Ask an assistant the same category question from Toronto, from London and from Stockholm and you will get three different shortlists. Not slightly reordered. Frequently a different set of names.
In most industries that is a footnote. In this one it is the entire measurement problem, because iGaming brands operate market by market, compete against different names in each, and are discussed in a different set of places in every one of them.
Any figure that describes your AI visibility as a single global number is describing an average of several unrelated situations. It will not be wrong exactly. It will just be useless.
Why does the answer change at all?
Three separate mechanisms are moving at once, and it is worth separating them because they call for different responses.
Retrieval is local. When a model fetches current pages to support an answer, what it fetches is shaped by where the request appears to come from and what language it is in. Different pages come back, so different brands are available to be named.
The source ecosystem is national. This is the big one for iGaming. The comparison sites, forums, trade coverage and review pages that discuss this industry are overwhelmingly organised by market. The set of publications that shapes answers in one country has almost no overlap with another. Being well covered in one market buys you remarkably little in the next.
The competitive set is genuinely different. Underneath both of the above sits the plain commercial fact that different brands operate in different places. A model naming four brands in one market and four completely different brands in another is often just describing reality accurately.
What this does to a single global number
Consider a brand active in three markets. Suppose it is a default name in one, occasional in the second, and effectively invisible in the third. Averaged, it looks like a moderate position everywhere. That average describes no market the brand actually operates in.
Worse, it hides the only two facts worth knowing. The strong market needs defending, not investment. The invisible one is either the biggest opportunity on the board or a market to stop spending in, and you cannot tell which without looking at it directly. The average obscures both and suggests a comfortable middle that does not exist.
This is why we run and report every question per market rather than rolling markets together. It produces a less tidy chart and a far more useful one.
| Varies by market | Travels between markets | |
|---|---|---|
| Who is named | Almost entirely. Different brands, different order | Very little, beyond the largest international names |
| Sources cited | Heavily. The publications differ by country | Occasionally, for genuinely international coverage |
| Phrasing of the question | Yes, including language and local terminology | The underlying shape of the question is stable |
| Your own site quality | Not much. It is the same site everywhere | Yes. Fix it once and it helps in every market |
| Category clarity | Not much, if you state it plainly | Yes. The single most portable fix on the list |
The useful split: portable work and local work
That last column is the practical takeaway, and it is genuinely good news.
The work on your own property is portable. Stating your category plainly, making your facts machine-readable, writing pages that answer real questions directly. You do it once and it improves your odds in every market at the same time. It is also usually the cheapest work on the list.
The work on third-party presence is local, and it does not transfer. Earning a place in the sources that shape answers in one market buys you nothing in the next, because it is a different set of sources with different people behind them. That work has to be done per market, which makes the order you do it in a real commercial decision rather than a technical one.
Which is why the sequencing matters so much. Do the portable work first, because it lifts every market at once. Then spend the local effort where the ground is softest rather than where the market is biggest, since a market where answers are volatile and no brand is anchored is a much cheaper win than one where an incumbent is cited everywhere.
How to test this yourself
You can see the effect in about ten minutes. Take one category question, ask it in a fresh session, and record the names. Then change nothing except the market you name in the question and ask again. Then again for a third market.
Most people are surprised by how little overlap there is. If you want the fuller method, including what to record and where doing it by hand runs out, it is written up in how to check what AI says about your brand.
Common questions
Can I just use a VPN to test other markets?
It helps and it is not the whole picture. Location is one input among several, and language, phrasing and the sources the model reaches for all move at the same time. A VPN gets you closer than testing everything from one desk, which is the more common failure.
How many markets should I track?
The ones where you actually compete, and no more. Tracking a market you do not operate in produces a number that cannot be acted on. Three or four real markets measured properly beats a dozen measured loosely.
Do the differences between markets narrow over time?
There is no particular reason to expect it. The differences are driven by genuinely different source ecosystems and genuinely different sets of competing brands, and neither of those is converging. If anything, better retrieval makes answers more locally specific rather than less.