Find the reason behind your edits
If you keep rewriting the same parts of an AI draft, save the lesson alongside the work. Start with one approved before-and-after example. The goal is fewer repeated corrections, measured on the next few drafts.
Sort feedback into three types
- Facts: the date, price, link, or claim was wrong. Correct the source information.
- Voice: the sentence did not sound like you. Supply an actual replacement and explain why it fits.
- Task: the draft missed the point or used the wrong format. Clarify the audience, purpose, or required output.
Give feedback the assistant can act on
Instead of “make this better,” say: “The opener takes too long to reach the editing problem. Start with the moment someone has footage ready but no clear handoff. Remove the revenue promise; I supplied no evidence for it. End with the workshop registration link.”
Compare my draft and approved revision. Identify up to three reusable lessons. For each, show the specific edit, explain the reason, and suggest a short instruction for future work. Separate general guidance from details that apply only to this draft. Do not update stored instructions until I approve the changes.Review the next draft against the lesson
Keep a simple record: draft, repeated issue, approved guidance, next result. If the assistant follows an instruction too literally, narrow it. “Always write short sentences” can make everything choppy; “keep the invitation brief and put the date near the link” is more useful for that task.
Use a final approval check
Does this address one clear audience situation?
Are the facts and links verified?
Would I say this out loud?
Is the promised outcome supported by what we deliver?
Is the next step obvious?
Has the person responsible approved it for use?Start with three drafts and look for repeated corrections. Keep the rules that help and remove rules that conflict. An approved draft is evidence of progress; a confident self-rating from the assistant is not a substitute for your review.