Why does an assistant answer about the wrong country?

Because most place names are not unique and the model has to guess which one you mean. Mansfield exists in England, Texas and Ohio. Without a country, a region or a postcode in the question, the model falls back on whichever version is better represented in its sources, which is frequently not yours.

Does adding the country to the question really change the result?

Substantially, and it is the single highest-value change you can make to how you test. Adding a county, state or country turns an ambiguous question into a specific one, and the set of businesses named usually changes with it. Without that qualifier you are measuring several markets blended together.

Are the same directories important everywhere?

No. The platforms that corroborate a tradesperson in Manchester are not the ones that corroborate one in Sydney or Toronto. Each market has its own dominant review sites and directories, and being absent from the ones that matter locally is a common reason a business is missing from answers in that market.

Do assistants behave the same in every language?

No. Answering in another language changes which sources are retrieved, because the model is searching a different corpus. A business with no content in the local language is often invisible in that language even when it is well established in the market it serves.

Should I track each country separately?

Yes. Averaging countries together hides the pattern that matters, because a strong result in your home market will mask absence in a market you are trying to enter. Run the same question set per country with local phrasing, and keep the scores apart.

Do AI assistants give different answers in different countries?

AI assistants give materially different answers in different countries, for four reasons: place names collide across borders, the directories and review platforms an assistant leans on vary by market, assistant and AI-search availability has rolled out country by country, and the words customers use for the same trade change between markets. An unqualified question therefore produces noise rather than a measurement.

Mansfield is in at least three countries

There is a Mansfield in Nottinghamshire, England. There is a Mansfield in Texas, in the Dallas-Fort Worth area. There is a Mansfield in Ohio. Ask an assistant about "the best joiners in Mansfield" and a real share of the answer can be about the wrong continent entirely — and you may not notice, because the answer will look perfectly plausible.

Mansfield is not unusual. Birmingham is in England and in Alabama. Boston is in Lincolnshire and in Massachusetts. Richmond is in south-west London, in Virginia, and in British Columbia. Perth is in Scotland and in Western Australia. Hamilton is in New Zealand, in Ontario and in Lanarkshire. There is a London in Ontario with a population in the hundreds of thousands. Newcastle, Windsor, Cambridge, Rochester, Wellington: the list is long, and in most cases the version with the larger online footprint wins the tie.

That is the part worth sitting with. Ambiguity does not produce an error message. It produces a confident answer about somewhere else, naming businesses that are real, well-reviewed and four thousand miles away. If you run a joinery firm in Nottinghamshire and your tracking says you are never named, you may be measuring the Texan market.

The right and wrong way to ask

Qualifying the location is not politeness or padding. It is the difference between a measurement and noise. Compare these:

Wrong: "best joiners in Mansfield" — ambiguous across three countries.
Right: "best joiners in Mansfield, Nottinghamshire, UK" — unambiguous, and in the vocabulary a British customer uses.
Wrong: "emergency plumber Richmond" — London, Virginia or British Columbia.
Right: "emergency plumber in Richmond, south-west London, UK".
Wrong: "who should I use for conveyancing in Perth" — Scotland or Western Australia, and the legal systems differ entirely.
Right: "conveyancing solicitors in Perth, Western Australia".

Three qualifiers do most of the work: the country, the region or state, and where possible a postcode or postal code. A postcode is the strongest of the three because it is almost never ambiguous, though it is also less like the way a real customer types. Use the realistic phrasing for the questions you want to win, and the postcode version as a control when you need certainty about what you are measuring.

A quick test: take your five most important questions and read them as though you knew nothing about where you are based. If a stranger could reasonably interpret the place name as somewhere on another continent, the question needs qualifying before it is worth recording an answer to.

The source set changes with the market

Assistants that browse lean heavily on third-party corroboration — directories, review platforms, trade registers, local press. Which of those dominate is a market-by-market question, and the differences are stark.

In the UK, trade work is corroborated by platforms such as Checkatrade, Rated People, Yell and Trustpilot. In Australia, hipages and ProductReview.com.au occupy a comparable space. In Canada, HomeStars does. In the United States, Angi, Yelp, Thumbtack and the Better Business Bureau carry weight that they simply do not carry in Britain. In France, PagesJaunes remains a fixture. In Germany, Trusted Shops and sector registers such as Jameda for doctors matter in ways no English-language equivalent does. In India, Justdial and Sulekha are part of the furniture.

The consequence for a business operating in more than one country is specific: your directory checklist is not portable. A strong presence on the UK platforms does nothing for your visibility in Sydney, and the Australian equivalents are not substitutes for the Canadian ones. Each market needs its own pass over which third-party signals actually count, and the answer changes by sector as well as by country.

Google Business Profile is one of the few genuinely global constants, which makes it worth getting right everywhere you have a presence rather than only in your home market.

Availability has never been uniform

AI features do not launch everywhere at once. Google began showing AI Overviews in the United States before expanding, and announced in October 2024 that the feature was reaching more than one hundred further countries, initially in a handful of languages rather than all of them. Availability in Europe lagged other regions. Anthropic introduced web search in Claude in March 2025, initially for paid users in the United States before widening to more countries.

Those specifics will date. The pattern will not. New capabilities arrive in large English-speaking markets first, spread outwards over months, and are sometimes delayed or altered in particular regions for regulatory reasons. Some assistants are unavailable in some countries altogether, and in a few markets the dominant assistant is a domestic product that most Western businesses have never tested against.

Practically, this means two things. A capability you have verified in your own country may not exist yet in a market you are entering, so a flat zero there might reflect the rollout rather than your visibility. And a market that looks quiet today can change character within a quarter when a feature lands, which is an argument for establishing a baseline before you need one.

Language is not translation

When an assistant answers in French, it is not translating an English answer. It is retrieving French sources, weighing French reviews, and summarising a different corpus. The businesses named can be an almost entirely different set from the ones named for the same question in English.

For a business in a bilingual or multilingual market this is the whole game. A law firm in Montreal that publishes only in English is competing for half its market. A supplier in Brussels, Zurich or Barcelona faces the same arithmetic. Translated service pages with proper hreflang markup are not a nicety in those markets; they are the condition of being retrievable at all in the other language.

The same point applies in reverse to businesses selling into a market from outside it. If every page describing your capability exists only in English, an assistant answering a German-language procurement question has no German source to cite, and will cite someone who does.

The words differ even within English

English-speaking markets are the easy case, which is exactly why the vocabulary gaps get missed. The same trade has different names in different places:

A British customer looks for a joiner; an American looks for a finish carpenter.
A British customer looks for a heating engineer or a Gas Safe registered engineer; an American looks for an HVAC contractor.
In the UK you instruct a solicitor and use an estate agent; in the US you hire an attorney and a realtor.
Moving house means removals in Britain, removalists in Australia, and movers in North America.
A British customer visits a chemist; an American visits a drugstore.
Australians search for a tradie in contexts where nobody else would.

Accreditation language varies too, and it is often the deciding filter in a search. Gas Safe, NICEIC and FENSA mean something in the UK and nothing elsewhere. A US customer may search for a licensed and bonded contractor. An Australian customer may search for a licence number. Those terms belong on the pages aimed at those markets, stated plainly, and nowhere else.

If you test with your own country's vocabulary everywhere, you will conclude you are invisible abroad when in fact you have never been asked about in the words local customers use. That is a measurement error, not a visibility problem, and the two call for completely different responses.

What to do if you trade in more than one country

Treat each market as a separate subject. Separate question sets, local phrasing, local place qualifiers, separate scores. Averaging hides the thing you are trying to see.
Give each market a page that can be quoted about that market, with the local address or service area, local terminology, local accreditation and local contact details stated in text rather than in an image.
Build the local corroboration. Claim the directories and review platforms that dominate that country, and keep your business details identical across all of them.
Publish in the local language where that language is how customers buy, and mark it up properly so the right version is retrievable.
Expect different competitors in each market. The firms named alongside you in Manchester will not be the firms named in Toronto, and the gap analysis has to be done per market.

The structural version of this problem — several locations, several markets, one brand — is worth reading about on its own in multi-location and multi-market visibility, because the content architecture question is genuinely different from the measurement one.

Measuring it so the numbers mean something

Country differences do not only change your score. They can invalidate it. A tracker that asks unqualified questions and averages the results across markets will report a number that corresponds to nothing in particular, and will move for reasons that have nothing to do with your business.

Two honest limits apply here. Nobody can see what an assistant told a real customer in any country, so every tool in this space is reporting how models respond to questions asked through public interfaces, not what any individual was told. And because models are probabilistic, a single run in a single market tells you very little — the signal is in the same qualified questions repeated over time, read as a trend. Those are the rules covered in measuring AI visibility honestly.

Start with the market you actually sell in, phrase the questions the way customers there phrase them, name the country, and keep the results apart. If you have never seen how assistants answer a properly qualified question about your business, a free check is a reasonable first data point.