Let me be clear: I use AI tools myself. I'm not anti-technology. What I am against is unsupervised AI translation, especially when the stakes are high — your book, your brand, your reputation.
AI translates words. I translate meaning, register, rhythm, and intent. These are not the same thing, and the gap between them is exactly where your translation can succeed or fail.
The 7 Most Common AI Translation Mistakes
These aren't edge cases. These are the errors I find in every single AI-translated manuscript I review. Every single one.
AI maintains English sentence structure in French. The result is grammatically correct but sounds unnatural — like reading a translated instruction manual, not a novel or brand message. French and English have fundamentally different rhythms.
'Eventually' translated as 'éventuellement' (which means 'possibly', not 'finally'). 'Actually' translated as 'actuellement' (which means 'currently'). These errors don't trigger spellcheck — they silently alter your meaning.
Is your narrator sardonic? Tender? Distant? AI assigns a default register — usually formal-neutral — and flattens every voice into the same polished blandness. A teenage protagonist and a Victorian scholar end up sounding eerily similar.
English idioms translated literally produce absurd French. 'Bite the bullet' doesn't become 'mordre la balle' — it becomes 'avaler la pilule' or 'serrer les dents' depending on context. AI picks one without understanding which fits.
French dialogue has very different conventions for speech tags, punctuation, and dialogue dashes (tirets). AI consistently uses English-style quotation marks and rigid he-said/she-said patterns that feel wrong to French readers.
American units, cultural references, and humor often land flat or confusingly in French. AI either leaves them in English (breaking immersion) or translates them literally (creating confusion). A human translator adapts with judgment.
This is the most dangerous mistake: AI-generated French is smooth enough that non-native speakers don't notice the errors. French readers will notice. And it will quietly undermine your credibility.
Side by Side: The Difference in Practice
Here's a simple example from a romance novel:
"She eventually came to understand that his distance wasn't coldness — it was protection."
"Elle a éventuellement compris que sa distance n'était pas de la froideur — c'était de la protection."
"Éventuellement" means "possibly" in French. The sentence now means the opposite of the original.
"Elle finit par comprendre que sa réserve n'était pas de la froideur — c'était une façon de la protéger."
Correct meaning, natural rhythm, and "réserve" is more emotionally precise than "distance" for a French reader.
This is a single sentence. Multiply this across 80,000 words, and you understand why post-editing an AI translation is not a light proofread — it's a substantial revision.
So When Should You Use AI?
There are valid use cases for AI translation:
- Internal documents that won't be published externally
- Getting the gist of a foreign-language document quickly
- First-draft generation for non-sensitive, non-literary content
- Translations that will then be professionally post-edited
What you should not do is publish AI-translated content to French readers without professional review — whether it's a novel, a product description, a website, or a legal document.
My approach: AI as a tool, not a translator
When I work on large translation projects, I sometimes use AI-assisted tools as part of my workflow — for terminology management, consistency checking, or to generate a rough skeleton on very technical passages.
Every single line is then read, evaluated, and rewritten by me. The AI assists the work. It does not replace the judgment. That's the distinction that matters.
If you've already run your manuscript through an AI and you're not sure of the result — send it to me. I offer a free sample assessment (up to 500 words) so you can see exactly what needs fixing before committing to a full post-editing project.