Home / Research / Local AI answers study
Research · November 2026 wave
Local AI answers study
I asked four AI assistants "Who is the best [profession] in [city]?" for 5 professions in 25 US cities, two times each. That is 1,000 answers. This page shows which businesses and professionals the assistants name, which sources they cite, and how much the answers change.
Key findings
Four assistants, four different lists
- 79.7%
- of the businesses and professionals named were named by only one of the four assistants (3,696 of 4,635)
- 1.5%
- were named by all four assistants (69 of 4,635)
- 0%
- of questions got the same list of names from ChatGPT two times in a row (0 of 124)
- 92%
- of Google AI Mode answers linked to a Google Maps listing (230 of 250). The other three assistants linked to none
- The assistants rarely agree. Across 125 questions they named 4,635 businesses and professionals. 79.7% came from one assistant only, 13.7% from two, 5.1% from three and 1.5% from all four.
- But almost each market has a shared name. In 92% of questions (115 of 125), one name or more came from three assistants or more. In 39.2% (49 of 125), one name came from all four.
- The same question gives a different answer the next time. When I asked again a few seconds later, the first name in the answer was the same in 52.7% of cases (255 of 484).
- Most of the different sites are local, but they get the smaller part of the citations. The answers linked to 1,153 different sites. 84.5% of them appeared for one city only (974 of 1,153). Sites that appeared for one or two cities got 36.9% of the citations (1,914 of 5,189).
- A small group of sites gets most citations. 124 sites other than Google Maps appeared for 3 cities or more. They got 58.7% of the citations (3,045 of 5,189). Google Maps listings got 4.4%.
- Yelp and Facebook are almost absent as sources. 2.6% of answers linked to Yelp as a source (26 of 1,000). ChatGPT and Gemini never linked to it. No answer linked to Facebook (0 of 1,000). An answer can still name Yelp or Facebook in its text.
Agreement
How many assistants name the same business?
| One assistant only | 79.7% | |
|---|---|---|
| Two assistants | 13.7% | |
| Three assistants | 5.1% | |
| All four assistants | 1.5% |
For a local business this means one assistant is not "AI". A business that ChatGPT names can be absent from Gemini, Perplexity and Google AI Mode for the same question on the same day. A person and that person's firm count as two names here, which makes agreement look lower than it is. The limits give the size of that error.
Consistency
Ask the same question two times
Each question went to each assistant two times, a few seconds apart. The table compares the two answers.
| Assistant | Same list of names | Same first name | Names in both answers (mean) | No name in common | Cited sites in both answers (mean) |
|---|---|---|---|---|---|
| Google AI Mode | 1.6% 2 of 125 | 31.5% 39 of 124 | 35.6% | 10.4% | 76.9% |
| ChatGPT | 0% 0 of 124 | 45.9% 51 of 111 | 29.6% | 13.7% | 51.5% |
| Gemini | 1.6% 2 of 125 | 48.8% 61 of 125 | 41.4% | 0.8% | 55.5% |
| Perplexity | 24% 30 of 125 | 83.9% 104 of 124 | 69.8% | 0.8% | 99.3% |
"Names in both answers" is the count of names in both answers divided by the count in one or the other. Perplexity used the same sources in the same order in 116 of 125 pairs, but it wrote new text each time. No assistant gave the same text two times.
Sources
What each assistant cites
I sorted each cited site into one of three groups. A Google Maps listing is a link to a business listing on Google. A multi-market site was cited for 3 cities or more in this study: directories, ranking sites, review sites and publishers. A local site was cited for one or two cities. In a sample of 40 of these, 33 were the website of one business and the others were local directories or publishers. Google AI Mode answers come from Google search results and the other three from each provider's API, so the form of the sources is not the same.
| Assistant | Google Maps listing | Multi-market site | Local site | Only multi-market sites | Sites per answer (mean) | Businesses per answer (mean) |
|---|---|---|---|---|---|---|
| Google AI Mode | 92% | 18.8% | 6% | 4.4% | 1.23 | 6.54 |
| ChatGPT | 0% | 99.2% | 68.8% | 31.2% | 5.29 | 11.05 |
| Gemini | 0% | 98.8% | 99.6% | 0.4% | 7.84 | 10.96 |
| Perplexity | 0% | 100% | 86.4% | 13.6% | 6.4 | 4.69 |
The 25 most cited sites
| Site | Share of answers | Answers | Cities | Google AI Mode | ChatGPT | Gemini | Perplexity |
|---|---|---|---|---|---|---|---|
google.com | 23% | 230 | 25 | 230 | 0 | 0 | 0 |
expertise.com | 12.9% | 129 | 24 | 0 | 83 | 10 | 36 |
energysage.com | 12.8% | 128 | 24 | 4 | 39 | 37 | 48 |
healthgrades.com | 11.4% | 114 | 25 | 0 | 60 | 0 | 54 |
clearlyrated.com | 11.3% | 113 | 24 | 2 | 28 | 35 | 48 |
doctor.webmd.com | 11.1% | 111 | 25 | 0 | 13 | 8 | 90 |
solarreviews.com | 10.2% | 102 | 25 | 4 | 36 | 38 | 24 |
clutch.co | 9.6% | 96 | 25 | 0 | 19 | 31 | 46 |
dentistsranked.com | 9.5% | 95 | 25 | 0 | 50 | 9 | 36 |
reviews.birdeye.com | 8.7% | 87 | 22 | 0 | 77 | 10 | 0 |
ecowatch.com | 8.6% | 86 | 24 | 0 | 4 | 38 | 44 |
justia.com | 8.6% | 86 | 25 | 0 | 0 | 42 | 44 |
attorneys.superlawyers.com | 7.7% | 77 | 25 | 3 | 17 | 41 | 16 |
bestinhood.com | 7% | 70 | 12 | 0 | 44 | 0 | 26 |
reddit.com | 6.8% | 68 | 22 | 0 | 43 | 25 | 0 |
bestlawfirms.com | 6.1% | 61 | 22 | 0 | 6 | 16 | 39 |
bestlawyers.com | 5.6% | 56 | 25 | 0 | 0 | 6 | 50 |
zocdoc.com | 5.1% | 51 | 17 | 3 | 8 | 8 | 32 |
bbb.org | 5% | 50 | 18 | 0 | 0 | 0 | 50 |
avvo.com | 4.5% | 45 | 24 | 0 | 43 | 0 | 2 |
dentascore.com | 4.5% | 45 | 19 | 0 | 35 | 0 | 10 |
forbes.com | 4.5% | 45 | 15 | 0 | 0 | 23 | 22 |
thumbtack.com | 4.4% | 44 | 18 | 13 | 1 | 12 | 18 |
serviceagent.ai | 4.1% | 41 | 17 | 0 | 9 | 32 | 0 |
hearingup.com | 3.8% | 38 | 17 | 0 | 0 | 28 | 10 |
The last four columns show the number of answers from each assistant that cited the site, out of 250. google.com here means Google Maps listings.
By profession
Results for each profession
| Profession | Named by one assistant only | A business named by 3+ assistants | Answers that cite a local site | Most cited site |
|---|---|---|---|---|
| Dentist | 79.3% | 96% | 65.5% | dentistsranked.com 47.5% |
| Audiologist | 78.4% | 88% | 71.5% | healthgrades.com 42% |
| Personal injury lawyer | 86.2% | 84% | 63% | justia.com 43% |
| Accountant | 77% | 92% | 64.5% | clearlyrated.com 56.5% |
| Solar installer | 74.9% | 100% | 61.5% | energysage.com 64% |
By city size
| City size | Named by one assistant only | A business named by 3+ assistants | Same first business on the second run |
|---|---|---|---|
| City population over 1,000,000 | 80.9% | 95% | 51.6% |
| City population from 300,000 to 1,000,000 | 80.4% | 91.1% | 51.7% |
| City population under 300,000 | 77.4% | 90% | 54.9% |
By assistant
Results for each assistant
- Google AI Mode6.54 businesses and 1.23 cited sites in a mean answer. Most cited:
google.com. - ChatGPT11.05 businesses and 5.29 cited sites in a mean answer. Most cited:
expertise.com. - Gemini10.96 businesses and 7.84 cited sites in a mean answer. Most cited:
justia.com. - Perplexity4.69 businesses and 6.4 cited sites in a mean answer. Most cited:
doctor.webmd.com.
Data
Download the data and cite it
The dataset has one row for each of the 1,000 answers: the question, the assistant, the model, the time, the number of businesses and professionals named, the number of cited sites, and the name of each cited site that appeared for 3 cities or more. The findings file holds each number on these pages with its count and its base.
The answer text and the names of the businesses are not in the public files. The methodology says why. To cite this study:
LiButti, Cameron. "Local AI Answers Study." November 2026 wave, published November 18, 2026. cameronlibutti.com. https://cameronlibutti.com/research/local-ai-answers/
Collected November 18, 2026. The plan is one new wave each month, with the same questions and cities. The methodology has the question list, the cities, the models, the limits and the corrections policy.
Author
Who ran the study
I'm Cameron LiButti, a former mechanical engineer and the founder of Bidview Marketing. My team works on how local businesses show up in AI answers, so I wanted numbers that do not come from our own clients. No Bidview client is inside a city of this study. Some clients are in suburbs of Houston, Denver and Chicago. More about my method is on the marketing engineer page. Found an error? Use the contact page.