Do AI assistants answer Australian city level supplier questions accurately, and the test to run on your own city

Our Australian reading was taken at country level, so it cannot answer the city question directly, and saying so is the first useful thing on this page. What we do have is a separate ten city scan that measured cities directly, a country level reading that shows what the engines reach for in Australia, and a method you can run on your own city in an afternoon. Those three together answer the question better than an extrapolation would.

What we measured, and at which level

Reading	Level	What it can tell you
Australian country audit, 22 October 2026, 64 blind readings, 32 per engine	Country	What the engines reach for when Australia is in the question, and who gets named
Ten city scan, 15 and 16 October 2026, 10 cities, 6 buyer questions each, both engines, 120 readings	City	That city specific sources win city questions, in the ten cities named
Anything about Sydney, Melbourne, Brisbane or Perth specifically	Not measured	Nothing. There is no number for those cities in either dataset

The ten cities measured were New York, Los Angeles, Chicago, Houston, London, Manchester, Birmingham, Mumbai, Delhi and Bangalore. No Australian city was in that scan. So for an Australian city the honest position is to carry the country figure and say it is the country figure, which is exactly what this page does.

The city finding that does transfer, because it is about mechanism

Across those 120 city readings, all usable, we were cited zero times in all ten cities. More usefully, every city had its own winner and they were small, city specific domains. localseo.nyc in New York, mainchester.co.uk in Manchester, generativeengineoptimizationuk.agency in London, searchmetric.in in Mumbai, onecity.co.in in Bangalore, houdigitalmarketing.com in Houston. Not large companies. They win because the domain and the page are specifically about that city.

That is a mechanism rather than a market statistic, and mechanisms travel further than numbers do. It predicts that in Sydney or Perth the sites winning city questions will be small and city specific, and it does not tell you who they are, because we did not measure those cities.

What the Australian country reading adds

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 per engine, returning 1,010 distinct hosts of which 435 carry an Australian domain.

Two things in that set bear on the city question. State government hosts appeared repeatedly: business.qld.gov.au in 5 readings, nsw.gov.au in 4, buy.nsw.gov.au in 4, service.nsw.gov.au in 3, business.vic.gov.au in 2, and single appearances from Western Australian, South Australian, Tasmanian and ACT portals plus two local councils. So the engines are already answering Australian questions at a sub national level when the question invites it. And the commercial sources were small Australian specialists, localdigital.com.au in 9 readings, digitalnomadshq.com.au in 8, 3pdigital.com.au in 7, which is the same pattern the city scan found elsewhere.

The test to run on your own city, in an afternoon

1. Write six buying questions a local buyer would ask, each naming your city. Their words. Six is the number the city scan used per city and it is enough to see a pattern.
2. Ask each one three times on each engine you care about. Put the city in the question text rather than in any location setting, because the location parameters answer HTTP 200 and do nothing.
3. Record four things per answer. Whether a business was named, which one, whether it actually operates in your city, and which sources appeared.
4. Score accuracy as a count, not an impression. How many of the named businesses genuinely serve that city. That is the accuracy question and it has a number.
5. Date the file and repeat monthly. One run is a snapshot. Two is a measurement.

Thirty six readings per city is a few hours and it gives you something no national statistic can: a count of how often the machine gets your own city right. Our page on how to check repeatably whether a business appears in an AI answer keeps the method consistent between runs.

The accuracy failure to watch for specifically

A business named for a city it does not serve. In a country where a two hour drive can cross a service boundary, an answer naming a Melbourne firm for a Geelong question is wrong in a way that costs somebody a phone call. The fix is on your own pages rather than with the engine: name the suburbs, the regions and the travel radius explicitly, because an engine cannot infer a service area you have not stated. Our page on drawing the service map first and then building in waves works through exactly that for one Australian city.

Questions about city level AI answers in Australia

Do you have a Sydney number

No, and neither does anybody we can check. Our city scan covered ten named cities and no Australian city was among them. For an unmeasured city, carry the country figure and label it as the country figure.

Are the engines better or worse at city questions than national ones

Different rather than better. The city scan found that city questions are answered by city specific sources, which means a small local site can win them, and that is not true of national questions where the competitive set is wider.

Why not just extrapolate from the ten cities

Because we did it once and had to correct the page. There is no Leeds number, no Dallas number and no Perth number, and inventing one would be the error this page exists to avoid.

Does it help to publish a page per Australian city

Only where the page can say something true of that city and nothing else. Suburbs served, travel radius, local conditions, a local reference. A set of near duplicates competes with itself.

How do I tell whether an answer about my city is wrong

Check whether each named business actually serves it. That is a count you can produce in an hour and it is the only accuracy measure that means anything commercially.

The reading that replaces the extrapolation

A business wanting to know whether the engines get its city right has two options: borrow a number from a city somebody else measured, or measure its own. The second takes an afternoon and produces something defensible. What our country reading can tell an Australian business is that in 28 of 64 readings no business was named at all, and that our own name appeared in zero of them, so the city layer is being built on top of a national layer where a competitor is already named more often than not. AI Knows Us runs a fixed question list at whatever level you set the questions, across ChatGPT, Gemini, Perplexity, Claude and Google AI Overviews, and reports named and cited separately so an accuracy problem can be told apart from an absence problem.