Make them explain it before the AI does

When a student asks a chatbot to explain something they have not tried to explain themselves, the answer lands on nothing and slides straight off. Reverse the order. Require a written or spoken attempt first, then let the AI respond to that attempt. Chi et al. (1994) showed that generating your own explanation produces markedly better learning than reading a good one, even when the self-explanation is flawed. It is also the cheapest integrity move available: the thinking happens before the tool arrives. (Manning's explain-first-interrogator skill turns this into a full dialogue protocol.)

In practice

I will explain [CONCEPT] in my own words first.
Do not correct me until I have finished.
Then:
1. Name the strongest part of my explanation.
2. Ask me questions about the weakest part.
   No direct corrections for three rounds.
3. If I am still stuck, give a bridge hint,
   not the full answer.
End by asking me to restate the whole idea.

• Set the rule for AI study sessions: your own explanation goes in the first message, and the AI may only respond to it, never replace it.
• Have the AI affirm the strongest part, then probe the weakest gap with questions rather than corrections, for up to three rounds before any hint.
• Collect the first messages occasionally; they are a better record of understanding than the polished output at the end.
• For students with genuinely zero footing, allow one orientation question first; repeated probing of a blank page feels punitive, not rigorous.