The problem
Board exams are marked against one specific textbook, but general AI chatbots answer from a global corpus: often right in general, wrong for the exam, and sometimes invented outright. Students cannot tell an exam-correct answer from a plausible one, and get no signal about what they are actually weak on.

