239 ChatGPT Answers for Local Trades: The Businesses It Showed Had a Median of 132 Reviews
ChatGPT named a business in all 239 answers, and the ones on its cards had a median of 132 reviews
ChatGPT named a specific business in all 239 first-run answers we collected on October 22, 2026, and in all 240 answers in a repeat run. Two in three first-run answers (158) showed a local business card with a star rating, review count, phone number, and website. The businesses on those cards showed a median of 132 reviews, so half had more and half had fewer.
Cards often show fewer reviews than Google does, so we also counted by Google's own numbers: the Google Maps results ChatGPT put on a card had a median of 236 reviews, and the ones it left off had 42.
We asked for a house cleaner, a lawn care company, a junk removal company, and a pressure washing company in 30 US cities, two ways each, then asked every question again about an hour and twenty minutes later. Unless we say otherwise, the numbers are from the first run.
| What we measured | Result |
|---|---|
| Answers with a local business card | 158 of 239 (66.1%) |
| Median reviews on the cards | 132 |
| Cards with 50 or more reviews | 76.7% |
| Cards for a business in the Google Maps top 3 | 15.3% |
| Maps results with 16 or fewer reviews shown on a card | 0.6% |
| Maps results with more than 192 reviews shown on a card | 32.9% |
| Answers that cited a Google page | 58.6% |
| Answers that cited a site like Angi or Yelp | 84.9% |
| Related searches that included "reviews" | 81.5% of 297 |
The per-answer data is in two CSV files with no business names, one row per answer: the first run and the repeat run. They hold counts and yes-or-no flags, such as whether a card showed. They give each answer's median review count, not every business's count, so the 132 across all cards cannot be rebuilt from them.
How we ran the study
We wrote 240 questions: four services, 30 cities, and two wordings. The services were "house cleaning service", "lawn care company", "junk removal company", and "pressure washing company". The first wording was "best [service] in [city]". The second was "who should I hire for [service] near [city]? I want someone reliable and fairly priced".
The 30 cities run from New York and Los Angeles down to Boise, Spokane, and Fargo: 10 large cities (about a million people or more), 11 mid-size, and 9 small (under about 300,000). We sent every question to ChatGPT with web search forced on, through DataForSEO's ChatGPT scraper, which DataForSEO describes as providing results from ChatGPT searches. The scraper labeled every answer gpt-6. One question timed out in the first run and is left out of its numbers, so the first run has 239 answers.
For the Google comparison we pulled the top 20 Google Maps results for each service and city within minutes of the first run. We searched the plain service name with the city as the location, on desktop. This is not the same query we gave ChatGPT.
Google Maps mixed six paid ads into the results, and we left them out of every Maps figure. We counted a business ChatGPT named as "in Maps" when its phone number or website matched a Maps result. A tracking number or a different website on the profile can hide a match, so the overlap numbers may undercount. Medians count only listings that show a review count.
What this study cannot tell you. It is two runs on the same day, from a scraper with no customer account behind it, for four trades in English in 30 US cities. It has no true small towns and says nothing about change over days or weeks. We also cannot prove that anything here causes a business to be named.
About two in three answers showed a local business card
In 158 of the 239 first-run answers (66.1%), ChatGPT showed a local business card listing about four businesses, nearly always with the rating, review count, address, phone number, and website for each. The other 81 answers named businesses in the text, with no card. A table appeared in 33.9% of answers, and a table with dollar amounts in 26.4%.
| Service | Showed a card | Median reviews on cards |
|---|---|---|
| House cleaning | 56.7% | 132 |
| Lawn care | 73.3% | 110 |
| Junk removal | 64.4% | 213.5 |
| Pressure washing | 70.0% | 129 |
Each service had 59 or 60 answers. House cleaning showed a card in 70.0% of repeat answers, against 56.7% in the first run, so treat gaps between trades as small samples. The order by median reviews held in the repeat.
| City size | Showed a card | Median reviews on cards |
|---|---|---|
| Large (about a million people or more) | 60.0% | 164 |
| Mid-size | 65.5% | 137 |
| Small (under about 300,000) | 73.6% | 120 |
The three sizes had 80, 87, and 72 answers. Town size made no steady difference to how often a card showed: small cities led in the first run, and mid-size cities led in the repeat. The way we asked mattered more. "Best [service] in [city]" produced a card in 78.3% of first-run answers (82.5% in the repeat). "Who should I hire ... I want someone reliable and fairly priced" produced a card in 53.8% (63.3%), and more tables with dollar amounts (39.5% of answers against 13.3% in the first run).
Most businesses on the cards show 50 or more reviews
The first-run cards listed 613 businesses, about 520 different ones, because the same business often showed up for both wordings. The median was 132 reviews counting every listing and 126 counting each business once. The 610 listings that showed a review count spread like this:
| Reviews on the card | Share of listings |
|---|---|
| Under 10 | 7.7% |
| 10 to 24 | 6.4% |
| 25 to 49 | 9.2% |
| 50 to 99 | 14.9% |
| 100 to 249 | 31.3% |
| 250 to 499 | 16.7% |
| 500 or more | 13.8% |
So 76.7% of the listings showed 50 reviews or more, and 61.8% showed 100 or more. Ratings on the cards were high: 88.0% were 4.8 or higher, and 5.7% (35 cards) were under 4.5.
The count on a card is not always Google's count. DataForSEO describes it as the "total number of reviews submitted for the local business", without naming a site.
We matched 351 first-run cards to a Google Maps listing by phone number. On 69% the card's count was within 5% of Google's. On 107 (30%) the card showed more than 5% fewer reviews than Google, and on 79 (23%) fewer than half as many. Across the 351 the median was 131 on the cards and 252 on Google. Of the 35 cards rated under 4.5, 30 matched a Maps listing, and Google rates all 30 at 4.5 or higher. So the counts and ratings above are the ones the cards show, and they may understate Google's.
A low count on a card is often not a low count on Google. 86 cards (14.1%) showed fewer than 25 reviews. We could match 58 of the 86 to a Google listing, and Google lists 25 or more for 54 of those 58. By Google's own counts, only 6 of the 306 Maps results ChatGPT showed (2.0%) had fewer than 25 reviews.
National brands (the same website in three or more cities, such as 1-800-Got-Junk, Junk King, MaidPro, and Molly Maid) were 10.6% of the listings, so most of the rest are probably local companies, though a chain seen in only one or two cities is counted there. Only 3.3% of the cards had no website.
ChatGPT's picks are not the top of Google Maps, but they have more reviews
We matched the 613 first-run listings against the top 20 Google Maps results for the same service and city. Only 15.3% of the businesses on ChatGPT's cards were in the Maps top 3, 38.3% were in the top 10, and 58.6% were in the top 20.
Now the other direction: of the Maps results, how many did ChatGPT show? We used the 109 service-and-city searches where either first-run answer showed at least one card, and only Maps results that list a review count. That is 2,077 results, split by review count into four groups of similar size: 543, 497, 518, and 519. We counted a result as shown when a card had the same phone number or website. ChatGPT showed 306 of them (14.7%).
| Google Maps result | Share ChatGPT showed on a card |
|---|---|
| 16 reviews or fewer | 0.6% |
| 17 to 57 reviews | 4.6% |
| 58 to 192 reviews | 21.0% |
| More than 192 reviews | 32.9% |
| Maps position 1 to 3 | 25.4% |
| Maps position 4 to 10 | 16.2% |
| Maps position 11 to 20 | 10.4% |
Position and reviews move together, so we compared mixed cases. A business in positions 11 to 20 with more than 192 reviews was shown 23.2% of the time (49 of 211), and one in the top 3 with 57 reviews or fewer 5.1% (6 of 117). In this snapshot, review count separated the shown businesses from the skipped ones more than map position did. The repeat, counted the same way on its 113 searches with a card, showed 34.7% of the results with more than 192 reviews and 0.7% of those with 16 or fewer.
By Google's own counts, the Maps results ChatGPT showed had a median of 236 reviews, against 42 for the ones it left off its cards and 106 for the Maps top 3 across all 120 searches. Another 103 results show no reviews and are left out of the table.
Listings with a website were shown 16.5% of the time (302 of 1,829) and listings with none 1.6% (4 of 248), but that mostly repeats the review gap. Listings with no website have a median of 5 reviews, against 73 with one, and 182 of the 248 are in the 16-or-fewer group, where almost nothing was shown either way.
This is a pattern in one day's answers, not proof that reviews cause a recommendation. Businesses with many reviews may also have longer histories and listings on the directories ChatGPT reads.
Angi turns up in almost every ChatGPT search, and Google is cited most
The scraper lists the pages ChatGPT retrieved in its searches and, separately, the pages its answer cited, which DataForSEO describes as the sources the model "actually cited or relied on in its final answer". We recorded both.
| Source | Retrieved in its searches | Cited in the answer |
|---|---|---|
| Angi | 94.6% of answers | 44.4% |
| HomeAdvisor | 74.5% | 14.6% |
| Thumbtack | 60.7% | 14.6% |
| Expertise.com | 48.1% | 35.6% |
| Birdeye | 45.6% | 18.8% |
| BBB | 24.3% | 6.3% |
| 12.1% | 2.1% | |
| Yelp | 0% | 22.2% |
| A Google page (Maps listing or search page) | 1.7% | 58.6% |
At least one Angi, HomeAdvisor, or Thumbtack page was among the retrieved pages in 99.6% of answers, and at least one of seven sites (those three plus Expertise.com, Yelp, Birdeye, and the BBB) was cited in 84.9%. The business's own website was among the pages ChatGPT retrieved or cited in 41.8% of answers that showed a card.
The most-cited source was Google itself: 113 answers (47.3%) cited a Google Maps listing and 39 (16.3%) a Google search page, and 12 cited both, so 58.6% cited a page on google.com.
Yelp never appeared among the retrieved pages in either run, yet it was cited in about one answer in five (22.2% and 22.1%). One guess is that Yelp's content reaches ChatGPT another way: Yelp's second-quarter 2026 shareholder letter, dated August 6, 2026, says its ratings and reviews "recently began powering ChatGPT's local experience in relevant categories." Treat that as a guess, not a finding.
Sam is the AI office manager that never sleeps.
He books jobs, sends invoices, and answers your customers, all by text. He checks with you before anything goes out.
Free plan, no credit card. Sam answers new customers on Pro. Or text (833) 365-7267.
Hi, I'm Sam. Text me what you need done and I'll take care of it.
How do I get started?
ChatGPT's related searches ask about reviews, prices, and the year
DataForSEO lists at least one related search for each of the 239 first-run answers, 297 in all. It calls them fan-out queries and describes them as related searches derived from the main question. One reads "best affordable reliable house cleaning services Chicago Illinois reviews pricing 2026".
| What the search included | Share of the 297 searches |
|---|---|
| The word "reviews" | 81.5% |
| Price, pricing, or cost | 50.8% |
| The year 2026 | 44.1% |
| Reliable, trusted, insured, or licensed | 36.0% |
Reviews came up most often, then prices, then the current year.
One lab test points the same way on price. In an arXiv preprint from May 2026, the authors ran 252,000 trials across six AI models, each choosing between two versions of a product-review article with made-up brand names that differed in one detail. The version that listed a price was cited first more often, and a 2026 date beat a 2019 date. A 2026 date against no date, and fewer reviews, did not show a consistent effect. It did not test local business pages, so read it as a reason not to hide your prices behind "call for a quote", not as proof.
A second run gave similar percentages but different names
ChatGPT rarely gives the same list twice, so we asked all 240 questions again about an hour and twenty minutes after the first run ended, with the same wording and settings. The repeat file has the same columns as the first-run file. In both files every row names a business, though the scraper's own business tag caught only 237 of the 239 first-run answers and 237 of the 240 repeat answers. The rest name businesses in the text.
The percentages held. Every answer named a business in both runs, and none of the percentages in the table moved by more than 7 points. Splits by trade and city size moved more (house cleaning went from 56.7% to 70.0% with a card), so read those as rough. Both runs use the same Maps pull.
| What we measured | First run | Repeat |
|---|---|---|
| Answers that showed a local business card | 66.1% (158 of 239) | 72.9% (175 of 240) |
| Median reviews on the cards | 132 | 141 |
| Answers that cited a Google page | 58.6% (140 of 239) | 57.9% (139 of 240) |
| Businesses on cards found in the Maps top 20 | 58.6% (359 of 613) | 55.2% (374 of 678) |
| Businesses on cards found in the Maps top 3 | 15.3% (94 of 613) | 14.0% (95 of 678) |
The lists changed. Of the 119 questions that showed cards both times:
- Only 15 gave the same businesses.
- On average, about 65% of the first run's businesses came back.
- The first card was the same in 66.
- 3 shared no business at all.
For 94 of the 239 questions, a card showed in one run and not the other. Two cards count as the same business when they share a phone number, else a website, else a name.
Asking in two ways changed the list too. Where both wordings showed cards, the two lists shared, on average, 37% of the businesses on either list in the first run (49 pairs) and 38% in the repeat (62 pairs).
So the percentages are steady, but who gets named changes. One answer is only one sample, so ask more than once.
What you can do with this
Here is what we would do, based on what the data showed:
- Ask every customer for a review. Review count was the biggest gap we found, in both runs: ChatGPT showed 0.6% of the Maps results with 16 or fewer reviews and 32.9% of those with more than 192. That does not prove reviews cause it. Ask everyone and offer nothing in return: Google's content policy help page, opened October 22, 2026, says merchants may not offer "payment, discounts, free goods and/or services" for a review, and may not "selectively solicit positive reviews from customers". See how to get your Google review link.
- Put your website and phone number on your Google profile. Almost every Maps listing ChatGPT showed had a website (302 of 306), though so do 88% of all results, so we cannot say the website itself matters. Our Google Business Profile setup guide walks through it step by step.
- Put your prices on your site, and keep the year current. Half of ChatGPT's related searches included price, pricing, or cost, and 44.1% included 2026. A page called "House cleaning prices in Columbus, 2026" that lists a starting price for each size of home matches the question. Our pricing guides for house cleaning, lawn care, junk removal, and pressure washing show how to work out the number.
- Check the sites ChatGPT cites. After Google, the most-cited were Angi, Expertise.com, Yelp, and Birdeye. Look yourself up on each, and where the site lets you, correct a wrong name, phone number, or website. If a card shows fewer reviews than your Google page has, the number may come from another site.
- Do the technical add-ons last. Schema markup is extra code that labels a page for machines, and llms.txt is a file that lists your pages for AI tools. An Ahrefs study of 1,885 pages that added schema found "no major uplift in citations on any platform", but every page in it already had 100 or more AI Overview citations, so it cannot say what schema does for a small site, and Ahrefs notes it has other uses, such as rich results. In another Ahrefs study of 137,210 domains, 97% of the roughly 38,000 with an llms.txt file saw no requests for it in May 2026.
- Try it for your own town. Ask ChatGPT the two questions we asked, with your service and your town. Write down who is named and how many reviews they have, and compare with your own. Ask a few times, because the names change.
Where ServicePro fits
ServicePro is home services software. It starts with getting you found. Nothing above needs it, and no tool, ServicePro included, can make ChatGPT name you. The Free plan builds a website from your Google Business Profile, with a booking page, unlimited leads with no per-lead fees, and a review widget. Pro is $99 a month, no contract. Pro widens the monthly rank scan to Google Maps and desktop searches, adds 25 town pages and AI reply drafts in the app, and each month asks ChatGPT and Google AI Overviews about your trade and town to show whether they name your business, one answer per question.
Sam is ServicePro's AI office manager. You text him, and he books jobs, sends invoices, and updates your website. On Pro, he also posts your work to Google, Facebook, and Instagram. He works by text and in the app, not by phone. The Sam page has how to get him. To see where you stand on Google first, run the free rank scan.
How this post was made: drafted with AI from the sources linked on this page. Figures and links were checked on October 22, 2026.
Frequently Asked Questions
Does ChatGPT recommend local businesses?
Yes. ChatGPT named at least one specific business in all 239 first-run answers, and 158 showed a card with a star rating, review count, phone number, and website. Which businesses it names changes between runs: when we asked all 240 questions again, only 15 of the 119 questions that showed cards both times gave the same businesses.
How does ChatGPT find local businesses?
We can see what it looked at, not how it ranks businesses. With web search on, the scraper listed 1.2 related searches per answer on average, and 81.5% of them included the word reviews. Angi, HomeAdvisor, or Thumbtack pages were among the pages it retrieved in 99.6% of answers, and it cited a Google page in 58.6%.
Does ChatGPT pick local businesses from Google Maps?
Not directly. Only 15.3% of the businesses on its cards were in Google's top 3 on Maps for the same service and city, though 58.6% were in the top 20, and the repeat run was close (14.0% and 55.2%). It showed 32.9% of the Maps results with more than 192 reviews and 0.6% of those with 16 or fewer.
How many Google reviews do I need to show up in ChatGPT?
There is no fixed number in our data. By Google's own counts, 92.5% of the Maps results ChatGPT showed had 50 or more reviews, and only 2.0% had fewer than 25. It showed 32.9% of the Maps results with more than 192 reviews and 0.6% of those with 16 or fewer. That is a pattern in one day's answers, not proof that reviews cause it.
Does paying for Angi or Thumbtack get me into ChatGPT?
We cannot tell from this data, because we did not check which businesses pay a lead platform. ChatGPT retrieved an Angi, HomeAdvisor, or Thumbtack page in 99.6% of answers, and we did not test whether a paid listing changes who it names. ServicePro is an alternative to lead platforms, so weigh our view accordingly. We would fix your Google profile and ask for reviews first, because most of the businesses ChatGPT showed had a website and 50 or more reviews.
Is this what my customers see when they ask ChatGPT?
Not exactly. We used DataForSEO's ChatGPT scraper, which provides results from ChatGPT searches, with web search forced on, a US location, and English. Answers can differ for a customer's own account, location, or chat history. We did not test that. Use this as a picture of what ChatGPT tends to look at, then check your own town.
Will ChatGPT give me the same answer if I ask again?
Not the same list. We asked all 240 questions twice, about an hour and twenty minutes apart. The percentages held: 66.1% and 72.9% of answers showed a local card. But where both runs showed cards, about 65% of the first run's businesses came back, and across all 158 first-run card questions 49% did. 94 of 239 questions showed a card in one run and not the other.

