Tracking local AI search in multilingual cities
In cities where buyers ask in more than one language, one set of prompts misses half the market. How to set up one monitor per language per location, keep prompts as true translations, and compare results by language.
In Brussels, a customer looking for a dentist might ask in French or in Dutch. In Montreal it's French or English, and in Barcelona it's Catalan or Spanish. Plenty of other cities have a large community that searches in a second language. If your AI visibility tracking runs in one language only, you're measuring part of the city and reporting it as the whole.
Language isn't a cosmetic setting here. The same question asked in two languages can produce two different answers, with different businesses named and different sources cited. Whether that happens in your category is something to measure, not assume. This guide covers how.
Decide which languages earn a monitor
Start from your customers, not the census. Which languages do your enquiries arrive in? What languages are your reviews written in? Which language versions of your site get traffic? A city can be officially bilingual while your particular customers mostly use one language, and a city with one official language can still have a big customer group that searches in another.
If a language accounts for a real share of your business, give it a monitor. If it's a small group, run it as a quarterly spot check before committing a monthly budget to it.
One monitor per language per location
The structure is a grid: each city crossed with each language. A clinic group with branches in two bilingual cities and two languages in each needs four monitors. Each one carries the city, the language, the same models and the same prompts (translated).
Don't put both languages in one monitor or mix translated prompts into one list. You'll lose the ability to say "we're named in French answers and missing from Dutch ones", which is usually the finding that matters. Name monitors clearly, for example "BE / Brussels / FR" and "BE / Brussels / NL", so the language is obvious in every report.
Before you plan around a language, check that your tool offers it as a monitor language. Coverage varies between products, and the time to find out is before the budget is set.
Keep prompts as true translations
The comparison only works if the French prompt and the Dutch prompt ask the same thing. Write the master set once, then have a native speaker translate each prompt the way a local buyer would type it. Avoid word-for-word machine output, which tends to read like nobody's real question.
"True translation" means the intent stays fixed while the phrasing becomes natural. If the English master says "dentist that's good with nervous patients", the translation should ask that in the way a local speaker would say it, not swap in a different question that seems more popular. Popularity is a separate decision. If one language community asks about something the other doesn't, add it to both sets, or to neither, and note why.
Keep a translation sheet: one row per prompt, one column per language, plus the date of the last edit. When you change a prompt, change every language version the same day. Otherwise a drop in one language could be a wording change and not a real shift.
Compare per language
Once the grid runs, read it in pairs. For each city, put the two language rows next to each other and look at three things.
Who is named. A competitor that dominates in one language and barely appears in the other often has strong content in one language only.
What is cited. The sources behind each language's answers tell you where that community's information comes from. If the cited pages in one language are mostly local directories and press you've never pitched, that's your list.
Which of your pages get used. If you have language versions of your location pages, check whether the right one is cited. An answer in Dutch citing your French page, or no page of yours at all, points to a gap in that language version.
Then check the site side. Crawler logs filtered to each language folder (/fr/, /nl/ and so on) show whether AI bots fetch both versions or only one. Visitor analytics on those same folders shows which language community actually arrives from AI answers and converts.
Budget for the multiplier
Languages multiply responses the same way locations do. Responses per month are prompts x models x days x monitors, and every language adds a set of monitors. A hypothetical 25 prompts on 4 models, run daily in 2 cities with 2 languages each, is 25 x 4 x 30 x 4 = 12,000 responses. Tier it if you need to: the full set in your main language, a shorter set of booking-intent prompts in the second.
Doing it in Promptwatch
Promptwatch sets language on the monitor next to the location, and monitors run their prompt panels daily per model, country and language. That's the grid above, built directly. Every paid plan unlocks all active models, so each language row can cover ChatGPT, Gemini, Perplexity, Copilot, Google AI Overviews and AI Mode alike.
Which plan depends on the level of location you need. If country level is enough, say a brand that wants French and Dutch answers across Belgium as a whole, any paid brand plan works, from Essential ($95/mo). For city monitors you need Business ($579/mo) or an agency plan from Kick-off ($199/mo). Business and Agency Scale also offer coordinate targeting for a specific district, where ChatGPT, Google AI Overviews and Google AI Mode answer as if asked from that point. The 12,000-response example above fits on Business (42,000 responses) or Agency Growth (25,000).
For the comparison work, citation analytics shows the pages and domains cited in each monitor's answers. Agent Analytics, from Professional and on all agency plans, has stackable path filters, so you can split crawler activity by language folder. Visitor analytics attributes AI-referred visits and conversions per landing page. Brand aliases help when your business name is written differently in each language.
AIrketing is another option that sets country, city and language per prompt. It runs weekly and is built for Spanish-speaking markets. For daily tracking across both languages with the crawler and conversion data alongside, we'd use Promptwatch. For a wider look at running several countries, see AI search visibility across international markets. Plans are on promptwatch.com.