We asked eight questions and the answers cited 89 different websites. Not one of them was cited on more than two of the eight. Nobody owns AI search in this market yet, and that includes us: tyso was cited zero times.
There is a second search happening now, and it doesn't hand back a list, it names a winner. So we pointed it at our own market and wrote down what came back. Here is the whole run.
What we measured
Why bother: BrightLocal's Local Consumer Review Survey 2026 found the share of consumers using AI to find a local business went from 6% in 2025 to 45% in 2026. Whitespark data reported by Search Engine Journal puts AI Overviews on 68% of local searches, rising to 97% for hybrid questions like "average cost of X in [city]". Both are overseas numbers. No published study measures how often AI Overviews appear in Australian results, which is the whole reason to measure your own market instead of quoting someone else's.
So, on 2 August 2026, we put eight fixed questions to two engines: Google's AI Overviews, pulled from a Canberra search, and ChatGPT search. The questions are the ones a trade-business owner actually asks while working out who to hire for marketing help, and the list is frozen so every future run is comparable to this one.
The eight:
- who does seo for tradies in canberra
- best website designer for tradies in canberra
- google business profile management for tradies canberra
- how much does a tradie website cost in australia
- is seo worth it for a small trade business
- best marketing for plumbers australia
- how do i get my trade business to show up in chatgpt
- how do i get more leads for my painting business in canberra
Eight questions across two engines is sixteen possible answers. Fourteen came back.
What came back
- 89 distinct domains across 107 citations. Seventy-seven of those 89 were cited exactly once.
- No domain appeared on more than two of the eight questions. The closest thing to a local incumbent is tradiemarketingcanberra.com.au, on two of them, and one of the pages the engine cited was its pricing page.
- The most-cited domain of the lot was reddit.com, five citations across two questions. Not an agency. A forum.
- ChatGPT answered four of the eight questions with no sources at all. It replied out of training rather than going and looking.
- Google showed no AI Overview at all for the two most commercial Canberra queries, "who does seo for tradies in canberra" and "best website designer for tradies in canberra".
- On the cost question the AI Overview cited 15 sources, and every one of them was a cost guide, most with a year in the title.
- tyso was cited on 0 of 8 questions and named in 0 of 14 answers.
One more, and it's the one I'd have preferred not to write. On "how do i get more leads for my painting business in canberra", Google's overview cited antpainting.com.au. That's ANT Painting, the painting business I run with my dad, so read it as a disclosed conflict rather than a clean result. What makes it worth reporting anyway is the company it kept: a Facebook community group, a paint manufacturer's advice page, a painters' training site, three Reddit threads, a field-service software blog, hipages, and a business-insurance blog. Not one marketing agency in the answer. The only trade business in there got in on a plain service page on its own website.
And a correction, because the run produced one. The script first recorded tyso as "mentioned" on that same question. It had matched the four letters of "tyso" inside "Tyson Farrington" in a page title. A substring, not a mention, so it goes down as zero. A measurement that never catches itself out isn't a measurement.
What it means if you're the tradie
Nobody has to be displaced. 89 domains, 77 of them cited once, no one on more than two questions. That's a fragmented citation set, and a fragmented set is the easiest kind to walk into. There's no incumbent here, only an empty chair.
The engines are citing guides, not service pages. On the cost question all 15 sources were cost guides with real numbers in them. On the Google Business Profile question they were how-to guides for tradies. Nothing cited was a "what we do" page. If you want to be quoted, the page has to answer the question, at a URL of its own, in words a stranger could act on.
The cost question is the one worth writing. It pulled the biggest AI Overview of the eight, 15 sources deep, and it lines up with the Whitespark figure, where hybrid "average cost of X" questions trigger an overview 97% of the time. Most trades won't publish numbers. That's exactly why the ones who do get cited.
For the queries closest to a sale, the old game is still the game. The two most commercial Canberra queries returned no AI Overview at all. Those are still won by ranking and by the map pack. Don't let the AI conversation talk you out of the fundamentals.
Where there are no sources, on-page work won't save you. ChatGPT answered four of eight out of training. You can't be cited in an answer that cites nothing. The lever there is being in the model's picture of the market at all, and that's earned on pages other people own: reviews, directories, being talked about. Not your own headings.
Reddit outcited every business in the market. Read that as a signal about what the engines trust, not as an instruction to go and post about yourself. The honest version of "get mentioned on Reddit" is doing work that people bring up without being asked.
How we keep measuring
A single run is a snapshot, not a trend. The answers vary between runs, an API answer isn't identical to what you'd see in the ChatGPT app, and none of this proves how many real customers ask these eight questions, because nobody publishes LLM query volumes. So the same eight questions get asked again on a schedule, and the change between runs is the only claim worth making. We've written before about why we won't promise more than that.
You can run the small version on yourself in about a minute. The free audit asks an assistant to name the businesses it would recommend for your trade in your suburb, then tells you whether you were named and who was instead. Same probe as the one above, pointed at you rather than at us. The full version, the frozen question set and the monthly re-run, is what AI search optimisation actually is.
Nobody owns this answer yet. That's a narrow window, and windows like it close quietly, usually while everyone is still arguing about whether they're real. The only question worth asking is who writes the page before it shuts.