preferly

Blog

Where ChatGPT, Claude and Gemini get their local recommendations

By the Preferly team · Published

Short answer: each assistant leans on a different kind of site. When ChatGPT answered a local question, 66.5% of the addresses it cited were a recommended business's own website. For Gemini it was 33.2%, and for Claude 7.4%. Claude cited directories and review sites most, and Gemini cited reddit.com and "best of" list sites.

These numbers come from our study, Do ChatGPT, Claude and Gemini recommend the same businesses?, in which we asked the three assistants the same 300 questions three times each, with web search on. The summary of the study has the other headline numbers.

How each assistant answers a local question

Assistant Businesses named per answer Cited addresses that are a named business's own site Most cited sites
ChatGPT 1.5 66.5% Mostly the recommended businesses' own sites (not listed: we name no business)
Gemini 4.4 33.2% reddit.com, threebestrated.com, fastexpert.com, smartasset.com, thecliniccompass.com, superlawyers.com
Claude 6.5 7.4% lantern.llc, reviews.birdeye.com, poyst.com, fresha.com, threebestrated.com, expertise.com

The three answer in very different shapes. ChatGPT named 1.5 businesses per answer on average and mostly pointed to their own websites. Claude named 6.5 and mostly pointed to sites that list many businesses at once. Gemini sat in between.

ChatGPT: the business's own site

Two of every three addresses ChatGPT cited for a local question were the website of a business it named. We don't list those sites, because this study names no business. The pattern held for software too: 84.2% of ChatGPT's cited addresses for software questions were a named product's own site.

Claude: directories and review sites

Claude cited a business's own site for only 7.4% of its local sources. Its most cited sites were directories, review and booking sites. For software, its most cited sites were toolradar.com, us.fitgap.com, g2.com, findstack.com and costbench.com, and 15.8% of its sources were a named product's own site.

Claude cited lantern.llc in 119, 124 and 123 of its 250 local answers in the three runs, and ChatGPT and Gemini cited it in none of their 1,800 answers. That depends on how Claude searched: asked 48 of the same questions through Claude's API with its built-in search, it cited lantern.llc in none, against 22 for those questions in the main run, in a small test that also changed how many questions were asked per request.

Gemini: reddit and list sites

Gemini cited a business's own site for a third of its local sources. Its most cited sites were reddit.com and sites that rank or list local businesses, such as threebestrated.com, fastexpert.com, smartasset.com, thecliniccompass.com and superlawyers.com. Not every Gemini answer had a source: 78 of its 750 local answers cited nothing, and for software it returned no citations for 147 of 150 answers.

What it means for a business owner

The sources differ, so what you'd check differs by assistant.

  • For ChatGPT, your own website is most of what it cited. Make sure it says plainly what you do, where, and for whom, and that it loads for crawlers.
  • For Claude and Gemini, third-party sites carried more weight in what they cited. Look up your business on the directories, review sites and local "best of" lists in your trade, and check that what they say about you is current.

To see what AI says when customers ask about businesses like yours, run the free check. It asks 3 AI models. To see who AI names instead of you, the sources it cited for them, and where it contradicts your website, get the full $99 AI visibility audit, paid once. It asks 10 AI models.

Limits

  • Cited isn't the same as caused. We measured which sites were cited. We didn't test whether a source caused a recommendation, or why the assistants read different sites.
  • These channels. ChatGPT was run in its app, Claude as agents with web search, Gemini through its API. In the small channel test on 48 questions, Claude's own-site share was 6% in the main run and 34% through its API, so sources depend on the channel too. For ChatGPT and Claude that test changed two things at once: the channel, and one question per request instead of 20 per file. The consumer Claude and Gemini apps weren't run.
  • No ranking of sites. The lists show which sites were cited most, not which are better.
  • US cities and English only, November 9 to 11, 2026.

Data: Preferly, "Do ChatGPT, Claude and Gemini recommend the same businesses?" (2026), https://www.preferly.ai/research/who-ai-recommends. 2,700 answers from ChatGPT, Claude and Gemini to 300 questions, November 9 to 11, 2026. The per-question data is free to reuse under CC BY 4.0 with that credit.

Questions

Does my own website matter for AI recommendations?

For ChatGPT it was most of what we saw cited: 66.5% of the addresses it cited for local questions were a named business's own site. For Claude it was 7.4%. We measured what was cited, not what caused a recommendation.

Which sites did Claude and Gemini cite most?

Claude's most cited sites included reviews.birdeye.com, poyst.com, fresha.com, threebestrated.com and expertise.com. Gemini's included reddit.com, threebestrated.com, fastexpert.com, smartasset.com, thecliniccompass.com and superlawyers.com.

Did every answer cite a source?

No. 78 of Gemini's 750 local answers had no citation. For software, Gemini returned no citations for 147 of its 150 answers.

Sources

Start free check

3 AI models, 5 customer questions, about a minute. Or check whether AI crawlers can reach your site.

Check your site free

No email. It reads your robots.txt and homepage, the way AI crawlers do.