Does AI localize brand recommendations by country?

A country-named prompt is a question that names a country, a US state or a US city as the place the buyer is in or is going to, such as "best online grocery service, I'm in the US" or "best afternoon tea in London". The first-named brand is the brand an AI answer names first. A local brand is one based in the country the prompt names.

Across 45,569 answers to 162 prompts that named a country other than the US, the first-named brand was local in 62.3% of answers. On the nearest prompts that named no place, only 5.3% of first-named brands were local to that same country.

The overlap ratio compares two numbers. The first is the share of brands two same-day answers have in common, one to the country-named prompt and one to its nearest generic prompt. The second is the same share for two generic prompts that are equally similar in wording. A ratio of 1.0 means naming the place changed nothing; below 1.0 means the brand list moved away from the generic answer.

When the prompt named the US, the overlap ratio was 1.49 (95% interval 1.22 to 1.78): the answer stayed closer to the generic answer than two comparable generic questions do. The default AI answer is already an American answer. Naming another country gave a ratio of 0.72, and naming a US state or city gave 0.46. A second method, which compares each prompt's usual brand set instead of same-day answers, gives 1.42, 0.66 and 0.54.

For scale, two answers to the same prompt on consecutive days share 41% of their brands. That noise floor is far above the overlap between any two different prompts, which is why the ratio uses equally similar generic pairs as its baseline.

The local first-named share was highest when the prompt named Australia (95.9%) or the UAE (89.9%), and lowest when it named Europe or the EU (13.9%) or Mexico (40.9%). Europe-wide prompts were mostly about software, travel gear and services sold across the region, where global platforms win. Mexico prompts mixed local categories, such as mineral water, coffee shops and phone companies, with money transfer, where Wise and Remitly were named first.

UK and US prompts show the difference clearly. A UK-named prompt got a local first-named brand in 70.5% of answers. A US-named prompt got a brand named almost only on US-named prompts in 8.2% of answers, because the US brands it named were the same ones generic prompts name.

Country splits rest on 5 to 25 prompts each, so read them as direction, not as a ranking with fine margins. India (2 prompts) and the Nordics (1 prompt) are left out.

We split country-named prompts by intent. A buying question names the country the buyer is in ("best online school for public procurement training in Canada"). A travel question names a place the person is visiting ("which patisserie in Paris is famous for its macarons").

Outside the US, travel questions got a local first-named brand in 67.8% of answers and buying questions in 53.0%. In a buying question, a global product that sells in that country can still be named first. In a travel question, the answer is usually a venue in that city.

These brands were named first at least 5 times on prompts for three or more countries. Travel marketplaces dominate: Viator, GetYourGuide and Tripadvisor were each named first for 11 country groups, from Japan to Spain. Wise is the main non-travel brand, named first mostly on money transfer prompts for Mexico, Europe and the UK.

A global brand wins a country-named question when it lists local inventory, such as tours in Kyoto or transfers to Mexico. The brand itself is not local, but its pages are.

The home share is the share of all answers naming a brand that came from prompts naming its home country. Each brand below was hand-checked as based in that country and was named first at least 220 times.

Skydive Dubai, Tour for Muggles in London and Bateaux Parisiens in Paris had home shares above 90%. AI almost never names them anywhere else. Some local brands are also known outside their country: Peñafiel was named first 437 times on Mexico prompts but appears in 9,256 answers overall, and SNCF Connect appears in 1,397.

Many US state and city prompts asked for local services: personal injury lawyers, truck accident attorneys and hard money lenders in a named state. The overlap ratio with generic prompts was 0.46, so naming a state did change the brand list.

But the brand named first was often not a firm from that state. Super Lawyers, a lawyer directory, was named first at least 5 times on 91 of 259 state and city prompts. Morgan & Morgan, a national firm, did so on 54, and Avvo on 43. Among lenders, Easy Street Capital led on 37 prompts.

A country-code domain is a web address that ends in a country's code, such as .co.uk, .com.au or .fr. On prompts naming a country outside the US, 10.6% of 426,548 citations pointed to that country's domains. On the nearest generic prompts, 0.51% did.

Australia had the highest share at 41.0%, led by sites like seaplanes.com.au. UK answers cited sites like royalparks.org.uk and tripadvisor.co.uk. Many local brands use .com domains, so this count understates local sourcing.

On the current engines, ChatGPT Search named a local brand first in 62.9% of answers outside the US and Google AI Mode in 58.6%. ChatGPT Search cited the country's own domains in 16.3% of citations, more than twice Google AI Mode's 7.3%. Google AI Mode quoted local currency slightly more often, 21.8% against 17.3%.

The earlier pair shows the same pattern: ChatGPT localized its first-named brand in 65.7% of answers and Google AI Overviews in 59.3%. Both current engines quote local prices far more often than the earlier ones.

We tested each answer for the named country's currency: £ or GBP for the UK, € or EUR for European countries, AED or dirhams for the UAE, yen, pesos and so on. 9.1% of 40,280 answers to country-named prompts quoted it. On the nearest generic prompts, 0.8% did.

Spain and France answers quoted euros most often. Italy answers rarely quoted any price. On prompts naming the US, 18.3% of answers quoted a dollar amount. Canada is left out because the CAD test also matches computer-aided design.

Some prompts imply a country without naming it as the buyer's location, through a local program or law: 401(k), FDIC, 1099 or an EU directive. Others name two countries in a cross-border task, such as hiring in one country from another.

Neither kind moved the answer away from the generic answer. Implied-term prompts had an overlap ratio of 1.62 and cross-border prompts 1.40, both above 1.0. These prompts are about a product category with a regulatory detail, and AI answers them with the category's usual brands.

We found country-named prompts with text rules: country names, US states and major cities as nouns, and codes like US, UK and UAE in capitals. Adjectives such as "German knife" or "Japanese whisky" describe a product's origin, not the buyer's location, and were not counted. Words like local, near me, global or international were not counted either. A prompt was kept if it had at least 20 answers on each of ChatGPT and Google AI Overviews, or 10 on each of ChatGPT Search and Google AI Mode. The tracked prompt set is US-heavy: 332 of 494 country-named prompts name the US or a US place.

A brand counts as local if at least 90% of all its answers come from prompts naming that country, or if its site uses that country's domain. We hand-checked the 100 most frequent brand and country pairs outside the US. The rule agreed with the hand labels for 85.6% of 97 checkable pairs and almost never called a foreign brand local, but it missed 13 local brands that are also named elsewhere, such as Peñafiel and Telcel. By hand, 71.9% of those answers had a local first-named brand against 57.2% by the rule, so the 62.3% headline is a lower bound.

Each country-named prompt was paired with the most similar prompt that names no place, by text embedding. Answers are an observed public data cut of the Parse mirror over the stated window; April 2026 is excluded because brand detection failed for most answers that month. The study describes what the engines answered, not why.

Track AI recommendations by country. Generic questions in this sample mostly surfaced US brands; questions naming the UK, UAE or Australia more often surfaced local brands.

If you sell outside the US, add prompts that name each country you sell in, and track them separately. Earn citations on that country's own domains, list local inventory if you are a marketplace, and publish prices in the local currency. In US local services, get listed on the directories AI names first.