AI
How ChatGPT and Gemini pick a local business
What AI assistants read before they name a local business: the signals that decide it in 2026 and the old tricks that now do nothing at all.
Ben Harvieupdated 6 min read

They read the same public sources a careful customer would, only faster. As of April 2026 the things that decide whether an assistant names your business are your Google Business Profile, your reviews and the replies underneath them, and a website that states plainly what you do and which towns you cover. None of that is secret, and none of it can be bought.
That is the whole answer. The rest of this is the detail, including the parts where the honest advice is to do nothing yet.
What do the assistants actually read?
Each one reaches the local world slightly differently, but the sources overlap almost entirely.
Google's AI Overviews and AI Mode, the AI answers Google shows above and alongside its ordinary results, are built on the pages Google's crawler (the software that reads the web for it) already visits, plus the Maps and Business Profile data Google already holds. For a "near me" question they lean hard on the profile: primary category, service area, opening hours, review text.
ChatGPT answers from its own library of web pages, kept up to date by a crawler called OAI-SearchBot. It will also open a page there and then when a question needs it, using separate software called ChatGPT-User.
Perplexity reads the web with PerplexityBot and opens pages on request with Perplexity-User. It lists the sources it used on screen, which makes it the easiest of the three to check.
Claude does much the same with Claude-SearchBot and Claude-User.
Underneath the branding, all of them are reading five kinds of thing about you:
- Your Google Business Profile, including the text of the reviews.
- Your own website.
- Review sites and trade directories.
- Local news, community pages and social posts that mention you.
- Anything else that repeats your name, address and phone number.
The shift worth noticing is where the value has moved. Google Business Profile actions, meaning calls, direction requests and clicks through to the site, grew roughly 41% year on year into 2026, while clicks from ordinary, unpaid search results to small business websites declined over the same period (Localo's analysis of AI's impact on local search). People are getting their answer without the visit. Your profile is now the shop window. Your website is the stockroom that proves the window is honest.
Which signals decide whether you get named?
These are the ones that carry weight, where each comes from, and the thing to actually do about it.
| Signal | Where it comes from | What to do |
|---|---|---|
| Name, address, phone consistency | Your profile, your site, directories | Make all three identical, down to the punctuation |
| Review volume and rating | Google Maps, mostly | Ask one customer at one fixed moment each week |
| Review responses | Your Business Profile | Answer every review, quickly, including the bad ones |
| Clear service and area statements | Your website | One page per service, naming the towns you cover |
| Structured data (hidden labels that tell Google what your business is) | Your site's code | Labels for your business details and FAQ that match what the page shows |
| Freshness | Profile updates, hours, posts | Change something real every month |
| Other sites backing you up | Directories, press, trade bodies | A handful of accurate listings beats fifty sloppy ones |
One published example gives a sense of the weighting. A brand with high Maps ratings, 98.3% of its reviews answered and an average response time of 0.6 days was measured at a 38.4% likelihood of being recommended by Gemini for its category (2026 local SEO trends). Treat that as one measurement rather than a law of nature. The useful part is the ranking of causes: rating, then answered reviews, then speed of answering. All three are free.
Structured data deserves its own sentence. It is a block of facts hidden in the page, written for software to read, stating your name, address, hours and prices. It does not make you rank. It makes you quotable, because an assistant reading it does not have to guess what a paragraph means.
Does keyword stuffing or a pile of new reviews still work?
No, and both are worse than doing nothing.
Writing "plumber Edinburgh plumber Midlothian emergency plumber" into your page footer was a bad idea in 2015. An AI reading that page does not pick out a search phrase. It makes a judgement, and the judgement is that this page is padded. Assistants are unusually good at noticing filler because filler is the thing they were trained to produce and to recognise.
Bought reviews are a bigger risk. They breach Google's review policies, they get listings suspended, and the pattern is easy to spot from the outside: eight five-star reviews in one week from accounts with no other activity, all written in the same tone. An assistant summarising your reviews reads the text, not just the stars, and a customer who checks will read them too.
There is no hidden instruction you can put on your website that makes an assistant recommend you. Text on a public page aimed at instructing an AI is treated as untrustworthy by every serious AI system, and the ones that would fall for it are the ones nobody is using to choose a joiner.
When does none of this matter to you yet?
Here is the unwelcome part.
If nobody searches for your category, skip all of it. A contract manufacturer with eleven named buyers does not have an AI visibility problem. It has a sales problem, and no amount of structured data touches it.
If you have fewer than about ten real reviews, do only that. Every other signal in the table above is a multiplier on a number you do not have yet. Ten honest reviews from real customers will move you further than a year of content.
If you are at capacity, stop. More enquiries are a cost when you cannot service them. Put the prices up first.
If your website is slow or five years out of date, fix that before anything clever. An assistant that reads a page stating last year's prices will confidently quote last year's prices to a customer, and you will find out on the phone.
The order that works for almost every small business in Edinburgh and the Lothians is dull: profile first, reviews second, clear service pages third, structured data fourth, everything else after that. There is a step-by-step profile setup that covers the first of those in about half an hour, and a plainer look at whether SEO is worth paying for at all if you are not sure this is your problem.
How do you check what they say about you today?
Fifteen minutes, no tools, no subscription.
- Open ChatGPT, Gemini and Perplexity in turn.
- Ask each the question a customer would ask: "who is a good electrician in Musselburgh", "best MOT garage in Livingston", whatever is true for you.
- Write down which businesses get named, and more importantly which sources they point to. Perplexity lists them. Gemini links them. ChatGPT will tell you if you ask.
- Ask each one directly about your business by name and note anything that is wrong.
- Repeat it in a fortnight, because these answers are not stable and one run tells you very little.
That list of sources is your actual to-do list. If three directories you have never updated are the sources an assistant is using to describe your business, that is where the work is, not on your homepage.
Being wrongly described is the common outcome, not being absent. Old opening hours, a phone number from two premises ago, a service you stopped offering in 2023. Fixing the record costs nothing and is the single highest-return hour most small businesses can spend on this. If you want a second opinion on where your business currently stands and what is worth doing first, that is exactly what the consultation (opens in a new tab) is for, and putting the record straight is the SEO and Google Business Profile service. If you are wondering whether an assistant on your own site would help, what it should never do is the one to read first.
Talk to Ben about SEO and Google Business Profile
Getting found on Google by people near you who are ready to buy, so the site earns more visits, enquiries and calls.


