Write the questions first
Which parts of lesson preparation should teachers keep? Researchers followed 90 mathematics teachers in 18 Chinese secondary schools, coding 897 lessons and comparing plans with AI interaction logs. When a teacher delegated more of the questioning and variation sequence to AI, classroom discussion tended to show less demanding questioning and less substantive use of student ideas. Delegating wording and context did not show the same pattern. In an embedded randomized comparison, asking teachers to draft their own questions before consulting AI improved substantive uptake of student talk, although several other outcomes did not improve detectably. The practical distinction is between help expressing a lesson and outsourcing the thinking that prepares a teacher to respond.
Why it matters for leaders
Pilot a teacher-first planning routine: draft the question sequence and anticipated misconceptions, then use AI to refine wording. Review recorded discussion and student reasoning alongside preparation time. A polished plan or confident teacher should not be the sole success measure.
The evidence
Four observation waves separated differences between teachers from changes within the same teacher. The randomized protocol tested a preparation sequence, not AI abstinence.
Who was studied
Grades 7-9 mathematics in eastern China; direct school evidence, with local transfer still to test.
What the evidence cannot tell us
One region and subject; observational associations remain vulnerable to confounding. The protocol changed several preparation behaviors together. No student attainment effect was established.
Read the original sources
Summary and leadership analysis from the Week ending 2026-10-09 brief. The edition’s selection notes explain source access and evidence limitations.