How often AI names a real ADU builder in London, Ontario
Short answer: We asked the AI models behind ChatGPT, Perplexity and Google Gemini which ADU, garden-suite, laneway and custom-home builders they'd recommend for London, Ontario, the same four queries on three runs between August 29 and September 21, 2026. Across 36 checks, a real local builder we track got named 33 times, 91.7%. Split by engine: Gemini named one on all 12 checks, Perplexity on all 12, and OpenAI only 9 of 12. This is a single dated reading from one client's tracker, not a market survey, and the method and every row are below.
## the_method
The method
This is real data from Advance Labs' own AEO tracker, the weekly cron that checks AI-search visibility for a client (Kuzyn Builds, an ADU/garden-suite builder in London, Ontario). Every week it asks three models with web search, Perplexity Sonar, OpenAI GPT and Google Gemini (the tracker defaults are perplexity/sonar, openai/gpt-5.6-sol and google/gemini-3.7-flash; the stored rows record the engine, not the exact model version), all through the Vercel AI Gateway. These are API calls, not the consumer ChatGPT app or Google Search results pages. It asks four fixed queries: "best ADU builder in London Ontario", "garden suite builder London Ontario", "laneway home builder London Ontario", and "custom home builder London Ontario". Each answer is checked against a roster of six real, named local builders (Kuzyn Builds plus five tracked competitors) for whether it names the client, ranks it, or names a competitor instead.
We pulled the three runs where all three engines ran all four queries: 2026-08-29, 2026-09-14, and 2026-09-21. The tracker is meant to run weekly, but the week of 2026-09-07 has no run, so these are not three consecutive weeks. That's 3 runs × 3 engines × 4 queries = 36 checks. Two earlier runs from 2026-06-27 (Perplexity only, before the other engines were wired up) are excluded, so every engine's sample size is identical: 12 checks each.
What "named a real local company" means here
A check counts as a real name when the engine's answer either cited Kuzyn Builds itself or named one of the five competitors we track (EpicMod, Suite Builders, Ontario Outbuilding, Gold Level, HŌM). We can only detect the roster we track. An engine that named a genuine seventh local builder outside that list would show up here as "no real name", so 91.7% is a floor on true share-of-voice for real businesses, not a ceiling.
## the_number
The number
Across 36 checks (3 engines × 4 queries × 3 runs, 2026-08-29 to 2026-09-21), a real, tracked local ADU/custom-home builder was named in the answer 33 times. Gemini: 12/12. Perplexity: 12/12. OpenAI: 9/12.
| Engine | Checks | Named a real local builder | Rate |
|---|---|---|---|
| Google Gemini | 12 | 12 | 100% |
| Perplexity | 12 | 12 | 100% |
| OpenAI (web-search) | 12 | 9 | 75% |
| All three combined | 36 | 33 | 91.7% |
The gap sits entirely on OpenAI: the three checks that came back with no tracked builder named were all OpenAI runs, and all on the same query, "custom home builder London Ontario", once in each run. Gemini and Perplexity named a real builder on every check in this window. Three misses on one query is a pattern in a small sample, not an explanation.
## caveats
What this number is not
- One client, one city, one category: This is the tracker for a single ADU/custom-home builder in London, Ontario. It says nothing about other trades, other cities, or businesses we don't track.
- Small, fixed sample: 36 checks across four fixed query phrasings and three Mondays. A different phrasing, a different week, or a different set of competitors could move the number.
- Bounded detection: "Real local builder" means one of the six names we track. An engine naming a real business outside that roster reads as a miss here, not a hit.
- Not projected forward: We're reporting what these 36 checks measured, not forecasting what next week's run will show.
The reason this is worth publishing anyway: most AI-visibility numbers cited in marketing come from vendor surveys with a category and city nobody can verify. This one is a receipt, not a survey, dated, with the raw rows in the repo.
## faq
Where does this ADU builder AI-visibility data come from?
Advance Labs' own AEO tracker, a weekly Vercel cron that checks AI-search visibility for a client (an ADU/custom-home builder in London, Ontario) across Perplexity Sonar, OpenAI GPT and Google Gemini via API with web search, all through the Vercel AI Gateway. It is not a third-party survey.
Why did OpenAI name a real builder less often than Gemini or Perplexity in this reading?
In this specific window (three runs, 2026-08-29 to 2026-09-21), OpenAI's answers to one query, "custom home builder London Ontario", came back with no tracked builder named on all three runs, while Gemini and Perplexity named a real builder on all 12 of their checks. That's a pattern in one dated sample, not a general claim about which engine is more accurate.
Does 91.7% mean AI search is mostly reliable for local business recommendations?
It means that, for this one client's four queries in this three-week window, an engine named a real local builder on 33 of 36 checks. It does not generalize to other categories or cities. Treat it as one dated data point, not a market-wide rate.
## sources
## related_guides
## want_your_own_reading
This is what the AEO audit's ongoing tracking looks like once it's running: real, dated checks against named competitors, not a one-time score. Want the same weekly reading for your business and category? The AI-visibility audit sets it up. CAD $750-1,500, three days.
See the AEO audit