Every conversation about AI in education eventually comes back to a more fundamental question: how long can students actually pay attention? We debate prompt quality, AI tutoring platforms, and assessment redesign. All of that assumes a student who is cognitively present. A 2025 study from Sharpe, Trotter, and Hale in Frontiers in Psychology tested something deceptively simple and found results that should matter to anyone designing a learning experience, with or without AI.
The researchers compared two break structures across 10 weeks of 90-minute university seminars: a traditional single 10-minute break at the 45-minute mark versus 90-second micro-breaks every 10 minutes. Two hundred fifty-three second-year psychology students participated across two consecutive cohorts (2021 and 2022), with the same instructor, identical slides, and counterbalanced conditions. The question was not whether attention declines over time. Decades of vigilance research have already answered that. The question was whether the structure of rest, the rhythm of pausing, could change the shape of that decline.
Micro-Breaks in the Classroom
Students in the micro-break condition averaged 65.13% on end-of-class quizzes. The traditional break group averaged 56.44%. Both groups started strong, scoring around 81–82% on questions covering the earliest slides. Both groups declined. But the trajectory differed in ways that reveal something important about cognitive architecture.
In the micro-break condition, significant performance drops from the initial baseline did not appear until time point 5, roughly 50 minutes into the session. In the traditional condition, significant drops began at time point 3, around 30 minutes in. That 20-minute difference in stable performance is not trivial. It represents a window where students receiving micro-breaks were still retaining material at near-baseline levels while their peers in the traditional condition had already started losing ground.
The micro-break advantage was most pronounced during the middle of each session (time points 3 through 6), where the micro-break group maintained an average 20.6 percentage point lead. Sharpe et al. note that “the micro-break condition appeared to minimize the decline in performance between time points” (p. 4). Consecutive micro-breaks smoothed out the performance curve. The decline still happened, but it was gradual, not catastrophic.
Why the Traditional Break Is a False Recovery
The most striking data point in the study involves the traditional break itself. What the traditional break actually reveals is the depth of the problem it tries to solve. Students had deteriorated so severely by the 45-minute mark that the break looked heroic by comparison. The micro-break condition never fell that far in the first place. The best intervention was not a bigger rescue. It was a design that prevented the collapse from happening.
That finding maps directly onto how I think about cognitive load management in learning environments. The EEG research on short-form video and attention has shown similar patterns: the brain’s engagement with content follows rhythmic cycles that reward periodic disengagement (Yan et al., 2024). Sharpe et al. ground their findings in the same neuroscience, citing rhythmic oscillations and default mode network activation as biological constraints that no instructional strategy can override. The brain needs these micro-pauses. Designing around that need is not a concession. It is good pedagogy.
What Micro-Breaks Mean for Instructional Design
Sharpe et al. frame their findings through cognitive load theory (Sweller, 1988; Sweller et al., 2019) and Mayer’s (2019) segmenting principle: people learn better when material arrives in chunks with pauses between them. The micro-breaks created natural spacing effects, brief consolidation windows where students could process what they had just heard before the next segment began.
This connects to a thread that runs through much of the AI and cognition research I’ve covered on this blog. When Kosmyna et al. (2025) measured neural engagement during ChatGPT-assisted writing, they found that students who thought independently before turning to AI showed stronger cognitive activation. The principle is the same: the brain needs processing time. Micro-breaks give it that space.
AI tools, used carelessly, can fill that space with more input before the consolidation is complete, which is exactly how Fan et al. (2025) describe metacognitive laziness taking hold. The cognitive offloading problem and the attention problem share the same root. Both involve a system, biological or technological, that skips the processing step.
The authors are careful to say that micro-breaks should not be treated as a standalone solution. As they put it, “the incorporation of such strategy, however, must not be considered a sole means maintain student attention” (p. 7). That tracks with everything the broader research shows. Micro-breaks are one structural element in a broader instructional design that should include active learning, retrieval practice, and intentional pacing. They are a necessary piece, not a sufficient one.
Two practical recommendations come through clearly in the data. First, front-load critical content. The performance curves show that students retain the most during the first portion of any session. The most complex, highest-stakes material belongs at the top of a class, not buried in the second half when cognitive resources have thinned. Second, rethink what breaks are for. A single long break halfway through a session is a rescue operation. Frequent short breaks are preventive design. The data favors prevention.
The study also found no significant differences between the 2021 and 2022 cohorts, despite the 2021 group being the first to return to face-to-face learning after COVID lockdowns. Sharpe et al. interpret this as evidence that attention constraints are driven by cognitive architecture, not by environmental factors or the learning context. I find that persuasive and a little unsettling. It suggests that the decline we see in classrooms is not a symptom of disengagement or poor teaching. It is a feature of how human cognition works. And that means no amount of better content delivery, AI or otherwise, will eliminate it. We can only design around it.
The authors go further, suggesting that “traditional extended examination formats may themselves be misaligned with human cognitive capabilities” (p. 6). That is a provocative claim, and one worth taking seriously as we redesign assessment for an AI-integrated world.
The technology we bring into classrooms will keep changing. The cognitive constraints of the students in those classrooms will not. Any instructional design framework, AI-augmented or traditional, needs to start from that reality.
References
- Fan, Y., Tang, L., Le, H., Shen, K., Tan, S., Zhao, Y., Shen, Y., Li, X., & Gašević, D. (2025). Beware of metacognitive laziness: Effects of generative artificial intelligence on learning motivation, processes, and performance. British Journal of Educational Technology, 56(2), 489–530. https://doi.org/10.1111/bjet.13544
- Mayer, R. E. (2009). Multimedia Learning (2nd ed.). Cambridge University Press
- Sharpe, B. T., Trotter, M. G., & Hale, B. J. (2025). Sustaining student concentration: The effectiveness of micro-breaks in a classroom setting. Frontiers in Psychology, 16, 1589411. https://doi.org/10.3389/fpsyg.2025.1589411
- Sweller, J. (1988). Cognitive load during problem solving: effects on learning. Cogn. Sci. 12, 257–285. doi: 10.1207/s15516709cog1202_4
- Sweller, J., van Merriënboer, J. J., and Paas, F. (2019). Cognitive architecture and instructional design: 20 years later. Educ. Psychol. Rev. 31, 261–292. doi: 10.1007/s10648-019-09465-5
- Yan, T., Su, C., Xue, W., Hu, Y., & Zhou, H. (2024). Mobile phone short video use negatively impacts attention functions: An EEG study. Frontiers in Human Neuroscience, 18, 1383913.