ChatGPT
Checking ChatGPT drafts before you publish them
The problem with a fluent draft is that it reads as finished. These are the four things to inspect before anything written this way goes near a publish button.
Take it apart before you sign it off.
Fix in the order that matters
Claims before style, always. Rewriting an unsupported statistic into vagueness removes the finding and keeps the misinformation, which is worse.
Then structure, then rhythm. A draft with sourced claims and a scannable outline is publishable; one with elegant sentences and no evidence is not.
What no check can tell you
Whether the argument is worth making, and whether the audience is the one you think. That is the editorial judgement, and it is cheapest when the mechanical findings have already been cleared.
What ChatGPT output typically gets wrong
Residue
Disclaimers, bracketed instructions and [insert company] left in text presented as final.
Unsourced numbers
Specific, correctly formatted statistics with no source a reader could open.
Hedged claims under bold headlines
The body qualifies what the headline promises, which is the shape of a claim nobody will defend.
Uniform rhythm
Sentences of near-identical length, the strongest single signal that text was generated.
What it produces
- Marketing and landing page copy
- Reports, briefs and internal documents
- Emails and sequences
- Summaries that compress away qualifications
One file
The findings are the same whoever wrote it.
Paste the output and the console returns what the rules observed, with severities you can argue with.