value of being in the compressed set and lowers the odds for any individual business. That is an argument for working on it, and an argument against paying for it as though it were a guaranteed placement. We laid out how to tell the difference in whether AI SEO is worth paying for. It is also an argument for not abandoning the surface that still produces the volume, which is ranking higher on Google Maps.

What the model is actually reading about you

It is not reading your llms.txt. Ahrefs analyzed 1.4 million ChatGPT prompts and found the general web search index accounted for 88.46% of cited retrievals, with news at 12.01%, Reddit at 1.93% and YouTube at 0.51%. The same study found URLs with natural language slugs were cited at 89.78% versus 81.11% for opaque ones, and the median age of a cited search result was around 500 days. Old pages get cited. Unreadable URLs get skipped.

The local picture is changing underneath that. On July 23, 2026, Search Engine Land reported that ChatGPT gained access to Yelp's reviews, ratings, photos and business information, including Yelp's Request a Quote flow so users can contact a provider from inside the answer. Yelp CEO Jeremy Stoppelman told Axios that Yelp sees value in distributing its content beyond its own properties when it benefits consumers.

Read that as an operator, not a marketer. A licensing deal means the profile you ignored on a review site you do not like is now feedstock for the answer a customer sees. Your Yelp page is no longer a page. It is a data source. The same logic applies to every directory you abandoned in 2019, which is the same argument we make about reviews across more than one platform.

What this is worth in dollars, honestly

Small, for now, as a traffic channel. Ahrefs measured traffic share across 76,000 sites and estimated Google sends 190 times more referral traffic than ChatGPT, at 39.98% versus 0.21%. Semrush's channel mix study across more than 50,000 sites put AI systems at 0.14% of total website traffic in 2025 against 16.04% for organic search, though AI grew 66% in the year while organic grew 2.38%.

So do not move your budget. Fund this the way we fund GEO marketing generally, as a slice alongside local SEO rather than a replacement for it. Do the work anyway, for a reason that has nothing to do with clicks: the calls are leaking from the other end at the same time.

Put the two together and the picture is uncomfortable but clear. Your Google Business Profile is producing fewer calls at the same ranking, and the surface absorbing that attention sends almost no clickable traffic. The person who read an AI answer about you does not arrive with a referrer. They arrive by typing your name into Google, or by dialing. Which is why the honest KPI here is not sessions from ChatGPT. It is whether you get named, and whether what gets said about you is true. If you want the measurement argument in full, we made it in how to tell if your SEO agency is working.

What we would actually do on Monday

Run band three first. Ten prompts, ten minutes, your own name. It is the cheapest audit in this entire category and it finds the errors that cost you jobs you already won.

Then run band two and write down who did get named. Those are your real competitors in the surface that is growing, and they are frequently not the same firms you fight in the local pack. If a name surprises you, go read what the internet says about them that it does not say about you. Usually it is a best-of list, an association page, a news mention, or a service page you never wrote.

Then leave it alone for sixty days. Ten prompts a week is not a program, it is a nervous habit, and the noise in these answers will happily convince you of a trend that is not there.

The rest of the plain-English series on this, including the local pack, reviews and paid, sits on the blog.