Which AI assistant matters most for an Australian business, and the entrenched set that is not a platform

There is no published Australian market share for AI assistants that names its sample and its method, so the ranking question has no answer we would print. The question that does have an answer, and a more useful one, is which engine behaves differently for your buyers, because we measured two of them side by side on the same Australian questions on the same day and they did not agree.

That disagreement is the finding. A business that optimises for one engine and assumes the other follows is working from an assumption our own data contradicts.

Two engines, the same 33 batches of questions, one day

On 22 October 2026 we put 33 batches of buying questions to ChatGPT and Gemini with Australia written into the text of each one, because the search interface we use accepts a country parameter, a language parameter and a location parameter, returns HTTP 200 for each, and ignores all three. The snapshot taken that evening holds 64 blind readings, 32 on each engine.

Outcome	ChatGPT, 32 readings	Gemini, 32 readings
Named a business that was not us	16	20
Named no business at all	16	12
Named us	0	0
Cited our domain	0	0

Gemini named a business in 20 of its 32 Australian readings and ChatGPT in 16 of its 32. For an Australian business trying to be named that is a real difference in where the opportunity sits, and it is a per market fact rather than a general one, because our Singapore reading taken the same day with the same method came out level at 19 of 32 on both engines.

The structural difference behind the split

Our wider work gives the mechanism, and it changes tactics rather than just numbers. In an earlier audit of 597 readings across three markets, ChatGPT's single most cited domain appeared in 176 of its 306 readings. Gemini's single most cited domain appeared in 19 of its 291. ChatGPT concentrates hard on a small set of entrenched documentation; Gemini spreads almost flat across many small sites.

For a publisher those are two different games. On ChatGPT you are competing against a short, entrenched list. On Gemini there is no entrenched list to displace, which is why a small specialist site can appear there on merit. In the Australian set the commercial hosts that came back repeatedly were exactly that kind of site: localdigital.com.au in 9 readings, digitalnomadshq.com.au in 8, 3pdigital.com.au in 7, soupagency.com.au in 7, odindigital.com.au in 6, profitgeeks.com.au in 5.

The result that repeats hardest in our data

Across two earlier audits, Gemini cited our own domain 0 times in 921 readings, 0 in 291 in the first and 0 in 630 in the second. In this Australian set it was 0 in 32 again. That is the most consistently reproduced single result in this programme and it is about us rather than about the engine: Gemini is reachable by small sites and it has not reached ours, which is a problem we own and publish rather than explain away.

The Australian wrinkle in the split

One thing is heavier here than in the other markets we read. Sixty six distinct Australian government hosts appeared 139 times across 25 of the 64 readings. Both engines reached for the state, which means in Australia the entrenched set a publisher competes against is not only documentation from the platform operators. It includes the Australian government, and no amount of editorial quality displaces a regulator on a question the regulator owns. The practical consequence is to publish beside those sources rather than against them: state the rule, name where it came from, and say what your business does about it.

What the split does not justify

Two conclusions people reach from an engine split that this data does not support. It does not justify writing different pages for different engines, because both engines read the same web and a page built to please one retrieval system is still just a page. And it does not justify ignoring the engine where you do worse, which is the commonest reaction and the most expensive, because the engine you are absent from is the one with the most room in it. The split is information about where the opportunity is concentrated rather than permission to pick a favourite, and the correct response to a weaker result on one engine is more attention to it rather than less.

What an Australian business should do with the split

• Measure both, always, and report them separately. One engine is a sample of one. Two is a comparison, and the comparison is where the decision lives.
• Treat Gemini as the nearer opportunity for a smaller Australian business. A flat citation distribution means the entry barrier is editorial rather than institutional.
• Treat ChatGPT as the harder, slower one. Competing with an entrenched set takes a genuinely original, checkable artefact rather than a better written page.
• Name the engines in anything you publish about this. Across 102 build spec conversations in our earlier work, 98 of 102 asked for the engines enumerated rather than the phrase all AI.

What we did not measure, and the figure to stop looking for

This run covered ChatGPT and Gemini. Perplexity, Claude and Google AI Overviews were not in it. Our product watches five surfaces and this dataset covers two, and merging those two facts into one sentence is the error we work hardest to avoid. There is also no Australian usage share here, because we measured what the engines say rather than how many people ask them. A supply side reading and a market share are different measurements and they are not comparable.

Questions about AI assistants in Australia

Which one do Australian buyers actually use most

We have no figure worth publishing. What we can say is that the two we measured answer the same Australian questions differently, which means the usage ranking matters less than most people assume.

Should I optimise for one engine

No, and the data is the reason. A page that satisfies a concentrated retrieval system is not automatically reached by a flat one, and our own result on one of the two is zero across 921 readings in earlier work.

Does Google AI Overviews behave like Gemini

They are different surfaces and we do not treat them as one. AI Overviews was not in this run, so anything we said about it here would be an inference rather than a reading. Our page on which sources appear in Google AI Overviews, audited with the links is the separate measurement for that surface.

How do I check my own split

Ask your own buying questions on both engines with Australia in the text, three times each, and record who gets named and which sources appear, engine by engine. Half a day gives you a reading you own.

How quickly does this change

Fast enough to matter. The scorecard behind this page was rewritten during the evening we were reading it, which is why these figures come from a frozen snapshot taken on 22 October 2026.

The incumbent an Australian publisher is actually competing with

In most markets the entrenched set a publisher faces on the concentrated engine is platform documentation. In Australia it is partly the state. Across our 64 blind readings on 22 October 2026, 66 distinct Australian government hosts appeared 139 times in 25 of them, with business.gov.au in 15 and the consumer regulator in 13, which places both engines' Australian answers partly on institutional ground no amount of editorial quality displaces. That changes the tactic rather than the ambition: publish beside those sources instead of against them, stating the rule, naming where it came from, and saying what your business does about it. Then measure both engines separately, because they did not agree in this market, naming a business in 20 of Gemini's 32 readings and 16 of ChatGPT's 32. AI Knows Us reads a fixed question list across ChatGPT, Gemini, Perplexity, Claude and Google AI Overviews and keeps the engines apart in the output, which is the only way to see a split rather than a flattering average. Our own split here was identical and unhelpful: zero on both.