Why your AI visibility is different in every country (and what to do about it)

An AI assistant does not know where the person asking is sitting. Unless the product it runs in passes a location along, or the question states one, the assistant answers as if the buyer could be anywhere. Three things change that: the question says where the buyer is ("for a company in Lyon"), the question is asked in a language that implies a market (French, Finnish, Portuguese), and, for assistants that search the web before answering, the search runs from a given country and returns that country's results.

Of the three, language does most of the work. An assistant answering in French retrieves and quotes French pages. The brands it names are the ones those pages name. A brand with English pages only is competing for a French answer with brands that have French pages, French reviews and French comparison articles, and it usually loses that answer even when it wins the English one.

Assistants build an answer from the sources they retrieve, and those sources are local. A German buyer's answer draws on German review sites, German trade press and German-language roundups. An Italian buyer's answer draws on Italian ones. A tool that is only ever written about in Italian appears in Italian answers and nowhere else, which is why a company that tracks its visibility globally can be unaware of the competitor that beats it in Italy.

The effect is strongest for comparison and shortlist questions, where the assistant leans on lists and reviews, and weaker for factual questions about a single brand, where it can answer from the brand's own site in any language. It also varies by assistant: those that search the web before answering follow local sources closely, while those answering from training data reflect how much has been written about the brand in that language over time.

A single visibility figure across all markets is an average, and averages hide the spread. A brand named in 141 of 220 answers in Germany and 24 of 110 in France reports a respectable combined number while losing most French buyers. The German result also dominates simply because more questions ran there.

The fix is to measure each country separately, in that country's language, and compare the countries side by side. A region can still be summarised, but as the plain average of its countries with the weakest country shown next to it, so a strong Germany cannot hide a weak France.

Start from the questions your buyers ask, but write them for each country rather than translating one set. A buyer in Germany, France or Brazil uses different terms, compares against different local tools, and asks about different rules: data residency, invoicing, consumer law, local certifications. A translated English question measures an answer no local buyer gets. Then run each question as a buyer from that country, in its language, and for search-backed assistants run the search from that country. Ask each question more than once, because answers vary from run to run, and record three things per answer: whether you were named, who was named instead, and which pages the assistant read.

A weak market almost always has one of three causes, and each has its own fix. First, you have no page in the local language that answers the question, so there is nothing for the assistant to quote. The fix is a page in that language, written for that question, not a translated homepage. Second, you are absent from the local lists and review sites the assistant reads. The fix is to get listed on the two or three sources it cites most in that country. Third, a local competitor is present in both and you are not. Their sources show you where to start.

Then re-measure. Because search-backed assistants pick up a new page within days of it being indexed, the French answers can change well before the German ones would need to. Answers that come from a model's training data move more slowly, so track both and expect the search-backed assistants to move first.

Every country and language is another version of every question, so a wide selection multiplies the checks. Start with two or three main markets, in their main languages. Add the rest of a region once you know which countries move the numbers, and keep an eye on the weakest country rather than the regional average.

Targetlytics tracks AI visibility by country and language: pick countries or whole regions, and for each one it writes a question set in that language the way a buyer there would ask, local and legal nuances included. Each question is asked as a local buyer on 12 assistants, and each weak market comes with the job that fixes it.