humaniseur-fr

An editing skill for removing AI-writing patterns from French text and adding voice, personality, and soul. It is used when editing, reviewing, rewriting, or cleaning up French content that reads like ChatGPT or Claude output, and triggers on terms like humaniser, deslopifier, and rendre plus humain as well as the English "make it sound human". It detects and fixes 27 patterns spanning AI vocabulary overuse, anglicisms from English-first models, copula avoidance, formulaic openings, superficial present-participle analyses, em-dash overuse, redundant adjective doublets, the rule of three, sycophantic tone, and typographic tells such as curly quotes instead of guillemets.

The task is structured as a six-step pass: identify AI patterns by scanning for all 27, rewrite problematic sections with natural French alternatives, preserve the core meaning, maintain the intended tone and register, add actual personality rather than only stripping bad patterns, and finish with an anti-AI pass that asks what still makes the text obviously AI and revises. It emphasizes French-specific context, noting that French professional writing is inherently more formal than English, that connectors like neanmoins and toutefois are legitimate in human French, that crucial is the number-one French AI word rather than "delve", and that anglicisms, strict typographic conventions, and the dissertation tradition are distinct tells.

A guiding rule is to never lower the register of the language: formal input must stay formal, because rewriting soutenu prose into casual French is its own detectable artificiality. The enemy is formulaic writing, not formal writing, so well-built subordinate clauses, precise connectors, and long periodic sentences are preserved as features of good French. Only genuinely mechanical traits are removed, such as inflated significance, copula avoidance, synonym cycling, and promotional filler. The documented patterns include significance and heritage inflation, notability and media-coverage insistence, present-participle pseudo-analysis, promotional language, vague weasel attributions, formulaic challenges-and-prospects sections, and overused AI vocabulary, each with before-and-after examples and a find-and-replace checklist. Use it to make French content read as authentic writing at its appropriate register.