From our data · AI search measurement
The same question has a different answer in every country
A brand selling into three countries usually tracks one number. That number averages three different realities, and it hides whichever one is failing.
By Priyadharshini Krishnan, Co-Founder & Chief AI Officer · Updated 16 September 2026 · 6 minute read
Google AI Overviews are genuinely country-specific. The same question asked from two markets returns different results, different sources and often different brands. Chat assistants work the opposite way and the difference matters: they have no country parameter at all, so a question without a place in it gets answered from a general view that may describe a market you do not sell in. Two rules follow. For Google surfaces, run each question once per market and keep the series separate. For chat assistants, put the market into the sentence, because it is the only place geography can live. And never average markets into a single score. The average will look reassuringly stable while one market quietly goes to zero.
Why is Google different from the chat assistants?
Because they are different kinds of system. A Google search runs against an index with a country setting, and that setting genuinely changes which results and which sources get used. The same query from two countries is two different queries as far as Google is concerned.
A chat assistant has no equivalent. There is no country field to set. Whatever it knows about your geography, it takes from the conversation, which means from the words in your question. Leave the place out and the answer comes from a general view weighted heavily toward whichever markets dominate the material the model was trained on.
The mistake this produces
A brand asks "what is the best X", with no country, gets a flattering answer, and reports it as their visibility. The answer described a market they do not sell in. The number was real, and about somebody else. This is the most common measurement error we see, and it hides well, because the answer itself reads as entirely reasonable.
What happens if you just pick a default country?
You get a wrong answer that looks right, which is worse than an error you can see. We pulled the default out of our own system for that reason. It had defaulted to the United States, and for a brand selling only in Canada and the UK every figure it produced described a market that brand had no presence in and no interest in entering. The system now refuses to run, and asks you to choose.
That refusal annoys people for about ten seconds, which is roughly how long it takes to set a market. A silently assumed one produces months of confident reporting about nothing.
How should a multi-market brand set this up?
- List the markets you actually sell into. Shipping to a country is a different thing. A market with no marketing behind it does not need a tracked score, it needs a decision.
- Run every Google question once per market and keep the results in separate series. Same question text, different market, different number.
- Write the market into your chat assistant questions. "What is the best X in the UK" is a different and far more useful question than "what is the best X", and it is the only lever available here.
- Report per market. Never average. A blended figure moves slowly and conceals the market that collapsed, which is the one thing you needed the report for.
- Expect the competitor set to change between countries. It usually does, and the brands beating you abroad are often not the ones you brief against at home.
What about language?
Treat language as a separate axis from market, because it is. A question asked in another language is a different question drawing on different source material, and the sources for a category in one language may barely overlap with the sources in another.
The pragmatic starting point is whichever language buyers research in, which is not always the one they speak at home and not always the one your marketing is written in. Add a second only once you have evidence buyers use it for this particular decision. Translating the whole set on principle doubles your costs for a number nobody reads.
One global visibility number for a multi-market brand is not a summary. It is three numbers averaged together until none of them can be acted on.
What to do this week
- Find out whether your current tracking has a country set at all, and which one. A surprising number of tools default silently.
- Take your three highest-intent questions and run each per market, kept separate.
- Rewrite your chat assistant questions so the market is named in the sentence.
- Compare the competitors coming back per market. That comparison is usually the most interesting thing the whole exercise produces.
If you are an agency
For any client selling in more than one country this is a scope expansion you can justify in a sentence, because the blended number they have today is demonstrably hiding something and you can show them which market in an afternoon. It also protects you. A flat-looking blended score while one market collapses is the sort of reporting that ends a relationship the month somebody finally notices.
How we do this ourselves
Markets are set per brand in our AI search visibility module and drive the country used for Google results. There is no default: a brand with no market set gets an error. Google AI Overview results are stored per country and appended, so every market keeps its own readable history.