How AI assistants decide which local businesses to recommend
When someone asks an AI assistant “who should I call about a leaking water heater,” the assistant interprets the question, looks up live sources about businesses near that person, weighs what those sources agree on, and writes back a few names with a reason for each. That whole decision runs on public information about your business. This post walks through it step by step, because every step is something you can influence.
What happens in the seconds after someone asks
The assistant does four things fast: it works out what’s being asked (a service, a place, an urgency), it pulls in sources (search results, listings, review pages, your site), it weighs what those sources say against each other, and it writes an answer naming the businesses the sources best support. There’s no secret ranking file. The answer is assembled fresh, from what’s publicly readable, nearly every time.
The lookup step is documented behavior, not our guess: OpenAI describes ChatGPT searching the web and linking sources1, and Google describes its AI answers issuing multiple related searches over its index before responding2. The two pipelines feeding those answers are worth understanding on their own, but the short version is that for local questions, the live lookup does most of the work.
What AI weighs before it names you
Three buckets of information, roughly in order of how much you control them.
What your own site says
Whether your pages state plainly what you do, where you do it, and answers to the questions customers actually ask. Brochure copy (“quality service since 1998”) gives the assistant nothing to work with. A page that says “we replace water heaters in [city], usually same day, licensed and insured” gives it exactly the sentence it needs.
What your reviews say
Volume, recency, and the words inside them. Review signals are their own subject, but the part that surprises owners is the text: when thirty reviews mention fast emergency response, the assistant can repeat that as a reason to pick you.
What third parties say about you
Directories, your Google Business Profile, press, forums, professional listings. This bucket matters because it’s corroboration: sources you don’t control agreeing with the ones you do.
Why AI names some businesses and skips others
Agreement wins. A business whose site, profile, listings, and reviews all tell one consistent story is easy for a machine to trust. A business whose sources disagree (two phone numbers, an old address, a category that doesn’t match the reviews) is risky to recommend, and the assistant has other candidates one line away. Skipping isn’t punishment. It’s a machine picking the names it can best defend from the sources in front of it, and conflicting information is also how it ends up wrong about you when it does name you.
What AI cannot see
Some businesses are invisible for mechanical reasons, not reputational ones. Content locked in PDFs and images of text. Sites that only render with JavaScript some crawlers don’t run. Crawlers blocked outright by a security plugin nobody remembers installing, which is worth checking on your own site directly. If the assistant can’t read you, everything else on this page is moot; readability comes first.
What this means for a real business
An illustration, labeled as one: picture two plumbers in the same mid-size city. One has 210 reviews with long gaps, a site that’s one big photo, and a profile listing the shop they moved out of in 2023. The other has 90 reviews arriving steadily, a plain-HTML site that answers the eight questions customers ask, and listings that agree everywhere. The second plumber is easier for a machine to defend, and machine-made shortlists reward defensibility over size. That’s the honest reason this work favors diligent small operators: the inputs are public, and most of them are fixable without anyone’s permission.