Guides

How AI assistants pick which businesses to recommend

What the evidence says about how ChatGPT, Google AI Overviews, Copilot and Perplexity choose businesses: Ahrefs' studies, Google's guidance, BrightLocal and our data.

The short version

Nobody outside the AI companies knows the exact recipe, and anyone who says they do is selling something. The evidence points the same way, though. Assistants recommend businesses that their search can find, that other sites already name (especially "best of" lists and review sites), and whose own pages state the facts plainly. For Google's AI features, Google says the requirements are the same as ordinary Search. Almost every reliable lever is one you'd pull for SEO anyway.

This page pulls together what's been published about how assistants choose, from the companies themselves and from the largest independent studies, and adds what our own crawler data shows. We've kept each claim tied to its source, because this is a topic where confident advice runs well ahead of evidence.

Step one: the assistant has to find you

Assistants don't recommend from memory alone when a question needs current, local information. They search, then answer from what the search returns.

Assistant Where its web results come from What that means for you Source
ChatGPT Sites surfaced by OAI-SearchBot, plus third-party search providers, which OpenAI says include Bing Allow OAI-SearchBot; be visible in Bing OpenAI crawlers, ChatGPT search help
Google AI Overviews and AI Mode The Google Search index Same as ranking in Google: indexed and eligible for a snippet Google
Microsoft Copilot The Bing index Allow Bingbot; be visible in Bing Bing's crawler docs, see the registry
Perplexity Its own index, built by PerplexityBot Allow PerplexityBot Perplexity
Claude Its own search crawler (Claude-SearchBot) and live fetches Allow Claude-SearchBot and Claude-User Anthropic

Google is the clearest about this. "There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary", and a page "must be indexed and eligible to be shown in Google Search with a snippet". Its May 2026 guide puts it more bluntly: "optimizing for generative AI search is optimizing for the search experience, and thus still SEO" (Google).

So the first test is mechanical. If robots.txt blocks the search crawler, or your text only appears after JavaScript runs, the assistant can't pick you because it never saw you. The access checker tests both.

Step two: other sites have to name you

Once an assistant has search results, it leans heavily on pages that compare businesses. Two Ahrefs studies are the best evidence we have.

"Best X" lists dominate what ChatGPT cites. Ahrefs ran 750 recommendation-style prompts through ChatGPT across software, products and agencies, and found that "best X" blog lists made up 43.8% of all page types cited. Brands placed higher on third-party lists were more likely to be recommended, though the effect was weaker on lists of ten or more. Freshness mattered too: 79.1% of the cited lists had been updated in 2025 (Ahrefs, Dec 2025). Ahrefs is careful to call this a correlation.

Brand mentions beat backlinks for AI Overviews. In a study of 75,000 brands, the factor most correlated with appearing in Google's AI Overviews was branded web mentions (0.664), ahead of branded anchors (0.527), branded search volume (0.392), Domain Rating (0.326) and the number of backlinks (0.218) (Ahrefs). Being talked about, by name, on other sites appears to matter more than link counts.

Neither study looked only at local businesses. For a plumber or a dentist, the equivalents of "best X" lists are local roundups, directories, review platforms and local news. BrightLocal's 2026 consumer survey notes that AI answers draw on reviews, local directories, business websites, social media and other third-party sources, depending on the tool and the prompt.

Step three: your own page has to answer the question

When an assistant does open your site, it's often doing so live, for one person. In our Observatory data, 35% of AI fetches on small-business sites were user-triggered (ChatGPT-User alone was 28.1% of the typical site's AI fetches). That's an assistant reading your page while a customer waits for an answer.

What it needs from the page is plain: who you are, what you do, where, for whom, roughly what it costs, and how to book. Google's guide says much the same in its own terms, asking for "non-commodity" content with original detail rather than generic copy, and it says you don't need special AI files, AI-specific rewrites or "chunked" content. Structured data helps state your facts without ambiguity, though Google says there's no special schema required for its AI features.

What customers do with the answer

BrightLocal's 2026 survey of 1,002 US adults found that 45% had used an AI tool for local business recommendations in the past year, up from 6% the year before, with ChatGPT the most used (31%) followed by Google AI Mode (23%). Trust is conditional, though: 97% of AI users said they sometimes double-check AI recommendations against real reviews, and 42% always do (BrightLocal).

So a recommendation is the start of a check, not the end of the sale. A business that gets named and then has thin or stale reviews loses the customer at the second step.

What we'd do, in order

  1. Fix access. Allow AI search crawlers and make sure your text is in the HTML. It takes an hour and nothing else works without it.
  2. Get onto the lists that get cited. Ask to be included in local "best of" roundups and relevant directories, keep your profiles complete, and earn mentions in local press. That's the clearest pattern in the Ahrefs data.
  3. Keep reviews current. Customers check AI answers against them.
  4. Write the page an assistant would want to quote: services, area, prices or ranges, hours, and answers to the questions customers ask you on the phone.
  5. Measure the right thing. AI assistants sent a median of 0.03% of visits in our sample, so your analytics won't tell you much. Ask the assistants directly every month, or use one of the tracking tools we compare.

What we don't know

No AI company publishes how it ranks businesses in an answer, and answers vary by wording, location and model version. The studies above are correlations from large prompt sets, mostly outside local search. Our own data shows what AI agents do on small-business sites, not what they say afterwards. We'll add to this page as better evidence appears, and we'd rather leave a gap than fill it with a guess.

Questions people ask

How do I get ChatGPT to recommend my business?

Be findable by its search (allow OAI-SearchBot, rank in the search engines it draws on), be named on the third-party lists and review sites it cites, and state on your own pages, in plain HTML, what you do, where and for whom. Ahrefs found 'best X' list pages were 43.8% of the pages ChatGPT cited for recommendation prompts.

What do AI Overviews need to show my business?

Google says there are no additional requirements: the page must be indexed and eligible to show a snippet in Google Search. Google's AI optimization guide says optimizing for its AI features is still SEO.

Do reviews affect AI recommendations?

Consumers check them. BrightLocal's 2026 survey found 97% of people who use AI for local recommendations sometimes double-check them against real reviews, and 42% always do. AI answers also draw on review and directory sites.

Do backlinks matter for AI visibility?

Less than brand mentions, in Ahrefs' study of 75,000 brands. Branded web mentions had the strongest correlation with appearing in AI Overviews (0.664), against 0.218 for the number of backlinks. That's a correlation, not proof of cause.