Short, practical things that work on Monday morning.
A standing rule for every AI answer: one place it could be wrong, one way to check it, one non-AI source to consult.
Release the class to independent work too early and they spend twenty minutes rehearsing errors. Wait for a high success rate first.
One genuinely vast fact at the top of a unit creates a need to know that the rest of the unit satisfies. Interest fades; awe lingers.
The student coming back from absence is deciding whether they still belong here. 'We missed you' answers it; 'where were you' does not.
Ten minutes marking a flawed sample as a class does more for quality than any comment you write afterwards.
Students read worked examples passively. Ask the AI to plant a why-question at each tricky step and reading turns into explaining.
Historians check author, date and purpose before the first sentence. Students dive straight in. Teach the pause and everything they read changes.
After an AI-assisted task, one honest question: which parts could you now do alone? The answer calibrates them better than any rule.
A diagram plus your explanation lays down two memory traces instead of one. Decoration doesn't count; structure does.
Not "what did we learn" but "what did you do that worked". The second question builds learners, not just learning.
Feedback describes the past. A specific, slightly-hard goal written in the student's own words points it forwards.
Any model can explain a topic. The rarer prep artefact is the expert's running commentary, doubts and checks included.
