Key facts
- Three separate sessions on 2026-09-26 asked gpt-5.4-mini the same question with web browsing switched off: who are reliable roofing contractors in Dallas, Texas for a full roof replacement, name specific companies.
- Each run named exactly ten companies, but the lists overlapped only a little: 25 distinct company names across 30 mentions.
- Baker Roofing Company was the only name in all three lists. Good Shepherd Roofing, Blue Ribbon Roofing and DFW Roofing Pro each appeared in two of the three.
- Two names may be the same business written differently, Hart Roofing Inc. in run 2 and Hart Roofing & Construction in run 3, and two more share a word without being the same, Elite Roofing in run 1 and Elite Roofing Solutions in run 3; the model's output alone cannot settle either case.
- All three runs opened or closed with the same kind of caveat, that the names were a starting list rather than an endorsement, and pointed the reader to licensing, insurance, warranties and recent reviews on Google, the BBB and Angi.
What we asked, and the settings
On 2026-09-26 we opened three clean sessions with gpt-5.4-mini, en-CA locale, browsing off, default settings, and typed the same prompt each time: "Who are reliable roofing contractors in Dallas, Texas for a full roof replacement? Name specific companies." Nothing else was said in any session. This is the second panel we have run this way; the first, on 2026-09-23, covered Chicago roofers and nine other prompts.
The point of asking three times is to see how stable the answer is. A homeowner asks once and takes the list as the answer. Asking three times shows how much of that list is the model's settled view of Dallas roofing and how much is whatever surfaced that session.
The three lists, in the order the model gave them
Run 1 named Good Shepherd Roofing, Quick Roofing, Elite Roofing, Legacy Roofing, Brite Roofing, Artisan Roofing & Solar, Baker Roofing Company, James Kate Roofing & Solar, Proctor Roofing & Construction and Blue Ribbon Roofing, as a plain numbered list with a short note before it that the model was not endorsing any one company.
Run 2 named C&W Roofing, Hart Roofing Inc., Baker Roofing Company, Strait-Line Roofing, Blue Ribbon Roofing, Good Shepherd Roofing, Highlander Roofing, DFW Roofing Pro, Pinnacle Roofing & Restoration and Roofer Adam Inc., each with a one-line description, and offered to point to roofers who specialize in insurance claims.
Run 3 named Pond Roofing & Exteriors, DFW Roofing Pro, Honest Abe Roofing Dallas, Baker Roofing Company, Brown Roofing Company, Schrader Roofing, Elite Roofing Solutions, Texas Star Roofing, Hart Roofing & Construction and Xtreme Roofing & Construction, and offered a list of questions to ask before signing a contract.
What repeated, and what only looked like it did
Counting exact names, Baker Roofing Company is the only company in all three runs. Good Shepherd Roofing and Blue Ribbon Roofing appear in runs 1 and 2; DFW Roofing Pro appears in runs 2 and 3. That is four companies with any repeat at all out of 25 distinct names, and 21 names that appeared once.
Two pairs need a careful note. Hart Roofing Inc. in run 2 and Hart Roofing & Construction in run 3 may be one business written two ways, or two businesses; we cannot tell from the output and do not count them as a repeat. Elite Roofing in run 1 and Elite Roofing Solutions in run 3 share a word and nothing else we can verify, so they are counted as different.
The practical reading is simple: a Dallas homeowner who asks this question once gets ten names, and a neighbour who asks the same thing gets a list that shares one to three names with it.
Patterns in how the model answered
Every run framed the list as a starting point rather than a recommendation, and every run told the reader to check licensing, insurance and recent reviews, naming Google, the BBB and Angi as the places to look. Run 1 gave names without descriptions; runs 2 and 3 attached a line to each name, describing years in business or a specialty in general terms that we have not verified and are not repeating here.
Two runs volunteered a follow-up: a shortlist of roofers who handle insurance claim replacements, which reflects how much Dallas roofing work follows hail and storm damage. All three mentioned getting several written estimates. None of the runs mentioned a review count, a rating or a price for any company, and none cited a source, which is expected with browsing off.
What this test cannot tell you, and what a roofer can do with it
This page records what one model said across three tries of one prompt on one date with browsing off. It does not verify licensing, insurance, workmanship or reviews for any company named, and a name appearing here is not a recommendation from SearchPod or from the model. A different model, a different date or browsing switched on would produce a different list.
For a Dallas roofing company, the useful part is the shape of the answer. The model draws on what it has read about a company: its name as written consistently across the web, the services it is known for, and the kind of pages that describe it. A roofer that wants to appear in answers like these is working on the same things that win local search: a consistent name and address everywhere, service pages that say plainly what the company does and where, reviews that mention the job type, and a Google Business Profile kept current. SearchPod tracks AI answers for clients with a fixed prompt panel like this one and reports which prompts name them and which do not.
Related questions
No. Across three runs it named 25 different companies. Baker Roofing Company was in all three lists, and three others appeared twice. The other 21 names appeared once.
This page cannot say. It records what the model named, not whether any company is licensed, insured or good at the work. Every run itself said the names were a starting point to verify, not an endorsement.
The model's reasoning is not visible to us, only the output. With browsing off it answers from what it learned in training, and the selection from that memory changes from session to session, which is what three runs are designed to show.
No. None of the three runs mentioned a price, a rating or a review count for any company, and none cited a source. Each told the reader to check reviews on Google, the BBB and Angi and to collect several written estimates.
By being described consistently across the web: one name and address everywhere, clear service pages, reviews that mention the work, and a current Google Business Profile. SearchPod runs a fixed prompt panel for clients to track which prompts name them over time.
Sources
- SearchPod AI prompt panel, run 2026-09-26 (gpt-5.4-mini, browsing off, three runs) (accessed 2026-09-26)
one model, no browsing, one date, three runs; names are recorded as the model wrote them and are not verified or endorsed; SearchPod sells AI visibility work.
Want to know how your business shows up?
Get a free, no-obligation proposal within one business day. We look at your site and your market and tell you plainly what we would do, and what we would not.
Get your free proposal