Nobody outside the AI companies knows exactly how ChatGPT or Gemini choose which businesses to mention, and anyone who claims otherwise is guessing. But after testing hundreds of local questions across Essex towns, we have a reasonably clear picture of what tends to matter. Here is what we have seen, and where we are still unsure.
Two ways an assistant finds a business
There are really two routes. The first is what the model absorbed during training: a huge snapshot of the web that is months or years old. If your business was well documented then, it may be known. If you launched last year, you may be invisible on this route.
The second is live retrieval. Many assistants now search the web as they answer, pull in a handful of pages and summarise them. Gemini leans on Google's own index and Maps data. ChatGPT, when browsing, uses a search provider. In both cases the pages that get pulled in are the ones that clearly answer the question asked.
Signals that seem to carry weight
From our testing, a few things come up again and again for businesses that get named.
- Clear service and location statements. A page saying you are a family-run electrician serving Southend, Rayleigh and Leigh-on-Sea is easy to quote.
- Consistent details. Name, address and phone number matching across the website, Google Business Profile and directories.
- Reviews with substance. Reviews that mention the actual job and the town give a model something to work with.
- Third-party mentions. Local press, trade bodies, chamber of commerce pages and reputable directories.
None of that is exotic. It is the same groundwork good local SEO has always needed, which is partly why we are sceptical of anyone selling a secret AI trick.
Why Gemini and ChatGPT often disagree
Ask both the same question and you will frequently get different businesses. Gemini is tied closely to Google Maps and the Business Profile, so a well-maintained profile does a lot of work for it. ChatGPT has no Maps feed of its own in the same way, so it relies more on what it can read on websites, directories and articles.
That is a real practical point. If you have only looked after your Google profile, you may do well in Gemini and poorly elsewhere. If your website is thin, the reverse can happen.
What a model cannot see
Assistants struggle with information locked inside images, PDFs, pop-ups and scripts. We regularly find Colchester and Chelmsford firms whose best content, such as price guides or service areas, sits in a graphic that a machine cannot read. If the text is not on the page, it effectively does not exist.
They also struggle with contradictions. If your website says you cover Harlow but an old directory says you are in Brentwood only, the model may pick either, or neither.
What we would do about it
First, write plain pages. One page per core service, one page per main area, each stating what you do, who it is for and where you work, in normal sentences.
Second, tidy your footprint. Search for your business name and fix every listing that is wrong or out of date. It is dull work, but it is the part most firms skip.
Third, add structured data so the facts on your pages are labelled for machines. And fourth, ask customers for specific reviews, not just stars, without scripting what they say.
If you would like a second pair of eyes on this, our Essex SEO service includes this kind of audit.
What we would not bother with
Stuffing pages with phrases like best plumber near me does nothing for an assistant and reads badly to people. The same goes for mass-produced pages for every village in the county. A model can tell when ten pages say the same thing with a different town name swapped in.
The Bottom Line
ChatGPT and Gemini mention local businesses they can find, understand and corroborate from several places. There is no switch to flip, but clear pages, consistent details and genuine reviews give you a better chance of being the name they reach for.