Higher education has a reading problem, and right now AI is making it worse. More than half of college students regularly use AI tools in their coursework, and many have found in AI-generated summaries an easy substitute for long-form reading assignments. The result is a widening reading preparedness gap: the distance between what courses ask students to read and what students actually work through. The question facing institutions is not whether to address AI; it is how to get students to use the tools they are already deploying to improve reading rather than replace it.
As Michael Hale, Ph.D., Chief Learning Officer at VitalSource and a learning researcher with more than 35 years of experience, puts it: "For students already uncertain about their reading skills, these shortcuts have become a replacement rather than a support."
Not all AI use in the classroom is the same. AI that summarizes content moves cognitive effort away from the learner. AI that generates practice questions inside the reading puts that effort back on the student, where learning actually happens.
The research case for AI-enabled reading support
One of the most established principles in learning science is the "Doer Effect," developed through research by Ken Koedinger and Carnegie Mellon University's Open Learning Initiative. The finding is straightforward: students who answer practice questions as they read achieve learning gains six times greater than those who read without them. That gap is not a correlation. It is a causal relationship.
The challenge for institutions has always been how to scale the “Doer Effect.” Writing high-quality formative practice questions takes significant time and expertise; resources most institutions cannot sustain across entire course catalogs. However, AI changes that equation, and the question institutions now face is whether they are deploying it in ways that actually reflect what learning science says works. For academic leaders thinking through that question, the Higher Ed Dive Playbook on scaling the Doer Effect with AI offers a practical framework.
"With AI, we can produce high-quality questions at a massive scale," Hale says, "integrating them directly into the reading experience rather than treating practice as a separate activity that competes with it."
Evidence from the classroom
Recently, learning scientists, alongside an instructor at California State Polytechnic University Pomona, embedded AI-generated formative questions directly into digital course materials for real-world classes across two semesters of a cognitive psychology course with more than 200 students. When instructors assigned the embedded practice before each major exam, students who completed it scored an average of two points higher than students from the prior semester.
The most significant improvements were among lower-performing students, reinforcing what broader research has consistently found: well-designed formative practice delivers the greatest benefit to students who need the most academic support. The Cal Poly study also confirmed for the first time that the "Doer Effect" works in practice when using AI-generated questions.
What academic leaders should ask
For institutions evaluating AI tools, the Cal Poly study points to three practical questions. Does the tool ask students to retrieve and apply information, or only to receive it? Is practice built into the reading experience, or separate from it? And what evidence shows it improves learning?
"Technology is simply a set of tools and does not automatically improve learning," Hale notes. "Like any tool, it only improves learning when it is deployed in service of what we already know works and when institutions hold it to the same evidentiary standard they would apply to any other instructional investment."
That distinction carries real consequences. AI that summarizes content on behalf of students moves cognitive effort away from the learner. AI that generates embedded practice questions puts that effort back on the student where it belongs, while making something previously too resource-intensive to scale viable across an entire curriculum.
The reading preparedness gap will not close through better summaries. It will close when institutions decide that AI's role in the classroom is to deepen engagement with material, not make it easier to avoid.
