Should You Let ChatGPT Read the Chapter for You?
You have 35 pages to read for tomorrow.
You could spend the next hour skimming, reading, rereading, and trying to figure out what the heck this chapter is actually saying.
Or.
You could upload the chapter to ChatGPT and type:
Summarize this for me.
Thirty seconds later: Voilà! Chapter summarized.
And honestly, why wouldn’t a student be tempted?
If AI can make 35 pages feel manageable in thirty seconds, that seems worth exploring.
But I also know many educators are super concerned when students let AI do the thinking for them. As one of the cognitive scientists I follow said (I can’t remember who, sadly): “Whoever does the thinking does the learning!”
Aren’t students skipping some important thinking when they let AI do the reading for them?
Yes. Clearly.
But does that mean students need to do all the thinking themselves all of the time?
Good question.
We don’t usually insist that a student do every calculation by hand while learning statistics. A calculator can take over some of the thinking so the student can focus on the thinking that matters more.
I’m thinking about all of this right now because, at the end of October, I’ll be teaching two masterclasses: one on how to teach students to take more effective notes, and another on how to help students skim and summarize their reading.
And in this grand (scary?) new world of AI, it seems irresponsible to teach either one without addressing the tools students can use to do some of that reading and note-taking work for them.
So I’ve been digging into the research, trying to answer a question that I suspect a lot of us are asking:
What parts of reading and note-taking actually help students learn—and when might letting AI take over some of that work actually help, especially for neurodivergent students?
That’s what we’re going to get into today, using one very common AI shortcut as our test case: asking ChatGPT to read something and summarize it for you.
What Are You Doing When You Read to Learn?
So first, what is the thinking students are actually doing when they read?
The older students get, the more often they’re simply assigned reading.
Read these 35 pages before class. Read Chapter 7. Read the article and come prepared to discuss it.
Educators who teach young children know just how complex reading is. But somewhere along the way, as students get older and reading becomes the vehicle for learning rather than the thing they’re explicitly learning how to do, all that complexity can disappear inside two little words:
Read this.
But reading to learn isn’t just about getting your eyes from the top of page one to the bottom of page 35.
To make sense of what you’re reading, you have to make decisions:
What matters here?
What’s the big idea and what’s just a detail?
How does this section connect to the section I just read?
What is the author actually trying to say?
How would I explain this in my own words?
Research on generative learning gives us a useful way to understand some of that cognitive work. Meaningful learning involves actively making sense of material: selecting what matters, organizing it into a coherent representation, and integrating it with what you already know (Fiorella & Mayer, 2021).
And often, we can actually see some of that thinking in the notes students take while they read—or while they listen to a lecture. Careful reading combined with good note-taking leaves behind a record of some of the thinking the student was doing along the way.
Give that 35-page chapter to ChatGPT and ask for a summary, and it will do some of that work for you: selecting what it thinks matters, organizing those ideas, and handing you a much more manageable version of the reading.
That might be incredibly useful.
But it also means that some of the cognitive work that could have happened while you were reading and taking notes has already been done for you.
And that distinction might matter for learning.
The finished product isn’t always where the learning happens.
Want to understand what your brain actually needs to do to learn? We’ve got a free guide that breaks down the research into practical strategies.
Explore the Science of Studying →But What About Skimming?
One of those decisions happens before we even get to summarizing: deciding where to put our attention.
“Skimming” sometimes sounds like the lazy cousin of reading. Like you’re reading, but not trying very hard.
But strategic skimming is work.
As a fan of skimming (I’m teaching a whole workshop about it, after all!), here’s something I found fascinating:
In a series of experiments with expository texts, researchers found that people who skimmed under time pressure remembered more of the important ideas than people who simply read half the text. Skimming didn’t offer the same advantage for less-important details or for making inferences. The researchers’ eye-tracking data also showed that skimmers weren’t simply speeding through every sentence equally; they were allocating their attention selectively (Duggan & Payne, 2009).
In other words, effective skimming requires judgment.
What is this text mostly about? Where does the important information seem to live? What deserves more of my attention?
Now imagine that I give the chapter to ChatGPT before I ever look at it and ask:
What are the most important ideas in this chapter?
That might be useful!
It also means ChatGPT just made one of those judgments for me, rather than me doing it for myself.
Again, I don’t think that automatically makes it bad.
For now, I just want us to notice that something has been outsourced: ChatGPT decided what deserved our attention before we ever encountered the text ourselves. (If, that is, we chose to let ChatGPT skim it before we did—which is an assumption we’ll tackle later.)
But Is Outsourcing Some of the Thinking Actually Bad?
Not necessarily!
Humans have always used tools to outsource some of our cognitive work. Researchers call this cognitive offloading: changing a task by using an external action or tool so that it requires less internal cognitive processing (Risko & Gilbert, 2016).
We write grocery lists instead of holding twelve ingredients in working memory. We put appointments in calendars. We use calculators. We draw diagrams. We set alarms.
In other words, we reduce the cognitive demands of tasks all the time.
And often, that’s a good thing.
Offloading some cognitive work can free up capacity for other thinking that matters more.
And the distinction between useful and unhelpful offloading matters especially when we’re thinking about neurodivergent students.
Imagine a student who opens a dense textbook chapter and encounters a wall of unfamiliar vocabulary, complicated syntax, distracting sidebars, fourteen headings, and thirty-five pages of tiny print.
There is cognitive work involved in learning the ideas in that chapter.
There is also cognitive work involved in simply getting through the dang chapter.
Those aren’t necessarily the same work.
For some neurodivergent learners, demands involving attention regulation, working memory, language processing, executive function, or sheer reading load may make accessing the material harder. We shouldn’t assume that AI affects every neurodivergent learner in the same way—or that we yet have strong research telling us exactly which AI supports work best for which learners.
But researchers are already asking an important version of this question for students with learning disabilities: Can generative AI serve as a compensatory support that improves access without bypassing cognitive processes students need to develop? Right now, that’s still an emerging research question, not a settled answer (Seung & Basham, 2026).
Which makes uses like these interesting:
“Explain this paragraph in simpler language.”
“Before I read this, give me three things I should be looking for.”
“I don’t understand the difference between these two concepts. Can you explain it another way?”
Those uses of AI by students might make the original text more accessible.
And if reducing that barrier frees the student to notice relationships, ask questions, connect ideas, and make meaning?
That’s very different from:
“Read this so I don’t have to.”
What Happens When Students Use AI to Read?
We don’t yet have decades of research telling us precisely what happens when students use generative AI while reading. But two recent studies give us some useful clues.
Easier Isn’t the Same as Better Learning
In one randomized study, 405 students ages 14–15 in English secondary schools studied two short history texts. They were assigned to one of three conditions: they could take notes while they read, use an LLM chatbot while they read, or use the LLM and take notes (Kreijkes et al., 2026).
Importantly, the students using the LLM were still reading the original text. The AI wasn’t reading instead of them. It was available as an interactive reading aid: students could ask questions, request explanations, get additional context, or use it in other ways as they worked through the material.
Then, three days later—and without knowing ahead of time that they would be tested—the students were assessed on how well they remembered and understood what they had read.
The students who took notes, either with or without the LLM, showed better comprehension and retention than the students who used the LLM without taking notes.
But guess which one the students preferred?
The LLM.
Students perceived it as more helpful, and many described it as making complicated material more accessible and reducing cognitive load. Meanwhile, students described note-taking as helping them engage more deeply and remember the material (Kreijkes et al., 2026).
I find that tension fascinating.
The AI may genuinely have made the reading easier to access. AND taking notes while reading and using the LLM appears to have supported learning in a way that interacting with the LLM alone did not.
AI as a Reading Replacement? It Depends on the Reader
A second study looked at something different.
Researchers gave 195 college-aged participants several kinds of GPT-based reading support, including an AI-generated summary, an outline, a Q&A tutor, and a Socratic chatbot.
The effects also depended on participants’ baseline reading comprehension. Researchers used performance on a separate, no-AI reading passage to identify relatively lower- and higher-performing readers. AI support tended to improve comprehension for the lower-performing group, while it tended to reduce comprehension for the higher-performing group. The Socratic chatbot produced the biggest improvement for lower-performing readers, while replacing the original passage with an AI summary produced the biggest decline for higher-performing readers. (Etkin et al., 2025).
But the summary condition had one particularly important feature:
Participants read the AI summary instead of the original passage.
Remember that assumption we noticed earlier, when we imagined giving ChatGPT a chapter before we ever looked at it ourselves?
This study tested a version of exactly that: participants in the summary condition read the AI-generated summary instead of the original passage.
And that distinction matters. The study tested what happens when an AI summary replaces the reading altogether. It did not test what happens when students use a summary before, during, or after reading the original text.
So I wouldn’t take these results to mean that AI summaries are bad for learning. What they do suggest is that how helpful AI is may depend on how much support a student needs with the original reading, how they use the AI, and what the AI is replacing.
Taken together, the studies point toward a more complicated picture. AI can make reading more accessible. It can also change or remove some of the cognitive work students would otherwise be doing themselves. And those effects may differ depending on how the AI is used and who is using it.
So, Should You Let ChatGPT Read the Chapter for You?
Should? I don't know about that. But is it a legit option from time to time? Sometimes.
I know. Annoying answer.
But I don’t think the emerging research supports a simple rule that students should—or shouldn’t—use AI to help them read.
AI can make difficult material more accessible. It can reduce cognitive demands that were getting in the way of learning. And it can also take over some of the selecting, organizing, and meaning-making that helps learning happen.
The trick is figuring out which is which.
And that’s where I think educators have work to do.
We can’t just tell students not to let AI do their reading for them. We need to teach them how to take useful notes, skim strategically, identify what matters, summarize what they’re learning, and make thoughtful decisions about when AI is helping them do that work—and when it’s doing the work for them.
That’s part of what we’ll be exploring in both of my upcoming masterclasses:
Beyond Cornell Notes: Non-Annoying Ways to
Teach Students to Take Notes
Once students encounter information, how do they capture and work with it themselves to better support their learning? And when they need a shortcut, how can they use AI without shortchanging their own learning?Beyond the Highlighter: A Smarter Way to Teach Students to
Skim & Summarize Tricky Texts
When students encounter a difficult text, what can they do with their own brains to help them understand it? And how can they leverage AI when they need extra support—especially for neurodivergent learners?
Because if ChatGPT is going to be sitting next to our students while they learn, I want them to know what to do with it.
Want both? Get them free when you join the Anti-Boring Learning Lab!
A Few More Questions About ChatGPT and Reading
Phew! You already read a whole article, and now an FAQ?! You’ve already read all the new content—we promise. This section just makes it easier for humans and robots to find what they need.
Is it bad for students to use ChatGPT to summarize a chapter?
Not necessarily. At the Anti-Boring Learning Lab, I’m more interested in what students do BEFORE they generate the summary, and then what they do afterwards WITH the summary. There’s a difference between replacing the original reading with an AI-generated text and using a summary to get oriented, make a difficult text more accessible, or check your understanding. The emerging research doesn’t give us a simple rule that AI summaries are either good or bad for learning.
Can ChatGPT help neurodivergent students read difficult texts?
Potentially—but we need to be careful not to make sweeping claims about neurodivergent learners. In the Anti-Boring Learning Lab, we’re interested in whether tools reduce barriers while leaving students with meaningful learning to do. AI might make complicated language or an overwhelming reading load more accessible, giving a learner more capacity to engage with the material. Research on how generative AI can best support neurodivergent learners is still emerging.
What’s the difference between cognitive offloading and letting AI do the learning?
Cognitive offloading simply means reducing internal cognitive demands by using something outside your brain, and we humans do it constantly. One of the principles behind the Anti-Boring Toolkit is reducing unnecessary cognitive overload so students have more capacity for learning. The important question with AI is which demands we’re offloading—and whether the tool is creating more capacity for meaningful thinking or doing that thinking for us.
How should students use AI when they’re reading and taking notes?
In the Anti-Boring Learning Lab, we want students to learn strategies they can use with their own brains and make thoughtful choices about when technology can help. That’s what we’ll explore in Beyond Cornell, where we’ll look at note-taking and appropriate AI shortcuts, and Beyond the Highlighter, where we’ll practice understanding difficult texts and using AI for extra support when needed—especially for neurodivergent learners.
Get access to both – and so much more – when you join
the Anti-Boring Learning Lab!
Works Cited
Duggan, G. B., & Payne, S. J. (2009). Text skimming: The process and effectiveness of foraging through text under time pressure. Journal of Experimental Psychology: Applied, 15(3), 228–242. https://doi.org/10.1037/a0016995
Etkin, H. K., Etkin, K. J., Carter, R. J., & Rolle, C. E. (2025). Differential effects of GPT-based tools on comprehension of standardized passages. Frontiers in Education, 10, 1506752. https://doi.org/10.3389/feduc.2025.1506752
Fiorella, L., & Mayer, R. E. (2021). The generative activity principle in multimedia learning. In R. E. Mayer & L. Fiorella (Eds.), The Cambridge Handbook of Multimedia Learning (pp. 339–350). Cambridge University Press. https://doi.org/10.1017/9781108894333.036
Kreijkes, P., Kewenig, V., Kuvalja, M., Lee, M., Hofman, J. M., Vitello, S., Sellen, A., Rintel, S., Goldstein, D. G., Rothschild, D., Tankelevitch, L., & Oates, T. (2026). Effects of LLM use and note-taking on reading comprehension and memory: A randomised experiment in secondary schools. Computers & Education, 243, 105514. https://doi.org/10.1016/j.compedu.2025.105514
Risko, E. F., & Gilbert, S. J. (2016). Cognitive offloading. Trends in Cognitive Sciences, 20(9), 676–688. https://doi.org/10.1016/j.tics.2016.07.002
Seung, Y., & Basham, J. D. (2026). Cognitive offloading in the age of generative AI: What does it mean for students with learning disabilities? Learning Disability Quarterly, 49(3), 121–133. https://doi.org/10.1177/07319487261439132