Visible in English, Invisible in German AI Answers

Ask ChatGPT, in English, who makes the equipment you sell, and it names you in the first sentence. Ask the identical question in German and either nothing comes back with your name in it, or a competitor you have never benchmarked against gets named instead. You did not imagine this. It is not a bug in the model, and a better German translation of your homepage will not fix it by itself. It is a diagnosable condition, and this piece is about running the diagnosis rather than guessing at the cure.

German is the example here because it is common: a large export market, a language most brand teams cannot read closely, and a web that is, for most companies, far thinner in German than in English. Everything below applies to any language pair. Swap German for Polish, Japanese or Portuguese and the mechanism does not change, only the corpus does.

Rule out the boring explanation first

Before anything else, confirm this is not a crawling problem. If a bot cannot fetch your site, none of your content reaches any model in any language, so a crawler block cannot explain why you appear in English and vanish in German specifically. Open yourdomain.com/robots.txt and check for GPTBot, ClaudeBot, PerplexityBot and Google-Extended. If none of them are disallowed, and you are visible in English, the crawler is not your German problem. Save yourself the detour. The cause sits somewhere else.

What "invisible in German" is actually measuring

A model does not hold one fact about your company that it then renders into whichever language you asked in. What it holds is the residue of text, and the text it read about you differs by language, sometimes drastically. Your English footprint might include trade coverage, forum threads, comparison pages built by people with no stake in your revenue. Your German footprint might be your own translated homepage and almost nothing else. Ask a German question and a model, particularly a retrieval-based one, leans on what German-language sources say. If that corpus barely mentions you, it answers from what it actually has: somebody else.

The test: mirrored questions, four platforms, both languages

This is the part you can do this afternoon, without buying anything.

• Write down a handful of questions an actual buyer would ask before they know your brand exists: not "what is [your brand]", but "who supplies X for a mid-sized manufacturer", "what are the leading German suppliers of Y", "compare the main options for Z in Germany". Phrase them the way a German buyer would frame the category, not as a literal translation of your English question. The everyday term for your product is sometimes a different word entirely.
• Run each question in English and in German, on ChatGPT, Claude, Gemini and Perplexity. Write down every brand named, in order, for every answer.
• Ask each question more than once per platform, on different days if you can. These models are probabilistic: one German answer that omits you could be noise, not a pattern. What you are looking for is a pattern across repeated asks, not a single unlucky roll.

Two outcomes are worth separating. One is a genuine gap: you appear reliably in English and never in German, across repeated asks and several platforms. The other is a coincidence: you are missing from one German answer on one platform, and present everywhere else. Only the first is a real diagnosis. Treat one bad answer as an anecdote and move on.

When you are missing, look at who is standing in your place

The names that show up instead of you are the most informative part of the exercise, and the part people skip because the absence itself feels like the whole finding. Read the substitute list carefully and ask three questions about it.

Is it a real competitor, or a distributor selling your own category under a different name? Imagine a pump manufacturer whose German market runs entirely through a distributor that resells the product line under its own brand. A model trained on the German-language web sees the distributor's name attached to the product category and never sees yours, because your name never appeared in that market's commercial text to begin with. That is not lost visibility. It is a name that was never associated with you in that language, which is a different problem with a different fix.

Is the substitute smaller than you, in a market you would expect to lead? That usually points to density rather than size: a trade publication profiled them, an industry association lists them as a member, a forum thread argues about their product. None of that requires them to be a bigger company. It requires someone writing in German to have covered them, and nobody to have covered you.

Does the same name show up across platforms, or does each one name someone different? Consistency across ChatGPT, Claude, Gemini and Perplexity suggests a genuinely dominant German-language source feeding all four. Disagreement between platforms suggests a thin corpus overall, where each model fills the gap with whatever fragment it happened to absorb.

Go find the German-language footprint directly

Once the mirrored-question test confirms a real gap, the next step is not to ask the model anything else. It is to leave the AI entirely and search the German web the way a journalist would, looking for the sources a model could have drawn on. Search your brand name alongside German commercial terms a buyer or journalist would actually use, such as Händler (dealer), Bezugsquelle (source of supply) or Hersteller (manufacturer), together with the German name of your product category. Check industry association member directories, distributor catalogues, and the kind of trade press document that gets indexed but rarely shows up in a casual brand search. Check technical documentation and standards sites relevant to your category, since industrial buyers often trust those over marketing copy.

What you are looking for is not your own site. You already know your site exists in German. You are looking for anyone else, and if the honest answer after a real search is close to nobody, you have found the cause rather than just confirmed a symptom.

What the diagnosis does and does not give you

A diagnosis tells you where the gap is and roughly why. It does not hand you a German trade mention, an association listing, or a distributor page that names you correctly, and nobody can promise you those on a schedule, because they depend on third parties who have their own reasons to write, or not write, about you. What a diagnosis buys you is the ability to stop guessing: knowing the German gap is real rather than assumed, knowing which platforms show it most starkly, and knowing whether the entity standing in your place is a competitor worth worrying about or a distributor selling your own product under a name you have never claimed.

Doing this by hand, for one market, once, is a reasonable afternoon. Doing it across every platform and every market you actually export to, on a recurring basis, is not something a person keeps doing by hand for long. That repeated version, per language and per market, tracked over time instead of checked once, is what PSentry runs: the same mirrored-question logic above, automated across ChatGPT, Claude, Gemini and Perplexity, with the competitors named in your place surfaced per market instead of buried in a folder of screenshots.

Frequently Asked Questions

Could the gap be caused by a name collision rather than a content gap?

Sometimes. If your brand name is also a common German word, a surname, or shares a name with an unrelated company, the model may be drawing on text about that other entity instead of yours. When your substitute list contains something with no plausible connection to your industry, check for a name collision before assuming the problem is a missing corpus.

If a distributor is named instead of me, is that actually bad?

Not automatically. A distributor you have a real relationship with getting named is a legitimate route to the buyer, and arguably a fine outcome. The problem is when neither you nor any channel you recognise appears, or when a distributor is named as if they were the manufacturer, which can misrepresent who actually makes what you sell.

How many times should I repeat each question before I trust the result?

There is no fixed count, but a single ask on a single platform is not enough given how probabilistic these answers are. Repeat each question enough times, across sessions and ideally across days, that the pattern stops changing. If you are still seeing a different answer every time, you have not tested enough yet to conclude anything.

What if I don't speak German? Can I still run this test?

Yes, but do not rely on machine translation to write the questions or to read the answers, since it can smooth over the exact qualifiers that decide who a model names. Ask a colleague or a local partner to phrase the questions and to read the results back to you, at least for the first pass.

If I find no gap in German, am I safe in my other export markets?

No. This diagnosis tells you about German and nothing else. Each language has its own corpus, so a clean result in German says nothing about how you look in French, Spanish or any other market. The test has to be repeated per language, not assumed to generalise from one.