Key findings
Texas markets (one trade in one city): 23 trades, 23 cities
ChatGPT answers: 3 questions per market, each asked up to 2 times
markets where ChatGPT cited Yelp / Facebook / any Google domain
of markets cited Angi; one ranking site (ContractorsRanked.com) was cited in 71.8%
of citations to each market’s 10 most-cited sites went to third-party “best of” and ranking sites
average share of answers naming a market’s #1 business
businesses per market named in at least 5 of 6 answers; 21.9% of markets had none
of named businesses appeared in just 1 of the 6 answers (a minimum: rarer names weren’t stored)
Consistency figures use the 356 markets with 6 answers; citation figures use all 390. Please read the caveats before quoting a number.
1. Which sites does ChatGPT cite?
Share of the 390 markets where ChatGPT’s answers cited each site.
742 different domains were cited in total; 467 of them in only one market. Site descriptions come from each site’s own pages. A citation shows ChatGPT read a page, not that the page decided who was named.
2. What kinds of sites?
Share of citations to each market’s 10 most-cited sites, by type of site (sorted by hand, approximate). Rarer sites weren’t stored, so local sites are under-counted.
ChatGPT leaned on sites that already list and compare local businesses, rather than on a business’s own website or the big review sites. Other studies disagree on Yelp, and our API-based setup may behave differently from the ChatGPT app.
3. How steady are ChatGPT’s picks?
ChatGPT names at least 4.8 businesses per answer (a minimum, because only each market’s 12 most-named businesses were stored; our full saved scans suggest about 6.2). It usually has one favourite per market, and a changing cast behind it.
Share of the 356 markets with 6 answers. On average only 2.6 businesses per market were named in a majority of answers, and at least 35% of named businesses showed up in just one answer (about two thirds in our full saved scans). That’s why one ChatGPT question isn’t a reliable test.
4. Differences by trade
Average share of answers that name the market’s #1 business (17 to 23 Texas markets per trade).
Specialist trades (painting, pest control, garage doors) were the most settled; general contractors the least, with nearly half of markets having no business named in 5 of 6 answers. Roofing, HVAC and plumbing sit in the middle: about a quarter of their markets had no reliably named business. This is a pattern in our data, not a tested explanation.
5. Differences by city size
The 15 trades covered in all 23 cities (321 six-answer markets). Populations from the 2020 Census.
| City size | Cities | Avg. share naming #1 | #1 in all 6 | No majority pick | Cited Angi | Thumbtack | Birdeye pages |
|---|---|---|---|---|---|---|---|
| Large (500k+) | 5 | 84.3% | 43.3% | 10.4% | 70.1% | 19.4% | 31.3% |
| Mid-size (150k–500k) | 8 | 87.2% | 50% | 5.8% | 86.5% | 47.1% | 58.7% |
| Smaller (under 150k) | 10 | 88.1% | 50.7% | 2.7% | 82% | 40.7% | 60% |
Bigger cities were slightly less settled, and their answers cited marketplaces less. They also cited city and government sites less: 16.4% of large-city markets under every definition, against 21–37% of mid-size and 25–29% of smaller-city markets, depending on which city-run sites count. The correlation between population and top-business share was moderate (−0.41) across only 23 cities, so treat it as suggestive.
What this means for local businesses
From the data, not a guarantee.
- Where you appear online matters more than your own page alone. ChatGPT’s citations came mainly from sites that list and compare local pros. Being listed, accurate and well reviewed there is the closest thing to a lever in this data. It doesn’t prove a listing decides the pick.
- Claimable profiles are in the mix. Some of the most-cited sites accept free profile claims, and some show ratings pulled from Google, Facebook and Yelp, so those reviews may reach ChatGPT indirectly (an inference we did not test).
- Google and Yelp reviews still matter to customers, and other AI engines cite those sites heavily.
- One answer isn’t enough. ChatGPT’s #1 is stable, but the rest of the list moves. Test any question several times.
- Your trade matters. General contractors, gutter and pool companies face a more shuffled field than painters or pest control.
Methodology and caveats
Data. LocalPick AI’s public “Who AI recommends” market snapshots: the 390 market rows that existed on 10 Nov 2026, read from our database (no new scans were run for the study). A market is one trade in one Texas city: 23 trades across Houston, San Antonio, Dallas, Austin, Fort Worth, Arlington, Plano, Irving, Garland, Frisco, Grand Prairie, McKinney, Mesquite, Denton, Carrollton, Pearland, Richardson, Round Rock, The Woodlands, Sugar Land, New Braunfels, Georgetown and Katy.
How answers were collected. Three customer-style questions per market (“Who is the best [trade] in [city]?”, “Can you recommend a good [trade] in [city]? Give me a short list.”, “What are the top-rated options for a [trade] in [city]?”) were sent to ChatGPT through OpenAI’s API (model gpt-5-mini) with web search on and the location set to the city, each twice: 6 answers per market (3 for 34 markets). Collected 10 Nov 2026, 04:48–07:04 UTC. Business names were extracted by a language model and merged when they referred to the same business; cited links were reduced to domains.
What was stored. Per market, only the top 12 businesses and top 10 cited domains, with counts. So rare sites are invisible, and consistency is measured from per-business totals rather than by comparing individual answers.
- One AI engine (ChatGPT via the API, gpt-5-mini). Gemini, Perplexity, Google’s AI and the ChatGPT app were not tested.
- One two-hour window on one day. Answers and sources change.
- Texas only, mostly the four big metro areas. Don’t generalise to other states.
- Words like “best” and “top-rated” may pull in ranking sites; other questions could cite other sources.
- Site categories are our judgement and approximate.
- A citation is not a cause.
- Hair salon, med spa and auto repair had too few markets to compare.
- LocalPick AI sells AI-visibility services, so we have a commercial interest in this topic. The numbers come straight from the stored snapshots.
How to cite: LocalPick AI, “Who does ChatGPT recommend for local home services in Texas? November 2026 snapshot” (390 markets, 2,238 ChatGPT answers, collected 10 Nov 2026), https://localpickai.com/study.
Free: the AI Visibility Checklist for Texas Contractors
12 things ChatGPT checks before it recommends a roofer, plumber or HVAC company, each with a 2-minute check you can do yourself.
Where does your business stand?
Free 60-second scan: whether ChatGPT names you, who it names instead, and which of these sites to get listed on. Or browse who AI recommends in your city.
Scan my business free