Why ChatGPT names four roofers in Dallas and skips the rest

Oct 12, 2026 · 6 min read · ontheshortlist.ai

A homeowner in Dallas types one sentence into ChatGPT after a hailstorm and gets four company names back. The old results page showed ten links and a map with three pins. The answer shows four names, in a paragraph, with a reason next to each one.

BrightLocal counted it across 200,085 local searches in September 2026: 4.1 businesses per answer in ChatGPT, 3.5 in Google's AI Mode and 2.5 in AI Overviews, the summary at the top of Google. On the old results page a roofer competed for thirteen visible slots, ten results and a three-business map. Now it competes for four, and there is no page 2.

The answer is one paragraph

The homeowner reads the answer, picks a name and calls. Sagapixel watched 54 people use ChatGPT for local services in January 2026: 40% felt ready to call and book from the answer alone, without opening a website. The tasks were in healthcare, so read it as behavior, not as roofing data. Ahrefs measured the same thing from the other side on its own site in June 2025. AI assistants sent 0.5% of the visits and 12.1% of the sign-ups. Few people, already decided.

So the question for a Dallas roofer is simple to state and hard to answer from the outside: when the homeowner asks, is my company one of the four?

Why the list changes

Ask the same question twice and the four names move. SparkToro and Gumshoe found in January 2026 that the same list repeats less than 1% of the time. A company is named some of the time, and the useful number is how often.

That is why we measure frequency and never publish a rank. For each city we ask the 12 questions a homeowner asks, to ChatGPT, Gemini, Perplexity and Claude, twice each: 96 answers. A roofer named in 2 of 96 answers has a different problem from one named in 40 of 96, and both have a different problem from one named in none.

What the four have in common

The models have never met a roofer. They read about one. BrightLocal traced 1.97 million AI citations for local businesses in September 2026. Yext traced 6.8 million in October 2025, and 86% came from sources the business controls: its own website (44%) and its listings and directories (42%). Reviews and social were 8%, forums 2%. For a roofer, that puts the website and the listings first. In Google's AI Mode, Ahrefs measured a 0.709 correlation between how often the web mentions a brand and its AI visibility (0.664 and 0.656 on the other two AI surfaces it measured; YouTube mentions and impressions scored higher), and a weak one for backlinks.

Reading roofing sites before measuring Texas, four things stand out as our working hypothesis for what gets a roofer named, and none of them is a badge inside an image:

Insurance and manufacturer certifications written as text on the site, with the certifying body named.
Reviews that say the city and the job. "Replaced our roof in Plano after the April hail" is a line a model can quote. "Great service" gives the model nothing to quote.
A page on the insurance claim: how they work with the adjuster, what happens with the deductible, how long after a storm.
The same name, address and phone on the website, the Google profile, Yelp, Angi and BBB. A mismatch reads as two companies.

The Texas measurement checks these four against which roofers actually get named. We publish the aggregate per city on the Texas roofing page, never a company name.

What we are measuring in Dallas

On October 12, 2026 we ran a first quick check on five Dallas roofers: 2 questions to the 4 models each, 40 answers in all. Two of the five roofers were named at least once, in 4 of the 40 answers; the other three were never named. The full measurement with the 12 questions comes next, and the Texas page carries the aggregate as it updates, never a company name.

If you run a roofing company in Dallas, the free check takes 25 seconds and tells you your own count, with the competitors' names hidden.

Does AI name your roofing company? Check free, 25 seconds, no sign-up.

Sources

BrightLocal, Local AI Visibility Study (200,085 searches, 1,342 locations) · Sep 16, 2026 brightlocal.com
Sagapixel, How users find local services using ChatGPT (54 participants, 177 sessions) · Jan 7, 2026 sagapixel.com
Ahrefs, AI search traffic conversions (own site, 30 days) · Jun 16, 2025 ahrefs.com
SparkToro and Gumshoe, via Search Engine Journal · Jan 2026 searchenginejournal.com
Yext, via Search Engine Land, AI citations come from brand-controlled sources (6.8M citations) · Oct 2025 searchengineland.com
Ahrefs, AI brand visibility correlations (75,000 brands) · Dec 12, 2025 ahrefs.com