The chatbot has been increasingly applied and investigated in education, along with many review studies from different aspects. However, few reviews have been conducted on chatbot-assisted learning from the pedagogical and implementational aspects, which may provide implications for future application and investigation of educational chatbots. To fill in the gaps, we reviewed relevant studies from the pedagogical and implementational aspects. Forty-six articles from Web of Science and Scopus databases were screened by predefined criteria and analysed step by step following the PRISMA framework. The finding showed diversified learning activities (i.e. exercise, instructions, role-playing activities, collaborative product design, independent writing, storytelling/book-reading, digital gameplay, and open-ended debates) that chatbots could support through presenting knowledge, facilitating practices, supervising and guiding learning activities, and providing emotional support. Chabot-assisted learning was applied in 14 disciplines, mostly in-class for one session, and had overall positive outcomes from academic and affective aspects. Based on the review results, we proposed a RAISE model of effective chatbot-assisted learning: Repetitiveness, Authenticity, Interactivity, Student-centredness, and Enjoyment. We identified eight theories that might be useful in analysing and supporting chatbot-assisted learning: constructivist theories, situated/contextualised learning theories, cognitive theories of multimedia learning, self-regulated learning theories, output hypotheses, flow theory, collaborative learning theories and motivation theories. Future studies on chatbot-assisted learning may be conducted on the use of theoretical frameworks, the application of various technological-pedagogical approaches and learning activities, and the long-term, out-of-class implementations in new areas.