Abstract: Optimizing learning feedback is a key step in the implementation of formative evaluation. It is vital to improve students’ autonomous learning ability, teachers’ teaching efficiency and the overall quality of classroom teaching. Learning feedback in formative evaluation has unique characteristics in fundamental purposes, interactive mechanisms, process forms, subject relationships and content orientations. It can effectively promote students’ cognitive deepening, ability development and independent learning. However, learning feedback in actual classroom teaching often suffers from one-way operation, low-quality content and low energy efficiency. These defects greatly limit the practical effectiveness of formative evaluation. To tackle these problems, this study proposes targeted strategies. Multi-subject dialogue feedback can facilitate students’ cognitive construction. High-quality feedback can support students’ knowledge understanding, learning reflection and self-regulation. It is also essential to improve the feedback literacy of teachers and students. With the support of artificial intelligence technology, teachers can deliver timely, personalized and traceable learning feedback to optimize classroom learning effects.