Revise AI-assisted writing without disguising it: a method and six prompts

The right goal: recover a voice, not pass a test

An AI draft can provide useful structure. The problem starts when the only goal is to give it the appearance of human authorship. Detectors are an unreliable compass, and Kaneme never promises to bypass them.

The useful move is to revisit every generic passage and add what the model does not own: a lived scene, a clear preference, a limit, a verifiable detail or a phrase that belongs to you.

A four-step method

Keep the draft when its structure helps. Run the pattern review, revise one section at a time, and finish by reading aloud. If you would never say the sentence to someone in front of you, it probably has not found your rhythm yet.

• identify stock phrases, flat sentences and weak transitions;
• choose one area to rewrite instead of smoothing everything;
• restore a real example, opinion or constraint;
• measure the result against your reference and keep what sounds right.

Six prompts to start revising

These prompts replace neither your judgement nor a voice measurement. They isolate one problem before you rewrite it in your own words.

• Generic-language diagnosis — “Review this paragraph: [paste text]. List every generic or hollow phrase without rewriting it.”
• Restore a scene — “Replace the general claim with one concrete, dated example and a verifiable detail.”
• Break flat rhythm — “Keep the meaning but alternate short, medium and long sentences.”
• Remove weak connectors — “Delete additive filler; replace it with a real pivot or nothing when the sentence stands alone.”
• Take a position — “Rewrite this with a clear position and one example that supports it.”
• Read-aloud pass — “Identify sentences too formal to say to a friend. Rewrite only those in natural, correct English.”

The loop ends with a check

A successful rewrite is one that recovers your reference without hiding that a model may have contributed to the draft. The pattern review shows passages to revisit; the stamp asks a harder question: does the text still hold up against your voice?