Original research · 1 September 2026

We asked ChatGPT for a local tradesperson in 40 US cities

200 questions. 5 trades. 1,097 businesses named. The most useful thing we found is that the answer is almost completely different in every town — which means this is a winnable game for a small local business.

200
questions asked
5.5
businesses named per answer
1,076
different companies named
9
names that appeared in more than one city

What we did

We asked ChatGPT the question a homeowner actually asks — "Who is the best plumber in Nashville, TN? Name specific businesses." — once for each of 5 trades in each of 40 US cities. 200 questions in total. Then we took every business name it gave back and checked it against our own database of 312,000+ US local businesses to see whether the company was one we hold in that city.

Finding 1 — it names about 5.5 businesses, and it commits

Across 200 answers, ChatGPT named 1,097 businesses — an average of 5.5 per question. It does not hedge and it does not hand back a directory. It gives a short list, the way a neighbour would.

That is the whole problem for a local business. A page of Google results has ten links and an advert block. An AI answer has 5 names. There is no page two.

Finding 2 — the list changes almost completely from city to city

We expected national chains to dominate. They do not. Of 1,097 mentions, 1,076 were distinct companies. Only 9 names appeared in more than one city at all.

The handful that did repeat:

BusinessCities
Chop Chop Landscaping7
Roto Rooter Plumbing Water Cleanup4
Bates Electric2
Mister Sparky2
Abacus Plumbing Air Conditioning Electrical2
Priority Roofing2
Cody Sons Plumbing Heating Air2
Cozy D Heating A C Llc2
Vh1 Roofing2

A note on the top entry. We checked the name that repeated most, in every city it appeared in. It is not a landscaping company. It is a lead-generation network running one city-specific website per market — a separate domain for Phoenix, Memphis, Sacramento, Philadelphia and the rest — which states on its own pages that the contractors are independent and not affiliated with the site. It matched none of the 312,000+ real local businesses we hold, because it is not one. We have left it in the table because ChatGPT did name it; but the honest reading is that the most widely-repeated "business" in this study is a directory wearing a tradesperson's name.

That is worth sitting with, because it cuts against the comfortable version of our own finding. The one operator that generalised across seven cities did it by publishing a page per city. Whether an assistant should be recommending a referral network to someone asking for the best local landscaper is a separate question, and not ours to settle.

This is good news if you own a small business. ChatGPT is not defaulting to the biggest national brand — it is assembling a local answer, city by city. Which means the seat is winnable.

Finding 3 — some trades are easier than others

We could match 493 of the 1,097 named businesses (45%) to a company we hold in that city. That share varies a lot by trade:

TradeNamed per answerMatched to a local business we hold
plumber5.463%
roofer5.349%
electrician5.842%
HVAC contractor5.539%
landscaper5.432%

An important caveat about that 45%

It is a floor, not a hallucination rate. Our database is large but it is not the whole of American small business. A name we could not match may be a perfectly real company we simply do not hold — a newer business, a different trading name, or one we have not scraped. We are reporting what we could confirm, not claiming the rest is invented. Anyone telling you they can measure AI hallucination precisely from the outside is overselling.

Where the answer was most local

CityBusinesses namedMatched locally
Dallas, TX3537%
Virginia Beach, VA3370%
Nashville, TN3250%
Jacksonville, FL3256%
Kansas City, MO3132%
Milwaukee, WI3148%
Oklahoma City, OK3145%
Charlotte, NC3148%
Phoenix, AZ3037%
San Diego, CA3033%
Tucson, AZ3040%
Minneapolis, MN2952%
Raleigh, NC2938%
Fort Worth, TX2821%
Colorado Springs, CO2829%

What it means if you run one of these businesses

Does ChatGPT name you?

Run the free scan. We will ask it for your trade in your town and tell you whether you were one of the names — plus what is wrong with your site. No card, about a minute.

Check my business
Method. 200 questions put to ChatGPT (gpt-4o-mini with web search) between 31 August and 1 September 2026, one per trade per city, using the prompt pattern "Who is the best {trade} in {city}, {state}? Name specific businesses." Cities were the 40 US cities where our own database holds the most local businesses with websites, so that every answer could be checked against real local data. Business names were extracted from the response and matched case- and punctuation-insensitively against our records for that city. Limitations: one run per city+trade, so this measures a typical answer rather than a stable ranking; a model update will change these numbers; our matching database is not exhaustive, so the matched share is a lower bound. Raw counts are stated throughout so you can check the arithmetic.

Free to cite or quote with a link to this page. Questions: get in touch.

Use this data

The full dataset is published under CC BY 4.0. Use it, check it, disagree with it. The only condition is attribution.

Summary CSV — 200 rows, one per question, with counts.
Mentions CSV — 1,097 rows, one per business named, with whether we could match it.
Full JSON — everything, including the caveats.

If we have got something wrong, we would rather know. The method is above and the raw answers are in the file; both are checkable. patrick@searchwrench.com.

Cite this study

SearchWrench (2026). ChatGPT local business recommendations across 40 US cities. 200 questions, 5 trades, 1,097 business mentions. https://searchwrench.com/research/chatgpt-local-business-study.html