Since the arrival of neural machine translation in 2016, and then generative AI, one question keeps coming up: “What’s the point of a translator anymore?” My clients ask me, sometimes a little awkwardly. It deserves an honest answer, far from both alarmist talk and miracle promises.
What AI does well
Let’s be clear-eyed: today’s tools produce fluent, often correct texts in a few seconds. For grasping the gist of an email in a foreign language, sorting a large volume of documents or getting a first draft, they are genuinely useful. Denying it would be absurd.
I use them myself, as one aid among others: checking a turn of phrase, exploring synonyms, checking terminological consistency across a long document.
What AI does badly
The problem is that AI makes mistakes with total confidence. A misinterpretation, an inverted figure, a missing negation: everything is written in impeccable English, so the error is invisible to anyone who can’t read the source language. Specialists call these “hallucinations”.
Other limits are less visible:
- Context: AI doesn’t know who your text is for, or why. It doesn’t ask questions.
- Tone and culture: a slogan, a training document or a sensitive letter requires adaptation, not word-for-word transposition.
- Confidentiality: pasting a contract or personal data into a free online tool is never harmless.
- Accountability: when an error is costly, nobody can answer for the machine.
An experiment conducted in Vienna by the linguist Waltraud Kolb even showed that reworking a machine translation creates an “anchoring bias”: you stick to the proposed text, and the language becomes poorer (source).
Post-editing: not a magic shortcut
More and more agencies offer “post-editing”: correcting a machine-produced translation at a reduced rate. On paper, everyone wins. In practice, it’s more nuanced. In a survey by the Société française des traducteurs (the French translators’ association), 70% of respondents considered post-editing problematic, notably because achieving professional quality often takes more work than expected (SFT).
Post-editing can work for internal, repetitive, low-stakes texts. For a text that affects your image or your liability, human translation, possibly assisted by tools, remains the best guarantee.
How I work in 2026
For me, the question is no longer “AI or human”, but “who stays in control”. In practice:
- I analyse the need: audience, purpose, stakes, deadlines. This diagnosis determines the method.
- I choose the tools: translation memories, terminology databases, and sometimes AI, in a secure environment.
- I translate and revise with a critical eye, sentence by sentence.
- I commit to the result: it’s my name and my responsibility.
I am transparent with my clients about the tools I use. It’s a matter of trust.
In short
AI is a powerful tool, not a translator. It speeds up certain tasks, but it understands neither your stakes nor your readers. The value of a professional translator is shifting: less typing, more judgement, checking and advice. And it is precisely that judgement you are paying for.
Unsure whether to use machine or human translation for a project? Write to me: I’ll help you choose the right solution for your needs and your budget.