Your AI translates into German, but your German audience hears English.
Both the quality scores and the LSP confirm the translation is correct. The German market hasn’t complained. Well, it hasn’t complained yet…
The patterns are consistent across tools, clients, and source texts. AI translates sentence by sentence – it has no model of the whole. A term rendered one way in the opening appears differently three paragraphs later. The German reader, who expects precision and internal consistency as baseline competence signals, will notice – whereas the English-speaking client can’t.
This is because the register is wrong, in a specific direction. US content arrives loaded with performative enthusiasm signals – the energy and urgency that English-speaking audiences treat as persuasive. In German B2B communication they sound noisy at best, unreliable at worst. Not because German audiences are humorless or resistant to ambition – but simply because claims without demonstration carry no weight. The louder the assertion, the more it invites skepticism.
Take a simple product testimonial. “I absolutely love CoffeeMaster2000. It’s just a wonderful addition to my kitchen” becomes “Der CoffeeMaster2000 passt wunderbar in meine Küche. Ich verwende ihn oft und gerne.”1 The enthusiasm is reduced, the functional aspect remains. German audiences generally want to know why a product is useful, not how much somebody else loves it.
This pattern isn’t new, but it has been studied. Endmark Communication ran a comprehension study in 2003, 2006, and 2009, asking more than 1,000 native German speakers to explain what English-language advertising claims actually meant.2 YouTube’s “Broadcast Yourself” was most commonly understood as “Mache Deinen Brotkasten selbst” – make your own breadbox. Less than a third of respondents got the intended meaning across all three waves – a result that remained largely stable over time. The Endmark researchers named the underlying mechanism doppelter Transferaufwand: the cognitive cost of decoding a foreign-language claim while simultaneously decoding the cultural assumption behind it. The issue is not the language itself, but the second layer – interpreting the cultural assumption behind the claim. The German reader, faced with both costs at once, often stops engaging with the text – quickly and quietly.
The Endmark researchers’ framing for what cross-language advertising costs the reader.
The first step is decoding the foreign language. The second is interpreting the cultural assumption behind the claim – what the speaker is doing by making it, and what kind of response or register it signals. A native English speaker reading “Broadcast Yourself” performs the second step automatically. A German reader, even one fluent in English, carries both costs simultaneously.
Across the three Endmark waves, this double cost predicted comprehension failure better than English proficiency did. Respondents who scored high on English vocabulary tests still failed to interpret claims that depended on Anglo-Saxon advertising conventions. The cognitive load was measurable and showed up consistently across all three waves, operating independently of language skill.
The implication for AI-translated German B2B content: the German reader of a fluently AI-translated text no longer faces the first step, since the language is German, but still has to process the second – and unexpected – step. AI primarily removes the linguistic layer, but the more demanding part – interpreting the cultural assumption – remains, and continues to affect how AI-translated content is received.
And it’s not just the words – the structure is wrong, too. US content leads with the benefit and follows with the support (sometimes). German B2B communication expects the reverse, establishing the evidence before drawing the conclusion. AI translating benefit-led content into German produces benefit-led German – and the structural inversion German readers expect never happens.
The adjustment is not correcting an error but correcting a cultural assumption.
These are not translation errors. They clear every automated quality check, they pass the review of every English-speaking editor, and they reach the German reader intact – only to land wrong.
This pattern also shows up in measurement, not just in observation. Wendler’s 2024 ACL paper analyzed AI translations of real German corporate texts across several models, comparing the source register to the output register systematically.3 The analysis shows that the performative and evaluative dimensions – the parts of language that signal what kind of claim is being made and what level of authority and trust it implies – get flattened across models. The propositional content is largely preserved, while much of the communicative weight is lost.
This is something automated quality checks are not designed to capture. They measure terminology consistency, grammar, fluency, segment-level accuracy. What carries the communicative weight – the performative and evaluative dimensions – sits outside these metrics. A German B2B text that has lost its evaluative dimension still scores well on every automated metric, because the metrics are designed to detect other types of problems.
The failure is structural rather than technical. Better prompts help at the margins, and a different engine produces slightly better results, but neither addresses what the failure actually is: content designed for one cultural communication model, reproduced in another language without transposition.
And yet the German market is already responding. Bitkom’s February 2026 survey of more than 600 German enterprises across industries found 36% naming poor AI quality as their primary barrier to broader AI adoption.4 The figure refers to AI output quality generally, not translation errors specifically – which is exactly the point. German enterprise readers register translation output under the same quality framework they apply to AI in general. The market isn’t complaining – it’s simply withdrawing.
Germany doesn’t have a Loi Toubon, France’s law mandating French in business and public life. Your German audience doesn’t need one. They have something more efficient: they simply stop reading.