Who Do ChatGPT and Perplexity Recommend? We Asked 40 Times
We asked ChatGPT and Perplexity for the best plumber, HVAC company, roofer, dentist, and injury lawyer in four US cities. Here is who they named, which sites they cited, and what local owners can do about it.
When you ask ChatGPT or Perplexity "who is the best plumber in Dallas," the answer is built mostly from review and directory sites plus the businesses' own websites. In our 40 answers, 85 percent cited at least one review or directory platform, 62.5 percent cited a recommended business's own website, and the two engines almost never agreed on who was best.
We ran this because local owners keep asking us the same thing: "Where does ChatGPT get its answer, and how do I get into it?" Opinions are cheap, so we collected answers instead.
How we ran it
Twenty questions, two engines, one run each, all on October 19, 2026.
- Trades: plumber, HVAC repair company, roofing contractor, dentist, personal injury lawyer.
- Cities: San Diego, CA; Phoenix, AZ; Dallas, TX; Charlotte, NC.
- Question, word for word: "Who is the best {trade} in {city}?"
- Engines: ChatGPT with web search, and Perplexity (Sonar), both set to the United States in English. We collected the answers through DataForSEO's API, not the consumer apps.
- What we recorded: every business each answer named, every source it cited, and whether a named business's own website was among those sources.
That is 40 answers. It is a snapshot, not a census, and the caveats are at the bottom of this page.
Finding 1: Directories and review sites decide who gets considered
34 of 40 answers (85 percent) cited at least one review or directory platform. Perplexity did it in 19 of 20 answers and ChatGPT in 15 of 20.
The sites cited most often across both engines:
| Site | Answers citing it (of 40) |
|---|---|
| expertise.com | 18 |
| consumeraffairs.com | 13 |
| forbes.com | 8 |
| bestprosintown.com | 7 |
| avvo.com | 6 |
| thumbtack.com | 6 |
| yelp.com | 5 |
| bbb.org | 5 |
| attorneys.superlawyers.com | 5 |
| doctor.webmd.com | 5 |
The mix changes by trade, and that matters more than the overall list:
| Trade | Most cited sources (answers, of 8) |
|---|---|
| Plumber | ConsumerAffairs (6), Expertise.com (5), BestProsInTown (4), Yahoo Local (4) |
| HVAC repair | ConsumerAffairs (7), Expertise.com (5) |
| Roofing contractor | TheRoofingBrief (5), BBB (3), Expertise.com (3), Thumbtack (3) |
| Dentist | WebMD (5), Zocdoc (4), Healthgrades (3) |
| Personal injury lawyer | Avvo (6), Super Lawyers (5), Expertise.com (5) |
Google Maps links showed up only in ChatGPT answers (3 of 20), and no answer in our set cited Reddit.
Finding 2: Your own website still gets read
In 25 of 40 answers (62.5 percent), at least one recommended business's own website was among the cited sources. ChatGPT did it in 13 of 20 and Perplexity in 12 of 20.
So the answers did not rely on directories alone. In most of them, a recommended business's own site sat right next to the review platforms in the source list. A site that says clearly what you do, where, and for whom gives an engine something accurate to repeat.
Finding 3: The two engines barely agree
Across 20 identical questions, ChatGPT and Perplexity shared an average of 0.3 business names per question. Fourteen of the 20 questions had no business in common.
The six names both engines agreed on:
| Question | Name both engines gave |
|---|---|
| Plumber, Dallas | Workman Plumbing |
| HVAC repair, San Diego | Bob Jenson Air Conditioning, Heating & Plumbing |
| HVAC repair, Phoenix | Day & Night Air Conditioning, Heating & Plumbing |
| Roofing contractor, Phoenix | Paragon Contracting |
| Roofing contractor, Charlotte | TruROOF |
| Dentist, Phoenix | Life Smiles Dental Care |
Each engine named three to four businesses per answer on average, so roughly seven slots per question across the two engines, and almost none overlapped. Being named by one assistant tells you little about the other.
Finding 4: You do not need thousands of reviews
ChatGPT showed local business cards in 13 of its 20 answers, 44 cards in total. The median card had 130 reviews and a 4.9 rating. The range was wide, from 1 review to 2,950, and 11 of the 44 had fewer than 50 reviews.
Reviews clearly matter. ChatGPT mentioned star ratings or review counts in 15 of its 20 answers. But a modest number of strong, recent reviews was enough for a quarter of the businesses it showed.
What this means if you own a local business
These are practical readings of 40 answers, not guarantees. Nobody controls what an AI model says.
- Find out which directories your trade's answers cite. For injury lawyers, that was Avvo and Super Lawyers. For dentists, it was WebMD, Zocdoc, and Healthgrades. Make sure your profile exists on those sites, is accurate, and matches your website.
- Make your website easy to quote. State your services, your service area, and what makes you different in plain sentences near the top of each page. Answers lean on your site to describe you.
- Keep earning reviews, steadily. A median of 130 is a useful reference, not a target you must hit before anything happens.
- Check more than one assistant, more than once. The engines disagree with each other, and the same question can return different names on a different day. One screenshot is not a ranking.
If you want to know where you stand today, our free audit shows how Google and AI tools read your site, and our SEO and GEO service is the work of getting the directories, reviews, and pages right. For the basics in plain English, start with Will ChatGPT recommend your business?
What this study does not show
- One run per question per engine. AI answers vary from run to run; the same question can name different businesses tomorrow.
- One date, one country, one language. October 19, 2026; United States; English.
- API answers, not the consumer apps. Results may differ from what a signed-in user sees.
- Named is not ranked. We recorded what each answer said. That is not an endorsement of any business, by us or by the engines.
- How we counted sources. Perplexity returns about ten source URLs per answer and its text references only some of them; we counted all returned sources. Matching is by exact domain, so a subdomain such as attorneys.superlawyers.com is listed separately from superlawyers.com.
- How we read names. ChatGPT business names came from its structured results. Perplexity names were read from the answer text, leaving out directories and publishers.
If you cite these numbers, please link to this page and note the collection date, October 19, 2026.
Frequently asked questions
Which sites do ChatGPT and Perplexity cite when recommending local businesses?
In our 40 answers, review and directory sites dominated. Expertise.com was cited in 18 answers, ConsumerAffairs in 13, Forbes in 8, and BestProsInTown in 7. Trade-specific sites mattered too: Avvo and Super Lawyers for injury lawyers; WebMD, Zocdoc, and Healthgrades for dentists; and TheRoofingBrief, BBB, and Thumbtack for roofers.
Do ChatGPT and Perplexity recommend the same businesses?
Rarely. Across 20 identical questions, the two engines shared an average of 0.3 business names per question. Fourteen of the 20 questions had no business in common at all.
Does my own website matter for AI recommendations?
Yes. In 25 of 40 answers (62.5 percent), at least one recommended business's own website was among the cited sources. Directories get you considered; your own site is often what the answer leans on to describe you.
How many reviews does a business need to be recommended by ChatGPT?
There is no fixed number in our data. The businesses ChatGPT showed as local cards had a median of 130 reviews and a median 4.9 rating, but 11 of the 44 had fewer than 50 reviews. Reviews help; they are not a hard cutoff.