Primary AI literacy pedagogy before and after generative AI: A critical review

Two recent reviews of AI literacy in primary education closed their searches in 2024, before most classroom research on generative AI had appeared. This critical review narrows that gap. It takes the pre-2023 literature from those reviews and adds studies published through October 2026 in which children roughly aged six to twelve used generative AI tools such as ChatGPT. Twenty empirical papers are examined: eight from before 2023, two on children’s conceptions of AI, and ten on generative AI (nine independent samples). The synthesis covers pedagogical approaches, learning outcomes, assessment, and challenges. Earlier studies mostly taught children about AI, while the generative-AI studies mostly test learning with AI, so the two literatures are not directly comparable. A continuity-with-adaptation framework is offered as a hypothesis, not a finding: hands-on, teacher-guided work is the common thread, and generative AI adds the need to judge fluent but unreliable output and to attend to privacy and age limits. Controlled studies have appeared only recently, cluster in a few settings, and none of those identified report a delayed follow-up. Early childhood was found to be underexplored. The review closes with a research agenda and cautious suggestions for teachers.