There's a lot of noise about AI "revolutionizing" education, and most of it is marketing. But a recent major international education report offers something more useful: actual research on when AI tutoring tools help students learn, and when they don't.

The core finding is encouraging but conditional. Generative AI tools can meaningfully improve the quality of tutoring and boost student learning outcomes - this shows up most clearly when the tutor using the AI tool is relatively inexperienced, and the AI is helping fill gaps in their own subject or pedagogical expertise. In other words, AI tutoring works best as a support system for the tutor, not as a replacement for one.

The report also flags an important caution: when AI tools are used without any pedagogical guidance - just handing a student a chatbot and letting them outsource an assignment to it - the research doesn't show real learning gains. It shows better-looking output, which isn't the same thing. The difference between those two outcomes comes down entirely to how the tool is used, not the tool itself.

For students using AI tools to study - whether that's an actual AI tutoring app or just ChatGPT for a tricky physics topic - the implication is direct: the tool helps most when you're using it to genuinely work through a concept (asking it to explain something a different way, quiz you, or check your reasoning) rather than to generate a finished answer you can copy. The research suggests that's not just an ethics preference - it's the difference between a tool that actually makes you better and one that just makes your homework look done.