How AI Assistants Decide Which Businesses to Mention

Two sources: memory and live retrieval

Most AI assistants answer from a mix of what the underlying model absorbed during training and what it looks up when you ask. Training data is a snapshot of the public web at some point in the past. Live retrieval is closer to a search: the tool fetches a handful of pages, reads them, and writes an answer from what it found.

The balance varies. Perplexity and Google AI Overviews lean heavily on retrieval and usually show their sources. ChatGPT, Claude and Gemini search the web for many questions, especially local or time-sensitive ones, and answer from memory for others. For a question like 'best family dentist in Tampa', retrieval tends to matter most, because the model's memory of a single local practice is thin or out of date.

Why different tools give different answers

Each tool uses its own search partner or index, its own way of picking which pages to read, and its own writing style. One may read your Google Business Profile and a local news article; another may read Yelp and a competitor's blog. That is why the same question can name you in one tool and leave you out in another on the same day.

It also means a single check in one tool tells you very little. A customer could be using any of them, and you have no way of knowing which. That is the reason Found in AI asks every tracked question to all five tools it covers: ChatGPT, Claude, Perplexity, Gemini and Google AI Overviews.

What retrieval tends to reward

When a tool reads pages to answer a question, it is looking for text that answers that question. Pages that state plainly what you do, where you do it and for whom are easier to use than pages full of slogans. A sentence such as 'We are a pediatric dental practice in South Tampa, open Saturdays, accepting most PPO plans' gives a tool something it can repeat.

Corroboration also seems to matter. When your own site, your Google Business Profile, a directory listing and a review site all describe you the same way, a tool has several independent sources pointing at the same facts. When they disagree on your name, address or services, it has less to go on.

• Clear, specific statements of service, location and audience
• The same business name, address and phone number everywhere
• Mentions on sites other than your own: reviews, directories, local press, associations
• Pages that crawlers can actually fetch and read as text

What you cannot influence

You cannot pay any of these tools to be included in an organic answer, and no one can promise what a model will say. Answers shift as models are updated, as the tools change which sources they read, and as competitors publish. Treat any vendor who promises placement with suspicion.

What you can do is remove the reasons a tool might skip you: a site it cannot read, facts it cannot confirm, a question your website never answers. The rest of these guides work through those one at a time.

Start by measuring, not guessing

Before you change anything, find out where you stand. Write down the questions a customer would ask before choosing a business like yours, and see what each tool says. Found in AI does this on a schedule: it records whether each answer names you, links your site, which competitors it names and in what position, and rolls that into a score, a trend and a list of gaps.

With a baseline you can tell whether a change you made coincided with a shift in the answers, instead of relying on one search you ran on your phone.