evidence boundaries

AI-supported lesson-design teams maintained immediate scores after withdrawal, but durable learning was not tested

Daniele Agostini, Anna Serbati, Federica Picasso, Anastasiya Lipnevich

Frontiers in Education

A crossover study organized 102 teachers-in-training into 21 lesson-design groups, with complete knowledge-test data from 82 participants. Teams used ChatGPT during either the first or second task. Scores rose during AI-supported phases and did not show a detectable immediate decline after withdrawal for the first sequence. Reflections also described efficiency gains and reduced peer idea-building. Short timing, weak test consistency and sequential tasks prevent interpreting this pattern as established long-term retention.