Skip to content

Research · October 2026

How AI assistants recommend local businesses: 29 questions, 10 countries

We asked AI assistants the questions customers ask in 10 countries, several times each, and checked what they said about 23 businesses. Eight findings stood out.

Countries
10
Industry × city questions
29
Requests to AI engines and search
461
Businesses checked by name
23
  1. 01

    Each AI engine has its own list

    For the same question on the same day, the businesses named by ChatGPT and by Gemini overlapped very little (Jaccard similarity 0.22 in the pilot, 0.20 in the US/DACH round, 0.17 in the UK/EU round). In Denver, no plumber appeared on both lists. Google AI Mode often named a third set.

    0.17–0.22

    overlap between ChatGPT and Gemini

  2. 02

    Answers change from run to run

    Asking ChatGPT the same question five times gave partly different lists every time. Overlap between runs ranged from 0.36 to 0.54 depending on the round. One screenshot of an AI answer, good or bad, proves little.

    44 of 90

    businesses named in only 1 of 5 runs (pilot)

  3. 03

    The ChatGPT card is your profile

    In the UK and EU, ChatGPT’s business cards showed exactly the rating and review count from the Google Business Profile. In the US, 159 of 341 cards (47%) carried Yelp’s logo and Yelp’s numbers. In Denver, all ten dental cards we checked showed Yelp ratings: 3.9 stars on average versus 4.75 on Google for the same practices.

    0 of 307

    UK/EU ChatGPT cards with a Yelp logo

  4. 04

    AI gets facts wrong: mostly because sources disagree

    When we asked about specific businesses, engines gave wrong opening hours, phone numbers from look-alike sites, outdated prices and another location’s hours. In 12 of the 21 businesses we checked in detail, the business’s own sources disagreed. The cleanest answers came from businesses whose website and Google profile agreed.

    12 of 21

    businesses whose own website and Google profile contradicted each other

  5. 05

    AI almost never knows prices

    Category answers (“a good dentist in Milan”) almost never included prices. Questions about a specific business did produce prices, and that is where most errors were: old pre-discount prices, another practice’s prices, dollars for a UK roofer.

    1 of 72

    continental-EU category answers that mentioned a price

  6. 06

    The question’s language changes the answer

    Asked in German and in English about dentists in Vienna, ChatGPT’s two answers shared only three practices. Practices that state the languages they speak were the ones named for “English-speaking dentist”.

    3 of 21

    dentists named in both German and English (Vienna)

  7. 07

    AI cites directories more than business websites

    The sources AI engines cited were mostly review sites and national directories: Checkatrade and Yell in the UK, Doctolib and Jameda in Germany and France, Herold in Austria, Doctoralia in Spain, MioDottore in Italy, ZnanyLekarz in Poland, Avvo and Justia for US lawyers, Angi and the GAF directory for US roofers.

    by country

    Checkatrade, Doctolib, Herold, ZnanyLekarz…

  8. 08

    Google’s AI Overviews rarely appear for local questions

    For local “who should I call” questions, Google mostly showed its map results instead of an AI Overview. AI Overviews did appear for price questions. Google’s AI Mode is the AI surface to watch for local businesses.

    3 of 29

    category + city questions with an AI Overview

One question, two engines

The clearest example from the pilot: the same plumbing question in Denver, asked five times in ChatGPT and three times in Gemini on the same day. Each engine was fairly consistent with itself, and the two lists had nothing in common.

For a business owner this means two things. Being named in one assistant says little about the others. And the work that moves all of them is the same: complete, consistent profiles and directory listings that every engine reads.

Question:“Can you recommend a good plumber in Denver?”

2026-10-03

ChatGPT5 answers

  • Plumber B
  • Plumber C
  • Plumber A
  • Plumber D
  • Plumber E
  • Plumber F
  • Plumber G
  • Plumber H

Gemini3 answers

  • Plumber I
  • Plumber K
  • Plumber J
  • Plumber L
  • Plumber M

0businesses on both lists

named in that answerPilot measurement, 3 Oct 2026, Denver (US). Business names anonymized.

Method

  • When: 3 October 2026, in three rounds.
  • Where: United States, United Kingdom, Germany, Austria, Switzerland, France, Spain, Italy, Poland, Netherlands.
  • Industries: dentists, med spas and aesthetic clinics, lawyers, plumbers, roofers, HVAC, movers, cleaning companies.
  • Engines: ChatGPT and Gemini as seen in their web interfaces (collected through DataForSEO), Google AI Mode, Google Search with its local results, and Perplexity’s API. ChatGPT was asked up to five times per question, Gemini up to three.
  • Questions: in the local language, with the city written into the question. Country was set for each engine; the city could not be set for ChatGPT.
  • Business checks: for 23 businesses, we asked each engine about the business by name and compared its answers with the business’s website and Google profile.
  • Overlap is measured as Jaccard similarity: shared names divided by all names in the two lists.

Limits

The samples are small and all from one day. Answers for logged-in users, paid plans or a user’s own location can differ, and we did not check Copilot, Claude or Grok. Business names are anonymized here; we report patterns, not judgments about specific companies. We will repeat these measurements on a schedule and publish changes.

Larger studies agree on the main points: BrightLocal’s analysis of 200,085 local prompts found that businesses present in Google Maps results appear in only about a third of ChatGPT answers, and that asking the same question again repeats only 20–33% of named businesses (September 2026).

Use this data

You can quote and chart these findings with a link to this page. Journalists and researchers can ask us for the per-question counts.

Run the same check on your business

Free, about a minute. Same method, your industry and city.

Free 60 seconds no call, no email for the first resultNo website? Use business name + city

How we measure