We asked ChatGPT to recommend a plumber, a dentist, a lawyer and 17 other local businesses across Orange County and the Inland Empire, then looked at every business it listed. Its written answer is the top of its list, read in order. Reviews help but do not decide the order, being in the city matters, and almost everything it cited was a directory, not a business website.
Everyone selling AI SEO has a theory about how ChatGPT picks a local business. We wanted numbers. On September 27, 2026 we asked ChatGPT, with web search on, to recommend 20 local businesses across Orange County and the Inland Empire: a plumber in Corona, a dentist in Costa Mesa, a personal injury lawyer in Irvine and 17 more. Then we looked at every business it listed, 228 of them, and every source it cited.
Some of what we found matches the usual advice. Some of it does not, and one finding changes where the work should go.
The findings in one place
- The written answer is the top of the list. ChatGPT's first written pick matched the first business on its list in 18 of 19 answers, and every written shortlist followed the list's order.
- Reviews help, but they do not set the order. The top three on each list had a median of 112 reviews against 72 for the rest, yet the most-reviewed business was first only 2 times out of 19.
- Being in the city matters. 80.7% of listed businesses were located in the city we asked about. Big, well-reviewed firms from the next town over sat low on the list.
- It cites directories, not your website. 45 of 53 cited sources were directories or review lists. Two were business websites.
- The bar is lower than you think in some trades. The median personal injury lawyer on the list had 13 reviews. The median plumber had 219.5.
How we ran it
Every question used the same wording, “Can you recommend a [trade] in [city], CA?”, so the answers can be compared. Ten questions were for Orange County cities and ten for Inland Empire cities, across 13 trades: personal injury lawyers, dentists, a med spa, HVAC, roofing, an immigration attorney, tax preparers, auto repair, plumbers, real estate, solar, electricians and landscaping. We asked each question once, on the same day, from a United States location, with web search on.
For 19 of the 20 questions ChatGPT showed a list of local businesses above its written answer, each with a rating and a review count. The lists held 7 to 16 businesses, 12 in the middle. The Tustin real estate question came back as text only. We recorded each business's position, rating, review count and whether it had a website, then which businesses the written answer actually recommended, and which sources it cited. We are reporting totals, not naming the businesses.
Finding 1: the written answer is the top of the list, read in order
This is the finding that matters most. When ChatGPT wrote its recommendation, it did not weigh the list and pick favorites. It started at the top and worked down. Its first written pick was the first business on the list in 18 of 19 answers. In all 19, the businesses it recommended from the list were the top ones, in the same order, with no position skipped. The deepest it ever went was the eighth.
So the question “how do I get ChatGPT to recommend me” is, for local searches, mostly the question “how do I get to the top of the list it shows first”. The written answer adds color, such as license numbers and a line about each business, but it rarely changes who gets named.
There is one exception worth knowing. In 4 of 19 answers ChatGPT also named businesses that were not on its own list, 6 names in total, and all 4 were Inland Empire questions. In Ontario its first two written picks were both missing from the list. Those names came from the directories it cited, which is the second route in, covered below.
Finding 2: reviews help, but they do not set the order
Across all 228 listed businesses, the median had 78 reviews, the middle half had between 30 and 223, and the range ran from 1 to 1,908. More reviews did help. The top three on each list had a median of 112 reviews, against 72 for everyone below them, and the businesses the written answer recommended had a median of 111 against 67 for the ones it passed over.
But the most-reviewed business was first on the list in only 2 of 19 answers. It made the written shortlist in 11 of 19, a little more than half. Rating did not separate them either: the median was 4.8 stars in the top three and 4.8 below it. Reviews are one input, not the ranking.
Finding 3: being in the city you were asked about matters
80.7% of listed businesses were located in the city in the question. The clearest example was the Mission Viejo plumber question. The three businesses with the most reviews on that list, with 1,908, 1,144 and 768, sat at positions 9, 10 and 12. All three were outside Mission Viejo, none was recommended, and every business above them had fewer reviews.
When a city has few businesses in a trade, ChatGPT reaches further out. For the Chino Hills landscaper question, only 1 of the 9 businesses listed was located in Chino Hills. That is an opening: in a thin category, a business that clearly says it serves that city has little competition for the top.
Finding 4: it cites directories, not your website
The 20 answers cited 53 sources. 45 of them, 84.9%, were directories or review lists. Another 6 were State Bar or county bar association pages, all on the lawyer questions. Only 2 were a business's own website, and they came from a single answer. 17 of the 20 answers cited at least one directory. One answer cited nothing at all.
| Most-cited source | Citations | Answers that cited it |
|---|---|---|
| Birdeye | 9 | 5 |
| Expertise.com | 7 | 6 |
| Angi | 5 | 4 |
| State Bar and county bar pages | 6 | Lawyer questions only |
For local services, the words in the answer come from the directories, and the directories are also where ChatGPT found the names it added from outside its own list. A business that is missing from the directories that matter in its trade, or listed there with a different name, phone number or category, gives the assistant nothing to quote.
This is different for businesses that sell to other businesses. When we asked ChatGPT who builds custom software near Chino Hills, it named us first and cited four of our own pages. B2B questions have fewer directories to lean on, so a clear website carries more of the answer.
Finding 5: the bar is lower than you think in some trades
About one in ten listed businesses, 10.1%, had fewer than 10 reviews, and the written answers still recommended some businesses with 3, 5, 8 and 9 reviews. Ratings were not a hard filter either: 13.2% of listed businesses were below 4.5 stars, and a few recommended ones sat at 3.2 to 3.8. Only 7% had no website, and 4 of those were still recommended, named without a link.
| Group | Median reviews of listed businesses |
|---|---|
| Plumbers | 219.5 |
| All 228 businesses | 78 |
| Tax preparers | 14.5 |
| Personal injury lawyers | 13 |
The trade decides the bar. A tax preparer or a personal injury firm with a few dozen genuine reviews is already ahead of most of its list. A plumber needs hundreds.
Orange County and the Inland Empire are different markets
| 114 listed businesses each | Orange County | Inland Empire |
|---|---|---|
| Median reviews | 112.5 | 65 |
| With 100 or more reviews | 62 | 39 |
| Located in the city asked about | 100 | 84 |
| No website | 6 | 10 |
Orange County lists are more crowded with well-reviewed businesses. Inland Empire lists are thinner, reach further across city lines, and are the only ones where ChatGPT added names from outside its own list. For an Inland Empire business that means a lower bar and more room to move.
It checks you against outside records
The written answers leaned on records outside the listing to vouch for a pick: contractor license numbers, “BBB currently lists them as A+”, AAA Approved, State Bar records. And they warned the reader not to rely on ratings alone: “I wouldn't pick one solely from online ratings”, and “get 2–3 written estimates rather than choosing solely on reviews”. A license, bar number or BBB profile that matches your listing is part of what gets you described with confidence.
What this means for a local business
- Aim for the top of the list, because the answer reads it in order. The businesses named in writing were the top of the list 19 times out of 19.
- Be clearly in the city you want to be recommended in. A real address or a stated service area there, the same on your listing and your site. Out-of-town businesses with far more reviews ranked below local ones.
- Measure reviews against your trade, not a national number. Look at who is on your list today and how many reviews they have. For some trades a few dozen puts you in the top three.
- Get onto the directories your trade is cited from. Birdeye, Expertise.com and Angi came up most here; for lawyers, the State Bar. Same name, address, phone and category everywhere.
- Keep your license, bar or BBB record consistent with your listing. The answers used those records to vouch for the businesses they named.
- Keep a website, and make it say plainly who, what and where. Most listed businesses had one, the listings link to it, and for B2B questions it is often the source itself.
- Check the answer every month. Ask ChatGPT the question your customers ask, with your city in it, and write down who is named. That is the only way to know whether any of the above is working.
What this study cannot tell you
It is 20 questions, asked once, on one day, from one location setting. ChatGPT's answers change from day to day and from person to person, so treat these as patterns, not rules. The listings also appear to come from more than one data source: review counts on ChatGPT's list sometimes differed from the counts on review sites cited in the same answer, by a lot in a couple of cases. And we only measured what was shown, not why. The order clearly weighs location and reviews, but it weighs other things we could not see.
Where we land
For local businesses, being recommended by ChatGPT is mostly a listings problem, not a content problem: be at the top of the list it shows first, be clearly in the city, and be on the directories it trusts, with every record agreeing. The website still matters, especially for B2B, but it is rarely the page it quotes. That is the work our AI SEO service for Orange County and the Inland Empire does, starting with an audit that asks the assistants the questions your customers ask and records who gets named instead of you. The broader mechanics are in how small businesses get mentioned by ChatGPT.
Frequently asked questions
How does ChatGPT choose which local businesses to recommend?
In our 20-search study, ChatGPT first built a list of local businesses with ratings and review counts, then wrote its recommendation from the top of that list in order. Its first written pick matched the first business on the list in 18 of 19 answers. The order leaned on being located in the city that was asked about and on review volume, but the most-reviewed business was first only twice, so reviews alone did not set the order.
How many reviews do I need to be recommended by ChatGPT?
Fewer than most owners expect, and it depends on the trade. Across 228 listed businesses the median was 78 reviews, but the median for personal injury lawyers was 13 and for tax preparers 14.5, against 219.5 for plumbers. About one in ten listed businesses had fewer than 10 reviews, and some businesses with 3 to 9 reviews were still in the written shortlist. Compare yourself with the businesses in your own trade and city, not with a national number.
Does ChatGPT cite my website when it recommends local businesses?
Rarely, for local services. Of 53 sources cited across the 20 answers, 45 were directories or review lists such as Birdeye, Expertise and Angi, 6 were State Bar or county bar pages, and 2 were business websites. For local questions the listings and directories carry the answer, which is why they need to agree with each other and with your site.
Do I need a website to be recommended by ChatGPT?
Not strictly. 16 of the 228 listed businesses had no website, and 4 of those were still shortlisted, named without a link. But 93% of listed businesses had one, the listings link to it, and a site is where your name, address, services and license details are confirmed, so going without one is a handicap rather than a plan.
Is Orange County harder than the Inland Empire?
By review volume, yes. The median listed business in Orange County had 112.5 reviews, against 65 in the Inland Empire, and 62 of 114 Orange County listings had 100 or more reviews, against 39 of 114 in the Inland Empire. Inland Empire answers were also the only ones that named businesses missing from ChatGPT's own list.