AI’s real test in education is outcomes

Generative AI arrived in education everywhere all at once. Today’s students are surrounded by tools that promise help, answers, and efficiency at every turn. But learning has never been about convenience alone. As AI reshapes how students engage with academic material, the questions are whether it is being built to support how humans actually learn and ultimately improve outcomes. New research on millions of actual higher education student interactions in digital course materials suggests that the answer lies in a deceptively simple idea: active reading (and in AI designed to support it, not replace it). Active reading is a well‑established concept in learning science. It describes how effective readers interact with text: testing their understanding, highlighting key ideas, asking questions, taking notes, and revisiting challenging concepts. These behaviors are strongly associated with better comprehension, retention, and academic performance. Reading, after all, is not a passive act. It is cognitive work. AI TOOLS CAN ENGAGE STUDENTS Yet digital learning environments, and now many AI tools, too often encourage the opposite: Skimming. Outsourced thinking. Letting the machine do the synthesis and interpretation work for the learner. An analysis of nearly 80 million student interactions across Pearson eTextbooks aligned to college courses over two semesters helps us understand how students actually behave when AI tools are built responsibly into learning materials. The findings were striking. Students who used these AI study tools were dramatically more likely to engage in active reading behaviors than those who did not. When students used AI study tools in their eTextbook, they were three times more likely than non-users to be active readers. Further, the data showed that students who used AI tools built into instructor-led digital platforms with assessment features and other interactive tools were over 20 times more likely to be classified as active readers, compar