Verify you can reconstruct AI-generated reasoning…
When reviewing AI-generated text, verify whether you could reconstruct the reasoning independently - if not, you have received polish without cognitive gain and should write your own version first.
The Test
After reading AI-generated text, close the window and attempt to reproduce the core argument from memory — not word-for-word, but the logical structure. If you can reconstruct why each claim follows from the previous one, the text reflects understanding you already had. If you stall, the gaps between what you can reproduce and what the AI wrote are exactly the areas where you received polish without comprehension.
When This Fires
- Reviewing an AI-drafted email, document, or code comment before sending
- Reading an AI explanation of a technical concept you asked about
- Editing AI-generated content for publication under your name
- Using AI summaries of meetings or research you should understand deeply
Common Failure Mode
The most dangerous pattern: editing AI text for tone and style while leaving the reasoning untouched. You feel like you "made it yours" because you changed words, but you never tested whether you understand the argument. Pennebaker's research shows that constructing the narrative — not polishing it — is what produces understanding. Copyediting AI output is the cognitive equivalent of retyping someone else's notes.
The Protocol
Write your rough version first, before prompting. Use the AI response to identify gaps in your thinking. Then write again, incorporating what the comparison revealed. Each write-compare-rewrite cycle is a generation event, and generation is where encoding happens.