When asked if they use AI to help with assessments, students self-report at only 15%, but the speaker believes the actual rate is 100%, indicating either pervasive underreporting or a gap between student perception of 'cheating' vs 'augmenting cognition'.