A better editing checklist for AI-assisted writing

A first draft from ChatGPT, Claude, or Gemini usually reads smoothly and says little. Most of the editing work is adding what the model couldn’t know and checking what it made up. Work through these in order.

Check every fact

Models produce statistics, quotes, and citations that look right and sometimes aren’t. Look up each number, name, date, and quotation. If the draft cites a paper, open it and find the sentence it supposedly supports. Cut anything you can’t confirm.

Add what only you know

A model writes from averages, so the draft is missing details that only you have: the problem you actually ran into, the option you rejected and why, a number from your own work. Put those in. If a sentence says “There are several factors to consider,” replace it with the factors.

Cut the habits readers notice

Some patterns show up in AI drafts often enough that readers now spot them:

• “Not just X, but Y” and “It’s not X, it’s Y” constructions.
• Lists of exactly three adjectives or verbs, whether or not there are three things to say.
• “Moreover”, “Furthermore”, and “Additionally” opening sentence after sentence.
• Paragraphs that end by restating what they just said.
• “Delve”, “crucial”, “robust”, and “landscape” where a plain word would do.

Removing these makes a draft shorter and easier to read. It won’t do much to how a detector scores the text, since detectors look at structure as well as vocabulary.

Read it aloud

You’ll hear sentences that all run the same length and phrases you would never say. Rewrite those. Let a complicated point take a long sentence. Keep simple ones short.

Check what you’re allowed to do

Before you submit or publish, read the AI-use policy that applies to you: your school’s, your employer’s, or the publication’s. If it asks you to disclose AI use, disclose it. Then ask whether you’d be comfortable telling an editor exactly how the piece was made. If there’s a part you’d rather skip over, that’s the part to fix.