One of the biggest challenges in AI detection is language. Most AI systems are heavily shaped by English and ignore a specification for different languages. That is not surprising: a large share of publicly available training data comes from the English-speaking web. So it’s the easy way to go for English. As a result, many detection systems perform best in English and less reliably in other languages.

But academic integrity questions do not exist only in English. Universities, teachers, and students across many languages face the same need: fairer, more reliable ways to assess written work. When language-specific differences are ignored, the quality of those checks can drop significantly.

What we are changing

With our latest update, Mentafy now takes a dedicated language-sensitive approach. Instead of treating every text through the same lens, we first differentiate by language and then run checks that are better aligned with that specific linguistic context. This improves quality for English texts and leads to a much stronger improvement for languages such as German, where generic English-centered systems often struggle.

In practice, that means a German text is no longer evaluated as if it were simply an English pattern translated into German. It is assessed with a model path that has been trained to better reflect how German actually looks and works.

Our first major step: German

With this release, we have gone all in on training our models with German-language samples. We see this as the first step in a broader language-specific strategy. Now we want to observe how the new German setup performs in real use, learn from edge cases, and improve it further with your feedback.

What comes next

Spanish is already high on our roadmap. We have built a strong base of training data there and expect it to be one of the next languages we expand more deeply. From there, we plan to add further languages based on two factors: customer demand and data availability. Our goal is clear: better AI detection quality across languages, not just in English.

Your feedback matters

If you are already using Mentafy and working with German-language submissions, we would love to hear from you. Feedback on edge cases, text types, and discipline-specific writing styles is especially valuable as we continue to refine the model. This release marks an important step for us: moving toward language-aware AI detection that is more useful, more scalable, and better suited to the real diversity of academic writing.