How ChatGPT, Claude & Perplexity decide which businesses to recommend.

AI tools don't keep a list of good businesses. When someone asks for a recommendation, they search the web, read a handful of pages and write an answer from what they found. Your website is one of the things they read. What other sites say about you is another.

When someone asks Perplexity for a good commercial electrician in Phoenix, it doesn't look up a ranking it keeps somewhere. It searches, reads and writes. Every answer is put together fresh from whatever it found at that moment.

That's good news for a business owner, because much of what it finds is yours to change. This post walks through how the tools pick who to name, and which parts you can do something about.

They turn one question into several searches.

Google calls the first step "query fan-out". The tool breaks a question into several related searches and runs them all. In Google's own example, a question about fixing a lawn full of weeds might set off searches for the best weed killers, for removing weeds without chemicals and for stopping weeds coming back.

A question about your trade works the same way. "Who should I hire to replace a warehouse roof in Denver?" might become searches for commercial roofers in Denver, for warehouse roof replacement, for coating versus replacing a roof and for reviews of local firms. A page that answers any of those searches can end up in the answer.

Each tool searches its own sources.

Two things follow. First, doing well on Google doesn't cover you everywhere. Bing feeds Copilot and is one of ChatGPT's sources, and many businesses have never checked whether Bing has their site. Bing Webmaster Tools is free, and since February 2026 it has had an AI Performance report that shows which of your pages Copilot cites.

Second, every one of these tools has to be let in. If your site blocks one of those crawlers, that tool is answering without your website. See is your website blocking ChatGPT and Claude.

They read your pages for plain facts.

Once the tool has your page, it looks for facts it can match to the question. For a business, those are mostly simple:

• What you do, in the words a customer would use
• Where you do it: towns, counties, states or regions
• Who you do it for: homeowners, facility managers, developers or public agencies
• Proof: projects, sizes, certifications, years in business and clients you're allowed to name
• How to reach you: a phone number and email in the page text

A site that states these on clear pages is easy to match. A site that spends its home page on a slogan, lists services only in a menu and keeps its phone number in an image makes the tool guess. We cover the reading problems in why AI tools can't read some websites.

Structured data helps at the edges. It states your name, address, phone and services in a format machines don't have to interpret. It's worth having, but adding it made no measurable difference to AI citations on its own when Ahrefs tested it on 1,885 pages. See does structured data help you show up in AI answers.

They read what other sites say about you.

A tool deciding whether to recommend you reads more than your own pages. It reads reviews, directories, trade association lists, local news and anyone else who has written about you.

The numbers point the same way. Ahrefs looked at 75,000 brands in 2025 and compared their visibility in Google's AI Overviews with things like backlinks and mentions. Mentions of the brand across the web had the strongest link, a correlation of 0.664. Backlinks came in at 0.218. Ahrefs points out that correlation doesn't prove cause. It still fits how the tools work.

For local questions it's more direct. OpenAI says ChatGPT may use a person's approximate location, and third-party providers of local business listings, to answer questions about nearby businesses. Google says its local results depend on relevance, distance and prominence, and that prominence comes partly from how many websites link to you and how many reviews you have.

Google also warns that chasing fake mentions doesn't help. The ones that count are the ones you'd earn anyway: reviews from real clients, listings in the directories your trade uses and coverage of real projects.

The answer changes every time someone asks.

Ask the same question twice and you'll often get a different list. In January 2026 SparkToro and Gumshoe published a study in which 600 volunteers ran 12 prompts through ChatGPT, Claude and Google's AI 2,961 times. The chance of ChatGPT or Google's AI giving the same list of brands twice was less than 1 in 100. The same list in the same order came up roughly once in 1,000 runs.

What held steadier was how often each brand appeared across many runs. That's the useful measure: the share of relevant questions where you're named. Your position in any one answer tells you very little. We explain how we measure it in what an AI visibility report shows.

Your website is the part you control.

You can't control the model or pay your way onto the list. What's on your website is yours to fix. What's said about you elsewhere you can influence over time.

Our free AI Visibility Check looks at the left-hand column. It takes about 30 seconds and shows what AI tools see on your site today. The AI Search Tune-Up fixes what it finds, from $1,500 for a site of up to 30 pages, and includes a before and after report of who gets named when your customers ask.