This work focuses on detecting AI-generated content in academic peer reviews, a crucial task for maintaining the integrity of the scholarly review process. We propose CoCoDet, which concentrates on semantic content rather than superficial stylistic features, leading to more reliable AIGC detection.