A B2B SaaS company gets its AI search working in the United States. The engines name it, describe it correctly, put it on the shortlist for its category question. Then someone in the room asks how it looks in Germany, or Japan, or Brazil, and the assumption underneath the question is that the answer should be roughly the same, minus some translation work.
It is not the same. Run the query in German and you can get a different shortlist, sourced from different pages, describing your company in different terms, or not mentioning you at all. The visibility you built did not travel. Nothing broke, and nobody deprioritized you. The engine is answering a question you have not done the work for.
Why the visibility does not travel
An engine composes an answer out of sources it can find and trust for the question being asked. Change the language and the region and you change that source pool almost completely.
The pages your English visibility rests on, the comparison articles, the review profiles, the practitioner write-ups, the roundups, are English-language pages, mostly published in your home market, mostly by outlets your buyers there read. A buyer asking in German is answered from German-language sources: local publishers, regional comparison sites, forums in that language, the local review ecosystem, translated documentation. Some of your English corroboration bleeds through, particularly for global brands and for buyers who ask in English anyway. Most of it does not.
So the accurate model is not “your visibility is weaker abroad.” It is that you are a separate entity in each market’s source graph, with a separate reputation, built out of different documents, and in the new market that reputation is close to empty. You are not underperforming in Germany. You have not started in Germany.
That reframe also predicts something teams find surprising: the failure modes reappear from scratch. The category you fought to be filed under correctly at home can be assigned differently in a market that uses a different word for it. The label you cleaned up can be stale again, because the local sources describing you are older or thinner. Every problem you already solved is waiting in the next market in its original packaging.
The translation reflex, and why it disappoints
The instinctive fix is to localize the website. Translate the pages, add hreflang, spin up regional subfolders, publish the product copy in each language.
Do it, because it is table stakes and because a buyer who arrives needs to read you. Just do not expect it to move the answers much, for a reason that will be familiar if you have read anything else on this site: your own site is the weakest input the engine has. Translating it produces a better-translated version of the least-trusted source. The engine is still going to answer the German question from German third-party pages, and you have not touched one of them.
There is a second, subtler problem with the translation instinct. It assumes buyers in the new market ask the same question in a different language. Often they ask a different question. The category may be framed differently, the constraints may be different, the incumbent everyone compares against is frequently a local player you have never had to concede to. Translating your English query map gives you a tidy list of questions nobody there is asking.
What the work actually is per market
Stripped down, entering a market in AI search is the same program you ran at home, rebuilt with local inputs.
Build the query map in the language, from how buyers there actually phrase it, not from translation. Find out who currently gets the answer, which will often include regional vendors invisible from your home market. Identify the specific local sources the engine leans on, which is the same source audit, run against a different set of pages. Then do the slow part: earn corroboration in that ecosystem, in that language, from parties who have no reason to know you yet.
None of that is exotic. It is just not free, and it does not compound across markets the way teams hope. Corroboration is local.
The decision most companies should actually make
Which leads to the strategic point, and the reason this post exists.
If each market is close to a cold start, then AI search visibility is not something you roll out globally. It is something you buy, market by market, at real cost each time. Most companies I talk to have between four and fifteen markets on the map and the budget to genuinely move two.
So the honest move is to pick. Rank markets by revenue concentration or by strategic intent, take the top one or two, and do the source work properly there while accepting that the rest run on translated pages and whatever your English corroboration happens to bleed through. That is a defensible position. What is not defensible is spreading a single market’s budget across nine markets, which reliably produces nine programs that all sit in the flat stretch forever and never get anywhere.
And check first whether the question is live at all. Some markets are further behind in AI adoption for B2B research than the coverage suggests, and in a market where your buyers still start on Google and a local portal, this is a real but future problem. Run the queries before you fund the program.
What I got wrong
I assumed the corroboration would carry. A client had genuinely strong AI visibility in the US, and their next priority was DACH, so I treated it as an extension of a working program: localize the key pages, translate the best-performing pieces, keep the same query set in German.
Months in, the German answers were essentially unchanged. Same competitors named, client absent from the main category question. All the translated content had done was give a small number of arriving visitors a better experience, which mattered, but was not what we were being paid to move.
When I finally ran the source audit properly in German, the picture was obvious and slightly humbling. The engine was leaning on a handful of German-language comparison pages and two regional publications, none of which mentioned the client, none of which had any reason to. The English corroboration we were quietly counting on was not in the pool at all. What eventually moved it was small and specific: coverage in one of those regional publications and a correction to a German comparison page that had the client filed under the wrong category. Two sources, in-language, aimed at the pages the engine was already reading.
The lesson I kept is that “we are strong in AI search” is never a global statement. It is a statement about one language, one region, and one set of pages, and it expires at the border.
Where to start
Pick your most important non-home market. Write down the ten questions its buyers actually ask, in their language, checked with someone who sells there. Run them through the engines a few times, because one run tells you nothing. Read who gets named and which pages get cited.
You will end up with one of two answers. Either the question is not being asked there yet, and you have saved yourself a budget line. Or it is, someone else is winning it, and you now know exactly which pages you would need to change. That list is short, local, and actionable, which is more than a global rollout plan has ever been.
The full sequence is on the methodology page, and if you want it run market by market rather than everywhere at once, that is what the retainer does.