How do AI assistants decide which local businesses to recommend?
Ask ChatGPT, Gemini or Claude for a good plumber, dentist or restaurant nearby and you get something a search engine never gave you: a short list of actual business names, with a sentence on why each one. It feels like the assistant has an opinion. It doesn't. It has sources.
Understanding where those answers come from is the difference between hoping you get named and knowing why you don't. This guide walks through what genuinely shapes an AI recommendation, without the jargon, so you can see the mechanics behind a moment that increasingly decides who your next customer chooses.
An assistant repeats the web, it doesn't invent it
A language model doesn't hold a private directory of the best businesses in your town. When you ask for a recommendation, it draws on two things: what it absorbed during training, and, increasingly, what it can pull from the live web while it answers. Both are just reflections of what the internet already says about local businesses.
That's the single most important idea to hold onto. An assistant is a mirror. If the web presents your business clearly and often, the mirror shows you. If the web barely mentions you, or describes you vaguely, there's nothing for the assistant to reflect, so it names someone else. Being good at what you do isn't enough on its own; the assistant can only work with what it can find, and it will confidently recommend a lesser competitor who happens to be more legible online.
It leans on the places the web already trusts
When an assistant looks something up, it gravitates toward sources that are structured, current and widely referenced: local directories, review platforms, established listings and clear business pages. These are the sites that already organise the world into "best X in Y" answers, so they're the easiest for a machine to read and repeat.
This is why a business can be genuinely excellent and still be invisible to AI. If the sources an assistant reads don't mention you, you effectively don't exist in that conversation. The assistant isn't ignoring you on purpose. It simply never encountered you where it was looking, and it can only recommend from what it encountered.
Reputation counts more than a polished homepage
A beautiful website reassures humans, but it's weak signal to an assistant deciding who to name. What carries more weight is corroboration: being mentioned, reviewed and described consistently across the places people and machines already trust. A steady, recent trail of real feedback tells an assistant you're active, real and worth surfacing.
Silence is the problem. A business with almost no third-party footprint gives an assistant no reason to pick it over a competitor with a visible track record, even if the quiet business is better. To a model weighing who to recommend, an absence of evidence looks a lot like an absence of the business.
Clarity gets you quoted
Assistants prefer information they can lift and restate confidently. Plain descriptions of what you do, who you help and where you do it are easy to quote. Vague, promotional copy is not; there's nothing concrete to repeat, so it gets skipped in favour of a business that spells things out.
It also matters that your details agree with each other everywhere they appear. When your name, location and services are consistent across the web, an assistant can trust and match them. When they conflict, it hedges and names someone it's more sure about, because a recommendation is a small act of vouching, and it won't vouch for a business it can't pin down.
Every assistant reads the web differently
ChatGPT, Gemini and Claude don't share one view of the world. They draw on different data, refresh it on different schedules and weigh sources differently. So the same question can produce three different shortlists. You might be named confidently by one and left out entirely by another.
Answers also shift from one run to the next, even within a single assistant. Ask twice and you can get two different lists. That's why a single lucky screenshot tells you almost nothing; what matters is how often you appear across many answers and across all three assistants, not whether you showed up once.
It's a moving target, not a setting
None of this holds still. As reviews accumulate, listings change and the assistants update what they read, the shortlist for your category quietly reshuffles. A business that's named today can fade as competitors gain ground, and one that's invisible now can break in as its online presence strengthens.
That's worth sitting with, because it cuts against the instinct to treat visibility as something you fix once and forget. It behaves more like your reputation than your website: always being written by what the wider web says about you, and always changing. The businesses that stay named are the ones that keep an eye on it, the way they already watch their reviews.
What this means for your business
None of this is about tricking a model. It's about whether the open web gives an assistant enough reason to name you when a customer asks. Most owners have never actually looked, so they don't know whether AI recommends them, recommends a competitor, or has never heard of them at all.
That's the gap a Peekl check closes. It asks the assistants the questions your customers ask, measures how often you're named across ChatGPT, Gemini and Claude, and shows you exactly where you stand today, before you decide what to do about it. You can't manage a recommendation you've never seen. For the flip side of this, see why so many good businesses stay invisible to AI.
Common questions
Can I pay to be recommended by ChatGPT?
Why does ChatGPT recommend a competitor instead of me?
If I rank well on Google, will AI recommend me too?
See where you stand
Reading about AI search is useful; knowing your own place in it is what helps. A free Peekl check asks the assistants the questions your customers ask and shows whether ChatGPT, Gemini and Claude recommend you today, who they name instead, and how often.