AI writing tools are noticeably worse in French, is this closing at all?
teach a french immersion program and every ai writing or detection tool we test performs worse in french than in english, sometimes significantly worse. grammarly’s french suggestions are still pretty basic compared to english, quillbot’s french paraphrasing sounds stiff and sometimes grammatically off, and detection tools seem to really struggle telling ai generated french from human french, probably because theres less french training data overall. process over detection, always, but its frustrating when even the process tools arent built with french in mind. wondering if anyone has seen real improvement in french language ai support over the last year or if were all just waiting on the big companies to catch up eventually.
5 Replies
Join the discussion.
Log In to Reply
YES THIS. i tried using an ai grammar checker with my students last semester and it kept flagging perfectly correct quebec french phrasing as errors because it was trained mostly on france french. so frustrating for students who are writing correctly for their region.
theres a 2025 paper out of a montreal university that quantified this gap, showing detection accuracy for french text was roughly 15 to 20 percent lower than english across three major tools. matches what everyone here is describing, its not just anecdotal.
@Greg Simard do you remember the name of that paper, would love to bring it to my department as evidence when we talk about tool purchases this year.
even if the gap closes on paper i still doubt these tools will ever be reliable enough in either language for anything high stakes. french just makes the unreliability more obvious right now.
twenty years teaching at a cegep and this has been a consistent frustration the whole time. french gets treated as an afterthought in almost every ai product rollout, detection or otherwise. the gap has narrowed slightly but its still very noticeable especially with quebecois expressions and regional variations.