Every translation is checked against a dictionary of false friends and risky expressions for each language pair, with a severity level, reason, and alternative. If there's something serious, a red band appears on the language card.
Errors that cost a client are almost never grammar mistakes: they are correct words that mean something else.
A translator gives you back a flawless sentence and tells you nothing more. If your email says «gift» and you translate it into German, it might come out as «Gift», which means poison in German. If you advertise a preservative and translate it into French, it might come out as «préservatif» (condom). If you want to say «exquisito» (exquisite) in Portuguese, it might come out as «esquisito», which means strange. These are well-written and well-conjugated words: no spell checker flags them, and you don’t see them because you don’t speak the language.
This app checks every translation in two layers before giving it to you. The first is a deterministic dictionary of risks per language pair: 23 rules distributed among the six target languages (7 in English, 5 in German, 4 in French, 4 in Portuguese, 2 in Italian, and 1 in Bulgarian). It covers classic false friends («gift»/«Gift», «preservativo»/«preservative», «embarazada» (pregnant)/«embarrassed», «esquisito» (Portuguese for strange), «burro» (Spanish for donkey/stupid), «salire» (Italian for to go up), «largo» (Spanish for long)), connotations that cheapen your brand («billig», «cheap»), and overly demanding phrases for a dunning notice («Wir fordern», «we demand», «nous exigeons»). Each rule has a high, medium, or low severity, an explanation of why it’s a problem, and a specific alternative. It doesn’t use AI: it runs always, on all translations, and gives the same result every time.
The second layer is the model’s qualitative evaluation of the already translated text, which comes with the back-translation report and can add up to twelve warnings that the dictionary doesn’t cover. Each warning is labeled with its origin—«dictionary» or «analysis»—so that the reliability of each is clear at a glance and you don’t have to guess.
And high risk doesn’t wait for you to open the report: it appears as a red band on that language’s card, with the reason written and a «View and correct» button. From the report, you can replace the segment with the alternative in one tap—the app rechecks the full text afterward—or silence that word if it’s correct for your business. If you copy text that still has an unreviewed high risk, the warning reminds you when you copy it.
Use it free without an account. If you want to save your result, you can sign up afterwards.
Save / Sign upTwo layers, and each warning tells you where it comes from
findings from the deterministic dictionary are marked as «dictionary» and those from the model as «analysis». Knowing how much to trust a warning is part of the warning itself.
High risk finds you
there’s no need to open any report. If there’s a serious false friend, a red band covers the header of that language’s card with the reason and the button to correct it.
Real false friends, not generic lists
the dictionary works by language pair and only alerts you if the suspicious term is actually in the translation and, when the rule requires it, if your source text was what triggered it.
Correction is a tap away
each alert provides a specific alternative—«Geschenk» for «Gift», «günstig» for «billig», «grávida» for «embaraçada», «conservateur» for «préservatif»—and applying it makes the app re-check the entire text.
Not just vocabulary, but also commercial tone
formulas that sound like a threat in a dunning notice are marked, such as «Wir fordern» in German or «nous exigeons» in French, and the commercial standard of the language is proposed.
What isn't a risk in your business stops bothering you
«Don't alert me about this word» silences that specific rule for your future translations, without affecting the rest of the dictionary.
Counts towards the language verdict
a high risk turns that language red and the report summary names it; medium and low alerts leave it amber. If there's nothing, the report also states it instead of remaining silent.
No account needed to see it work
the dictionary layer applies to all translations, even in guest mode and without using AI; the analysis layer comes with back-translation, one per translation run while you're a guest.
Paste your text, choose the languages, and check the traffic-light indicator before sending anything. No account needed for the first translations.
In the features that use AI, responses are generated by artificial intelligence and are not a substitute for professional advice.