Recently, alternate reality games (ARGs) have been applied in museum learning to encourage learners' active engagement through playful problem-solving activities. However, most learners have insufficient prior knowledge and metacognitive skills to complete the learning tasks in such games. To support learning with ARGs, there is a need to provide proper feedback so that the learners are capable of self-regulated learning and solving the problems encountered during the learning process. Considering the influences of individual differences, this research developed a chatbot as a learning partner in ARG-based learning to support students' learning with adaptive feedback in a museum. A quasi-experimental study was conducted to evaluate the effectiveness with and without the proposed approach. The results showed that the AI-based chatbot approach could significantly improve learners' metacognition awareness, emotional engagement and behavioural engagement. Besides, it is helpful to facilitate students' double-loop learning.