Engine disagreement
When ChatGPT, Claude, Perplexity, and Gemini name a business for the same question, how often they name the same one, per trade, with the questions compared and the questions left out.
Across the trades
Across 1 of 3 trades, week of Sep 21, 2026: every engine that named a business agreed on 8 of 8 questions (100%); no two engines that named a business agreed on 0 (0%).
Across 1 of 3 trades, week of Sep 21, 2026, not counted: 9 questions where fewer than two engines named a business.
Across 1 of 3 trades, week of Sep 21, 2026: each pair of engines, over the questions where both named a business.
| Engines | Questions where both named a business | Same business named most | Share |
|---|---|---|---|
| Claude and Gemini | 8 | 8 | 100% |
Windows and doors in Phoenix
In the week of Sep 21, 2026: every engine that named a business agreed on 8 of 8 questions (100%); no two engines that named a business agreed on 0 (0%).
Not counted: 9 questions where fewer than two engines named a business.
Which engines named a business, per question compared
| Engines that named a business | Questions | All of them agreed | No two agreed |
|---|---|---|---|
| Claude, Gemini | 8 | 8 | 0 |
Each pair of engines
| Engines | Questions where both named a business | Same business named most | Share |
|---|---|---|---|
| Claude and Gemini | 8 | 8 | 100% |
Download: CSV · JSON · dataset page. CC BY 4.0.
How this is measured
For each question a homeowner asks in the trade's newest weekly run, we take each engine's most-named businesses (the ones named in the most of that engine's answers to the question, ties included) and compare them across engines. Only the engines that named a business on a question take part in it: an engine that answered but named nobody is left out of that question. A question is compared when at least two engines named a business; the questions where fewer than two did are counted apart and not compared. Every engine that named a business agrees when all of their most-named sets share a business; no two agree when no pair of them shares one; a pair overlaps when its two sets share a business, counted over the questions where both named a business. The questions are not published. Results describe what was measured, not a promise of ranking.