When someone asks ChatGPT for "the best plumber near me" or "a good family dentist," the answer names two or three businesses. There’s no results page to scroll, no ranking to check, no click to count — just an answer, and everyone who wasn’t in it.
That makes the decision feel opaque, but it isn’t arbitrary. AI engines lean on a small set of signals, and most of them are things a business already controls.
Two different ways an engine can "know" you
The first distinction that matters: some engines search the live web when they answer, and some answer purely from what the underlying model already learned during training. Perplexity, Google’s AI Overviews and Microsoft Copilot lean toward the first kind — they can pick up a change to your site or your listings in weeks. A model answering from training data alone only updates when that model is retrained, which is a much slower clock.
That’s why a business can fix something today and see its technical audit score move within hours, while the AI answers themselves take longer to catch up. They’re measuring different things, on different clocks.
The signals that actually feed the answer
Underneath both kinds of engine, four things tend to decide whether a business gets named: whether AI can technically read the site at all (clean markup, structured data, an llms.txt file); whether the content on the page actually answers the questions people ask, instead of just describing the business in marketing language; whether the business has an established, verifiable identity — the kind of authority signal Google’s Knowledge Graph or a directory profile provides; and what its reputation looks like across the reviews AI can read.
None of these are secret. They’re the same four pillars Findelle’s AI Readiness Audit checks — infrastructure, content, authority and reputation — because they’re the same things an AI model is implicitly weighing when it decides who to name.
What a business can do about it
The honest version: there’s no way to force an engine’s hand, and anyone who promises guaranteed placement in an AI answer is overselling. What’s actually possible is making a business as easy as possible for AI to read and as easy as possible to trust — clean technical signals, content written in the way customers actually ask questions, a verifiable identity, and a reputation that’s structured so a model can quote it with confidence. That’s the whole job. The rest is patience while the slower-moving engines catch up.