Do AI assistants recommend businesses, or just big names?
Someone in your circle has probably said it by now — "oh, I just asked ChatGPT" — about something you'd expect them to Google. A tradesperson, a supplier, an accountant. It's a small sentence that raises a bigger question: is that thing actually recommending real local businesses, or is it just reaching for whichever names it already knows, the way a stranger at a dinner party reaches for the one brand they can remember?
The honest answer is: it depends entirely on what you ask it to do, and most people never notice the difference. Before you can work out whether you should show up in these answers, it's worth understanding what's happening when someone asks.
Two very different things are happening behind one text box
When you type a question into ChatGPT or Copilot, one of two things happens, and the assistant doesn't always make it obvious which.
It might answer from memory. Large language models are trained on a huge snapshot of text and can produce an answer without looking anything up. A plain, un-searched reply may simply be reciting whatever pattern of names it absorbed during training — which can be out of date, generic, or in the worst case, entirely made up. This is where the well-documented habit of AI chatbots inventing plausible-sounding businesses that don't exist comes from: no lookup happened, so nothing kept the answer honest.
Or it might search first. ChatGPT's search function and Copilot both work by querying a live index, reading what comes back, and writing an answer grounded in those results — with sources you can click through and check. That index, for both of them, is built on Bing rather than Google. For local questions specifically, Copilot leans further still: it pulls from verified Bing Places listings to construct answers to things like "find a plumber near me who's open now", with structured details — name, address, phone, reviews — feeding the response.
Neither mode announces itself loudly. The tell is whether the answer comes with sources attached. No sources, no citations, no little link icons — treat the names in that answer as memory, not research.
Run the test yourself, properly
Don't rely on one go. Open ChatGPT and ask it plainly:
"I need a [what you do] in [your town]. Who would you recommend and why?"
Note whether it shows sources. If it does, click through — are those real listings, real reviews, your actual competitors? If it doesn't, ask a follow-up: "Did you search the web for that, or is that from memory?" It will tell you honestly.
Then do the same in Copilot, and in whichever assistant your phone defaults to. Ask a second way — "who's well reviewed for X near me" reads differently to a search index than "recommend me an X". You're not looking for your own name to appear once; you're looking for a pattern across a few honest attempts.
What decides who gets named, once it's actually searching
Once an assistant is genuinely retrieving rather than recalling, the businesses it names are the ones its index can describe with confidence. Generic copy doesn't just fail to persuade a reader — it fails to get retrieved at all. Both Google and the Bing index that ChatGPT and Copilot lean on have to work out what a business actually is before they can hand it to anyone, and a page that describes itself the way a thousand competitors describe themselves gives the index nothing distinctive to hold on to. Specificity isn't a style choice at that point; it's the difference between being findable and being noise the system quietly skips.
That's also why a business can be comfortably placed on Google and still miss entirely from these answers. Google and Bing are separate indexes built by separate crawlers, and years of attention paid to one doesn't automatically transfer to the other. A directory listing that's accurate on Google Business Profile but unclaimed or thin on Bing Places is invisible to exactly the systems now fielding "who should I use" on your customer's behalf.
What this means before you do anything else
Two honest possibilities cover most businesses reading this. Either the assistants you tested are answering from memory — in which case nothing you do about your website will change today's result, because nothing was looked up — or they're genuinely searching, and what you saw is a reasonably accurate read of how findable you currently are on the index that matters. Telling those two apart is the whole point of running the test properly rather than panicking at the first paragraph.
If you'd rather not run the prompts across three assistants and guess at which mode you're seeing, that's exactly what our free visibility check does — we look at where your business actually stands on Google and on the AI assistants people are increasingly asking instead, and send back a plain-English report on what's helping, what's invisible, and what we'd fix first. We've written before about the specific reasons a business goes missing from ChatGPT's answers once you know it's genuinely searching — worth a read if this test has you a little rattled. Either way, guessing is optional. Finding out isn't hard.
← All posts