2 articles.
Research on learning techniques shows that practice testing and spaced practice work far better than re-reading. AI can generate practice questions, but passive AI summaries are the modern version of a low-utility habit.
The evidence on self-made flashcards is unusually clean: generating your own beat using premade decks in five of six experiments, but only if you paraphrased. Copy-and-paste was the single method where generating conferred no advantage. Here is why that makes AI flashcard generators structurally weaker than they sound, and what the measured error rate is.