We asked AI 1,040 questions about local contractors. It almost never named the one we asked about.

The numbers

What the questions looked like

The questions were the ones a homeowner types or says: “plumber Boise”, “furnace repair Spokane”, “best electrician sparks”, “drain cleaning Reno”, “emergency plumber”. Each business got about 11 queries built from its own services and towns, asked of each assistant.

In every one of the 16 market-and-trade groups we sampled, a short list of the same three to five companies came up again and again. That shortlist is the new first page of Google, except there is no page two.

Being named is not enough: check the link

Update, November 11, 2026. We kept asking ChatGPT the same questions each day for the businesses we track, and read every answer that did name one of them with a website. Across November 8–11 that was 90 links for 14 businesses.

At least 12 of the 90 links pointed to a web address that does not exist. The assistant built a plausible address from the company name (the name plus “heating”, or the name plus the city) and presented it as the company’s site. For one landscaper, all 6 mentions carried an invented address. Two more links went to a different company with the same name in another state.

So a buyer can be told to call you and still land on an error page or on someone else. The quick fix is to register the addresses the assistant keeps inventing and point them at your site. The lasting fix is making your real address the one every listing, profile and page agrees on.

Small sample, one assistant, four days; “does not exist” means the address had no DNS record when we checked on November 11.

What we’d fix first

This study measured who gets named, not why. These are the fixes we make for clients, based on how these assistants describe and cite local businesses.

• Pages that answer the question. Six of the sites we could read had a main page under 200 words: a phone number and a list of services. Give each service a page that states cost, timing, service area and process in plain text.
• A consistent footprint. The same name, phone, address and description on the Business Profile, directories, and the site. Mismatches make a business harder to trust and easier to skip.
• Being cited by sources the assistant reads. Local press, trade associations, detailed reviews and “best of” roundups are the kinds of sources these assistants read.
• Reviews that say what the job was. “Fixed our water heater same day in Meridian” gives a model something to match a question to; five stars with no text does not.

Check your own business in 30 seconds

Open ChatGPT, Claude or Gemini and ask “best [your trade] in [your city]”. Then ask it two more ways: a cost question and an emergency question. If your name doesn’t come back, a buyer asking the same thing isn’t hearing it either.

We write up the full version for local service businesses: every question, what each assistant answered, who it named instead, and the page changes that would put you in the answer. Ask for yours, or read how AI search optimization works.

Method and limits

• Sampled November 8, 2026 through each provider’s API (ChatGPT via gpt-5.5, plus Claude and Gemini), with a short instruction to answer as a local search assistant. Consumer apps can personalize by location and history, so your own results will vary.
• A business counted as “named” only when it was presented as a business (by name or domain), not when its words appeared in passing.
• Empty or failed responses were excluded, never counted as “not named”. AI answers change from run to run; this is a snapshot, not a ranking.
• The 34 businesses were established companies we were evaluating as prospects. We don’t name them, or the businesses the assistants recommended instead.