8 min read Brenden Parker
What Makes AI Assistants Name a Local Business: Real Data
We ask four AI assistants the same buyer questions every week. Here is what actually decides which businesses get named, including the weeks we were left out.
AI assistants name the businesses whose own pages say, in plain words, exactly what they do and where. That is the single clearest pattern in our data, and it is not what most people expect. Not the biggest brand, not the most reviews, not the prettiest site. The page that answers the question gets quoted.
We know because we measure it on ourselves every week and publish the result whether it is good or bad. Since 10 September 2026 we have asked four assistants, ChatGPT, Claude, Gemini and Perplexity, the same five buyer questions. Over the last four weeks that is 60 scored answers to questions a local business owner would really type. Our own business was named in two of them.
How do AI assistants decide which businesses to name?
They search the web live, read a handful of pages, and then repeat what those pages say about themselves. In the answers we scored on 12 October 2026, every single recommendation was justified by a line from the recommended business’s own website.
That is worth sitting with. We asked the consumer version of ChatGPT which agencies a Salt Lake City and Provo service business should shortlist. It named eight. For each one it gave a reason, and each reason was a quotation or paraphrase from one page on that company’s own site: what it says it serves, which cities it says it covers, which AI platforms it says it works on.
Not one of the eight was backed by an independent result, a case study or a third-party verification. The assistant said so itself, describing its own list as “a research-based shortlist, not a claim that any agency has independently proven” anything. The same held on the API answers from all four assistants.
So the mechanism is simpler and more mechanical than the mystique suggests. An assistant runs a search, opens a few pages, and summarises the clearest claim it finds. If your page does not state plainly what you do and where you do it, there is nothing to summarise, and you are not in the answer.
Which page gets read?
A page scoped to one service and one place, with the answer near the top. Six of the eight businesses named on 12 October were picked from a city or state page, not a home page.
This is the most actionable thing in our data. The pages that got their owners named were addresses like a city service page, a state service page, or a service page with the AI angle in the address. The home page won in only two of the eight cases.
That makes sense once you think about what the assistant is doing. It searched for something like “Salt Lake City SEO agency Provo”, and the page that matched was the one about Salt Lake City and Provo. A home page has to be about everything, so it is about nothing in particular. A page about one service in one place is an exact match for how buyers ask.
It also explains an uncomfortable result on our side. We retired one of our own exact-match pages during a site rebuild on 30 September 2026. On 12 October, a competitor was named as the single best answer to “best AI SEO agency in Utah” by one assistant, and the reason given was that it had a page at an address naming that exact phrase. We had the equivalent page and removed it. That is on us, and we are writing it down so we do not repeat it.
Do reviews decide whether you appear?
They decide whether you appear in the map panel next to the answer, and that bar moves. On 12 October the three businesses in ChatGPT’s panel had 27, 86 and 74 reviews.
There are two separate things going on in one answer, and mixing them up is the most common mistake we see. There is the prose, where the assistant names businesses based on what their pages say. And there is often a small local business panel, drawn from Google’s local data, where review count matters a lot.
We are absent from the panel every week, and the reason is simple: we have fewer than five reviews. No content fixes that.
But the bar is less fixed than it looks, and this is where a single week’s reading misleads. On 28 September, the top seat in Google’s own local map for “seo salt lake city” was held by a business with one review. On 5 October, the three seats had 27, 74 and 27. On 12 October, two of the three were back to one review and no reviews at all. We reported in early October that this had become review-gated. A week later that reading was too confident. It oscillates, and anyone telling you there is a fixed review threshold is guessing.
What does hold across every week we have measured: Google says local results come down to relevance, distance and prominence (Google Business Profile Help), and reviews are the part of prominence you can actually affect. Low review counts win seats sometimes. They do not win them reliably.
Does being mentioned mean being cited?
No, and the gap between the two is the most useful number in this whole exercise. Being named in the text and having your site linked as a source are different events, and they come apart constantly.
We score them separately for that reason. An assistant can describe your business accurately while citing someone else’s listicle about you. It can also link your page and not name you, which happened to us for three consecutive weeks earlier in the autumn on the same question from the same assistant.
Here are our own rates, pooled across the last four weekly runs to 12 October 2026: 60 scored buyer answers, 2 that named us, 2 that cited us. That is 3.3 per cent on both. We publish the pooled figure rather than the weekly one because the weekly one is noise. In one week we were named twice. In the three weeks after that, zero. Those are not a rise and a collapse. They are the same low rate sampled four times.
If an agency shows you a single screenshot of an assistant naming your business, you have learned almost nothing. Ask for the rate, the number of questions behind it, and the weeks where the answer was no.
What is the fastest thing that changes an answer?
Rewriting the sentence that describes your business on a page the assistants already read. In our experience that lands in about one to two weeks, which is far faster than anything else in search.
We tested this on ourselves in September 2026. We rewrote our own about page to say plainly that we are a done-for-you local SEO agency for service businesses, rather than describing a software-sounding category. Seven days later one assistant was repeating that description back, close to word for word, as its answer to what our company is. Write the sentence you want quoted, because it will be quoted.
The limit on this is crawling, not writing, and it is the part nobody warns you about. An assistant can only repeat the version of your page it has read. On 12 October, one of the four described a service we retired on 30 September as a live offering, quoted a phrase we had deleted, and gave our location as two different towns in one answer. Every error traced to an old copy. There was nothing on the live site to fix.
Our own pages were in the same state. On 12 October our main service page had not been re-read by Google since 10 September, and our free audit page not since 18 September, despite a re-crawl request on 5 October. So the honest sequence is: fix the sentence, then wait two to four weeks for the index to catch up. Anyone promising a change in AI answers this week is selling something.
Does any of this get you customers?
We do not know yet, and we are not going to pretend otherwise. We have had zero mentions in three of the last four weeks and we have had AI assistants send us a handful of visits, not customers.
In the 28 days to 12 October 2026, assistant referrals sent us six sessions, from Copilot and, for the first time, ChatGPT. Six. Over the same period our search impressions rose 34.8 per cent to 5,317 and our clicks stayed at zero, because almost all of that movement is on page two.
We think this channel matters because more buyers start there every month, and because the work that earns a mention is the same work that earns a ranking: clear pages, a complete profile, real reviews, honest claims. Google’s own guidance on what deserves to rank is a list of questions about originality and real experience (Google Search Central), and mass-produced pages built to game either system are specifically what its spam policies name (Google Search Central). The two goals point the same direction.
But we are not going to show you a chart of mentions and call it revenue. When we have leads we can trace to this channel, we will publish those instead.
What we would actually do
Three things, in this order, and none of them need an agency.
First, make one page per service per place you work, and put the plain answer in the first paragraph. Say what you do, who for, and which cities. That is the page an assistant reads, and it is also the page that ranks.
Second, ask every happy customer for a review, for the next six months. It is the only lever on the panel beside the answer, it is free, and nobody can do it for you.
Third, measure it the same way every time. Pick five questions a customer would ask, ask them of two or three assistants on the same day each week, and write down whether you were named, whether you were linked, and who was named instead. After four weeks you will have a rate. Before four weeks you have anecdotes.
If you would rather see where you stand before doing any of it, our free visibility audit checks your Google Maps position, your reviews against your closest competitors, your Business Profile, your phone site speed and one AI answer about your business, in about two minutes.
More on the same decision: can ChatGPT do SEO for your business covers what you can honestly do alone, what a free SEO audit should tell you covers how to read an audit without being sold to, and our local SEO service page lists what a month of this looks like done for you.