Study · October 2026
Do AI assistants agree on which local businesses to recommend?
We asked ChatGPT, Gemini, Perplexity and Claude who they recommend for five local services in 27 US cities, three times each, and counted every business they named. This page adds up all 134 markets.
Collected October 3–5, 2026 · 1,608 answers · 3,301 business names counted
- 68%of the businesses named were named by one assistant only
- 5%were named by all four assistants
- 33/100agreement when the same assistant is asked again
What the count shows
- The assistants mostly name different businesses. Of 3,301 business names, 2,250 (68%) came from one assistant only. 181 (5%) were named by all four.
- Most markets have many names and little agreement. The middle market had 23.5 businesses named and 2 named by three or more assistants. In 19 of 134 markets, none was.
- Asking again changes the answer. When the same assistant got the same question three times, its lists agreed 33 out of 100 in the middle market. Agreement was highest for hotels (48) and lowest for plumbers (22).
- The assistants answer differently. Gemini named 7.0 businesses per answer on average and Perplexity 2.6. Claude named no business in 109 of its 402 answers, Perplexity named no business in 54 of its 402 answers.
- Web search was offered on every question and did not always run. ChatGPT ran a search every time; Gemini answered 93 of 402 without a search; Claude answered 161 of 402 without a search.
- A few sites supply most of the citations. 1,403 of 1,608 answers cited at least one source. consumeraffairs.com, expertise.com, bbb.org, reddit.com and yelp.com were cited most.
How many assistants named the same business
Each business is counted once per market. A business named by one assistant is one that the other three never mentioned in any of their three answers for that market.
| Named by | Businesses | Share | Share, drawn |
|---|---|---|---|
| One assistant only | 2,250 | 68% | |
| Two | 595 | 18% | |
| Three | 275 | 8% | |
| All four | 181 | 5% |
The business named most in a market came up in 58% of that market’s answers, in the middle market. So even the most-named business was absent from 42% of them.
How much an answer changes when you ask again
For each market we compared the three lists one assistant gave to the same question. 100 means the same businesses every time and 0 means no business in common. The figure for a market is the average over its assistants; these are the middle values.
| Assistant | Agreement | Drawn |
|---|---|---|
| Gemini | 44.5 | |
| Claude | 33 | |
| Perplexity | 25.5 | |
| ChatGPT | 22.5 |
| Service | Agreement | Drawn |
|---|---|---|
| Hotels | 48 | |
| Personal injury lawyers | 36.5 | |
| Dentists | 32 | |
| AC repair | 25 | |
| Plumbers | 22 |
Across markets the figure ran from 9 to 75. One answer from one assistant is a sample, not the answer.
How each assistant answers
| Assistant | Businesses per answer | Drawn | Named none | No search run |
|---|---|---|---|---|
| Gemini | 7.0 | 0 (0%) | 93 of 402 | |
| Claude | 6.1 | 109 (27%) | 161 of 402 | |
| ChatGPT | 4.2 | 4 (1%) | 0 of 402 | |
| Perplexity | 2.6 | 54 (13%) | not reported |
The answers that named no business were not errors. Nearly all said that no single choice suits everyone, then explained how to choose or pointed to a directory. For lawyers, Perplexity often named a person and no firm, which is not counted here. Where no search ran we can see that it did not, and not why.
What the assistants cite
Counted in answers: a site cited twice in one answer counts once. In 47% of answers, at least one cited site was the website of a business named on that page.
| Source | Answers | Share | Drawn |
|---|---|---|---|
| consumeraffairs.com | 269 | 17% | |
| expertise.com | 226 | 14% | |
| bbb.org | 187 | 12% | |
| reddit.com | 186 | 12% | |
| yelp.com | 170 | 11% | |
| superlawyers.com | 142 | 9% | |
| angi.com | 129 | 8% | |
| serviceagent.ai | 127 | 8% | |
| tripadvisor.com | 127 | 8% | |
| trustanalytica.org | 106 | 7% | |
| thumbtack.com | 96 | 6% | |
| bestpickreports.com | 93 | 6% | |
| hotels.com | 92 | 6% | |
| expedia.com | 90 | 6% | |
| webmd.com | 89 | 6% |
By service
AC repair · 324 answers
- consumeraffairs.com37%
- expertise.com32%
- bbb.org31%
- reddit.com19%
- serviceagent.ai15%
- yelp.com15%
Plumbers · 324 answers
- consumeraffairs.com46%
- expertise.com32%
- angi.com29%
- bbb.org26%
- serviceagent.ai24%
- thumbtack.com22%
Dentists · 324 answers
- webmd.com27%
- zocdoc.com27%
- toprateddentist.com22%
- healthgrades.com20%
- dental.me19%
- opencare.com18%
Hotels · 324 answers
- tripadvisor.com39%
- hotels.com28%
- expedia.com28%
- marriott.com16%
- roommaster.com13%
- cntraveler.com12%
Personal injury lawyers · 312 answers
- superlawyers.com46%
- bestlawyers.com27%
- bestlawfirms.com18%
- justia.com12%
- lawleaderboard.com12%
- calbar.ca.gov8%
Share of each service’s answers that cited the site.
Is your business in these answers?
Leolens asks ChatGPT, Gemini, Perplexity and Claude the questions your customers ask, every month, and shows whether they name you and whether what they say about you is true.
One location: $799 per month. Cancel any time from the billing portal.
Every market
Named: businesses named at least once. 3+: named by three or more assistants. Agreement: how much repeat answers agreed, 0–100. Each city links to its full count.
AC repair · 27 cities
| City | Named | 3+ | Agreement |
|---|---|---|---|
| Atlanta, GA | 20 | 2 | 33 |
| Austin, TX | 25 | 3 | 30 |
| Baltimore, MD | 26 | 2 | 26 |
| Boston, MA | 22 | 1 | 34 |
| Charlotte, NC | 29 | 3 | 21 |
| Chicago, IL | 17 | 2 | 32 |
| Dallas, TX | 29 | 1 | 32 |
| Denver, CO | 30 | 1 | 16 |
| Detroit, MI | 18 | 2 | 26 |
| Houston, TX | 27 | 3 | 22 |
| Los Angeles, CA | 20 | 2 | 24 |
| Miami, FL | 25 | 1 | 22 |
| Minneapolis, MN | 19 | 3 | 45 |
| New York, NY | 22 | 0 | 41 |
| Orlando, FL | 26 | 1 | 24 |
| Philadelphia, PA | 34 | 2 | 9 |
| Phoenix, AZ | 29 | 1 | 20 |
| Portland, OR | 30 | 2 | 22 |
| Riverside, CA | 14 | 3 | 22 |
| Round Rock, TX | 20 | 2 | 38 |
| San Antonio, TX | 28 | 2 | 21 |
| San Diego, CA | 23 | 1 | 34 |
| San Francisco, CA | 23 | 1 | 28 |
| Seattle, WA | 30 | 2 | 12 |
| St. Louis, MO | 21 | 2 | 25 |
| Tampa, FL | 21 | 4 | 25 |
| Washington, DC | 23 | 0 | 26 |
Dentists · 27 cities
| City | Named | 3+ | Agreement |
|---|---|---|---|
| Atlanta, GA | 30 | 0 | 33 |
| Austin, TX | 28 | 1 | 47 |
| Baltimore, MD | 20 | 2 | 32 |
| Boston, MA | 15 | 0 | 51 |
| Charlotte, NC | 26 | 0 | 36 |
| Chicago, IL | 24 | 0 | 54 |
| Dallas, TX | 30 | 2 | 18 |
| Denver, CO | 16 | 1 | 29 |
| Detroit, MI | 21 | 2 | 28 |
| Houston, TX | 25 | 1 | 25 |
| Los Angeles, CA | 23 | 1 | 35 |
| Miami, FL | 25 | 0 | 38 |
| Minneapolis, MN | 21 | 7 | 14 |
| New York, NY | 22 | 0 | 40 |
| Orlando, FL | 22 | 0 | 28 |
| Philadelphia, PA | 23 | 0 | 34 |
| Phoenix, AZ | 32 | 1 | 18 |
| Portland, OR | 22 | 0 | 44 |
| Riverside, CA | 24 | 2 | 22 |
| Round Rock, TX | 23 | 3 | 32 |
| San Antonio, TX | 34 | 1 | 40 |
| San Diego, CA | 24 | 0 | 13 |
| San Francisco, CA | 27 | 0 | 18 |
| Seattle, WA | 27 | 1 | 40 |
| St. Louis, MO | 29 | 2 | 12 |
| Tampa, FL | 23 | 1 | 34 |
| Washington, DC | 21 | 0 | 20 |
Hotels · 27 cities
| City | Named | 3+ | Agreement |
|---|---|---|---|
| Atlanta, GA | 41 | 8 | 49 |
| Austin, TX | 25 | 11 | 54 |
| Baltimore, MD | 25 | 7 | 51 |
| Boston, MA | 32 | 8 | 56 |
| Charlotte, NC | 21 | 8 | 62 |
| Chicago, IL | 32 | 9 | 48 |
| Dallas, TX | 23 | 11 | 63 |
| Denver, CO | 29 | 8 | 48 |
| Detroit, MI | 27 | 9 | 54 |
| Houston, TX | 22 | 8 | 59 |
| Los Angeles, CA | 52 | 13 | 33 |
| Miami, FL | 35 | 10 | 42 |
| Minneapolis, MN | 32 | 8 | 48 |
| New York, NY | 57 | 14 | 24 |
| Orlando, FL | 48 | 11 | 37 |
| Philadelphia, PA | 34 | 8 | 47 |
| Phoenix, AZ | 31 | 8 | 48 |
| Portland, OR | 37 | 11 | 41 |
| Riverside, CA | 20 | 5 | 67 |
| Round Rock, TX | 27 | 9 | 43 |
| San Antonio, TX | 23 | 7 | 56 |
| San Diego, CA | 34 | 7 | 54 |
| San Francisco, CA | 36 | 9 | 44 |
| Seattle, WA | 32 | 7 | 52 |
| St. Louis, MO | 26 | 10 | 48 |
| Tampa, FL | 29 | 8 | 43 |
| Washington, DC | 35 | 12 | 52 |
Personal injury lawyers · 26 cities
| City | Named | 3+ | Agreement |
|---|---|---|---|
| Atlanta, GA | 29 | 0 | 40 |
| Austin, TX | 20 | 2 | 32 |
| Baltimore, MD | 27 | 3 | 22 |
| Boston, MA | 27 | 4 | 25 |
| Chicago, IL | 16 | 6 | 51 |
| Dallas, TX | 16 | 3 | 30 |
| Denver, CO | 29 | 2 | 16 |
| Detroit, MI | 16 | 5 | 33 |
| Houston, TX | 15 | 2 | 36 |
| Los Angeles, CA | 21 | 0 | 36 |
| Miami, FL | 29 | 1 | 44 |
| Minneapolis, MN | 15 | 3 | 30 |
| New York, NY | 14 | 3 | 42 |
| Orlando, FL | 19 | 2 | 48 |
| Philadelphia, PA | 15 | 3 | 37 |
| Phoenix, AZ | 26 | 0 | 48 |
| Portland, OR | 23 | 2 | 34 |
| Riverside, CA | 12 | 2 | 32 |
| Round Rock, TX | 17 | 3 | 37 |
| San Antonio, TX | 27 | 1 | 17 |
| San Diego, CA | 16 | 1 | 51 |
| San Francisco, CA | 12 | 0 | 75 |
| Seattle, WA | 20 | 3 | 20 |
| St. Louis, MO | 20 | 0 | 40 |
| Tampa, FL | 25 | 2 | 43 |
| Washington, DC | 12 | 4 | 44 |
Plumbers · 27 cities
| City | Named | 3+ | Agreement |
|---|---|---|---|
| Atlanta, GA | 21 | 2 | 20 |
| Austin, TX | 25 | 1 | 21 |
| Baltimore, MD | 26 | 2 | 29 |
| Boston, MA | 19 | 2 | 21 |
| Charlotte, NC | 23 | 6 | 20 |
| Chicago, IL | 18 | 3 | 32 |
| Dallas, TX | 25 | 3 | 33 |
| Denver, CO | 25 | 3 | 13 |
| Detroit, MI | 20 | 1 | 21 |
| Houston, TX | 19 | 7 | 30 |
| Los Angeles, CA | 17 | 1 | 62 |
| Miami, FL | 23 | 5 | 21 |
| Minneapolis, MN | 23 | 3 | 25 |
| New York, NY | 20 | 1 | 32 |
| Orlando, FL | 27 | 1 | 32 |
| Philadelphia, PA | 19 | 6 | 22 |
| Phoenix, AZ | 25 | 3 | 21 |
| Portland, OR | 19 | 3 | 35 |
| Riverside, CA | 31 | 1 | 18 |
| Round Rock, TX | 26 | 3 | 35 |
| San Antonio, TX | 29 | 2 | 19 |
| San Diego, CA | 21 | 3 | 22 |
| San Francisco, CA | 19 | 4 | 44 |
| Seattle, WA | 23 | 3 | 19 |
| St. Louis, MO | 19 | 3 | 44 |
| Tampa, FL | 20 | 3 | 16 |
| Washington, DC | 28 | 3 | 30 |
How this was collected
From October 3 to October 5, 2026 we sent one question per market to ChatGPT (gpt-5.6-terra), Gemini (gemini-3.8-flash), Perplexity (perplexity/sonar) and Claude (claude-sonnet-5), 3 times each, through each provider’s API with web search turned on. Every request was a new conversation with no account, history or location. The questions were:
What's the best AC repair company in [city], [state]?
Who is the best plumber in [city], [state]?
Who is the best dentist in [city], [state]?
What are the best hotels to stay at in [city], [state]?
Who is the best personal injury lawyer in [city], [state]?
A separate model listed the businesses each answer named. We checked every name against the answer’s own text, merged different spellings of one business, and left out directories and review sites named as if they were businesses. A market is published only when at least three assistants answered, at least five businesses were named, and at least one was named by two assistants or more.
What this does not show
- What you see in the apps. They use your location and history, and the apps’ models are not always the API’s.
- Whether a business is good. This is a count of mentions, not a rating or a review by Leolens, and no business paid to appear.
- Other wordings. Each service was asked one way; a different question gets a different list.
- People. Where an answer named a lawyer or a dentist and no firm or practice, nothing was counted.
- Certainty about spellings. 183 pairs of names that might be one business were kept apart because we could not tell. Counting every such pair as one business moves the share named by one assistant from 68% to 66%.
- Other places. 27 cities, most of them large US metros.
Quote these numbers
You are welcome to quote or chart anything on this page. Please link to it as the source:
Leolens, “Do AI assistants agree on which local businesses to recommend?”, October 2026. https://tryleolens.com/ai-visibility/study/2026-10/
The data behind the tables: every market (CSV) and every cited source (CSV). Questions or corrections: hello@tryleolens.com.