More and more people are using AI tools to translate text – and it’s easy to see why. They’re fast, accessible and often good enough for getting the gist. But if you don’t speak the target language, how can you be sure the translation is accurate? Tone, nuance and intention can shift quickly, even with simple sentences. So we tested three popular chatbots – ChatGPT, Gemini and Copilot – to see how they handle everyday phrases.
How we tested ChatGPT, Gemini and Copilot
We used the free versions of each tool and supplied the same short sentences. The aim wasn’t to find a “winner”, but to look at the kinds of issues that can crop up when a model works with limited context. At one point, when working with Arabic, Gemini even asked for additional information before it could commit to an answer – a reminder that these tools can be unsure without context.
Below, you’ll find three types of sentences that frequently cause problems: diplomatic business language, idiomatic expressions and scientific nuance. For each one, we show how a free model might mistranslate the meaning, followed by a more accurate version that keeps tone and intent intact.
1. Business tone
Does the model recognise diplomacy or translate word-for-word?
Original:
“While the proposal didn’t quite hit the mark, we appreciate the effort and look forward to future collaboration.”
German
Free-model translation:
„Während der Vorschlag nicht das Ziel traf, schätzen wir die Mühe und freuen uns auf eine weitere Zusammenarbeit.“ → literally “missed the target”, overly blunt.
Better translation:
„Auch wenn der Vorschlag noch nicht ganz unseren Vorstellungen entsprach, schätzen wir die Mühe sehr und freuen uns auf die weitere Zusammenarbeit.“
French
Free-model translation:
« Bien que la proposition ait raté la cible, nous apprécions l’effort. » → reads as critisising the proposal.
Better translation:
« Même si la proposition n’a pas tout à fait répondu à nos attentes, nous apprécions l’effort et espérons poursuivre notre collaboration. »
Arabic
Free-model translation:
« رغم أن الاقتراح لم يصب الهدف، إلا أننا نقدر الجهد. » → “did not hit the target”, too direct.
Better translation:
« رغم أن الاقتراح لم يلبِّ جميع التوقعات، فإننا نقدر الجهد ونتطلع إلى تعاون مستقبلي. »
2. Idiomatic phrase
Does the model recognise figurative language?
Original:
“She finally decided to take the plunge and start her own company.”
German
Free-model translation:
„Sie beschloss endlich, in das Wasser zu springen und ihr eigenes Unternehmen zu gründen.“ → literally “jump into the water.”
Better translation:
„Sie fasste endlich den Mut, ihr eigenes Unternehmen zu gründen.“
French
Free-model translation:
« Elle a enfin décidé de plonger et de créer sa propre entreprise. » → could be taken literally.
Better translation:
« Elle a enfin décidé de se lancer et de créer sa propre entreprise. »
Arabic
Free-model translation:
« قررت أخيرًا أن تقفز في الماء وتبدأ شركتها الخاصة. » → again literal.
Better translation:
« قررت أخيرًا أن تخوض التجربة وتبدأ شركتها الخاصة. »
3. Scientific precision
Is factual nuance maintained?
“The vaccine is effective in reducing the risk of severe illness, but does not provide absolute immunity.”
German
Free-model translation:
„Der Impfstoff verhindert schwere Krankheiten, bietet aber keinen vollständigen Schutz.“ → “prevents” overstates efficacy.
Better translation:
„Der Impfstoff verringert das Risiko schwerer Erkrankungen, bietet jedoch keinen vollständigen Schutz.“
French
Free-model translation:
« Le vaccin empêche les maladies graves, mais ne protège pas totalement. » → “empêche” = prevents.
Better translation:
« Le vaccin réduit le risque de formes graves, mais n’offre pas une immunité totale. »
Arabic
Free-model translation:
« اللقاح يمنع الإصابة بمرض خطير، لكنه لا يوفر مناعة كاملة. » → too absolute.
Better translation:
« اللقاح فعّال في تقليل خطر الإصابة بأمراض شديدة، لكنه لا يمنح مناعة مطلقة. »
AI is useful, but context matters
Free AI tools are great for quick checks and early drafts. They’re improving rapidly and offer real value when you need a fast sense of meaning. But as these examples show, they can miss tone, intention and factual nuance – especially when working without context.
For anything that requires accuracy, clarity or a specific communication style, combining AI with professional translation gives a far more reliable result. AI provides speed; the human translator ensures the message lands as intended.
If you’d like to explore how this could support your next project, get in touch
