Why rereading no longer works now that you use AI. If it has stopped working at all

Today it's rereading's turn. My suspicion is that if your first contact with a book comes pre-chewed by an AI summary, the second reading loses almost everything that made it worth anything. But was it worth that much? Here I have to declare my bias, and it's a big one: I lived through it, and educational psychology has been saying for decades that rereading is good for rather little. I may be defending something that never worked the way I thought. Let's look at it with the notebook in one hand and the studies in the other.

For three years I've kept a list of books I've reread —essay, novel, poetry— and a notebook where I write down what I found the first time, what I found the second and what I found the third, if there was one. Before I used assistants heavily, rereadings were revealing. After a year of heavy use, they started to feel flat, and I had to rebuild the practice from scratch.

A notebook isn't a study, I know. It's one person, with no control group and a great deal of eagerness to find what he's looking for. But it's where this text comes from and I'm not going to hide it.

What a second reading does that the first didn't

The explanation I give myself is this one, and I offer it as a hypothesis, not a finding. On a first reading, nearly all your attention goes into building the surface: who's who, what happens, what the author is arguing. You have to hold on to what you've already read while the new stuff comes in, guess where the text is heading and correct yourself at every turn. It's work, and it leaves little room to look at how the book is made.

On a reread, that surface is already in memory. Whatever's freed up can go to structure: why this order, what goes unsaid, which image returns at the strategic moments. Maryanne Wolf develops this in Reader, Come Home (Harper, 2018) with the concept of deep reading: the slow processes of inference, analogy and critical thought that only show up when the reader has the time and the footing for them.

If that's how it works, the second reading depends on how the first was done. And that's where AI comes in.

The summary that arrives before the book

When you ask for a summary before reading, you receive the surface ready-made. Plot, characters, thesis. You save time and decide better whether the book is worth it. If you decide to read it, you read it already knowing what it's about.

What you don't receive is the effort of building that surface, and there's recent evidence that this effort matters more than it seems. Shiri Melumad, of Wharton, and Jin Ho Yun published seven experiments with thousands of participants in PNAS Nexus in 2025. Some people had to learn a topic by reading a language model's synthesis; others, by navigating links from a traditional search engine, with the same underlying information. The model group put in less effort, said they'd learned less and, when they wrote advice on the topic, it was shorter, with fewer facts and more alike from person to person. Evaluators who didn't know where each piece of advice came from found it less useful and less trustworthy.

The MIT study we went through in another article points the same way from the writing side: people who produce with an assistant remember much worse what they've produced. Neither study is about rereading, and it's best not to stretch them. What they say is that knowledge that reaches you pre-synthesised settles less deeply than knowledge you build yourself. My hypothesis adds one step: if the first layer set loosely, the second reading has nothing to lean on.

The ones who say rereading was never all that

Now the other side, which has better credentials than my notebook.

In 2013, John Dunlosky, Katherine Rawson, Elizabeth Marsh, Mitchell Nathan and Daniel Willingham reviewed ten study techniques in Psychological Science in the Public Interest. Rereading ended up in the low-utility group, alongside highlighting and —irony for this article— summarising. What really works for remembering is testing yourself and spacing out your study. Rereading gives a feeling of mastery because the text feels familiar, and that familiarity is deceptive.

There's a second argument, an older one. In 1960 David Ausubel proposed advance organizers: giving the reader, before the text, a general frame to help them place what they're about to read. His experiments found that the frame improved retention. From that angle, a prior summary doesn't spoil the reading, it prepares it. And it's what any good teacher did before assigning a book.

The third goes against one of my favourite facts. For years I cited the meta-analysis by Pablo Delgado, Cristina Vargas, Rakefet Ackerman and Ladislao Salmerón (Educational Research Review, 2018) as proof that reading on paper is understood better than on screen. It does say that, but with small print I used to skip: paper's advantage shows up with expository texts and above all when reading in a hurry, and not with narrative. A later meta-analysis by Annika Schwabe and colleagues (Media Psychology, 2022) found no negative effect of screens on comprehension of narrative texts. The novels I reread most are exactly that kind.

And a fourth, common-sense one: most people never reread anything. If a summary lets you rule out five bad books and read one good one all the way through, the maths comes out in the summary's favour.

Why they don't entirely convince me

I accept the third argument without a fight: format wasn't the problem, and I've stopped citing it that way. With the others I have caveats.

Dunlosky and his colleagues were measuring something else. Their question was which technique helps a student remember a syllabus for an exam. Rereading a biology textbook the night before pays off poorly, agreed. But someone rereading an essay that marked them isn't trying to pass. I'm looking for what I didn't see, and a recall test doesn't measure that. The honest thing is to admit I don't have a study measuring what I'm after either, and that my argument rests on something unmeasured.

With Ausubel the difference lies in who makes the summary and what for. An advance organizer in its original sense is a frame more abstract than the text, devised by someone who knows the reader and wants them to read better. What an assistant gives back when you ask it to summarise a novel is something else: the condensed plot, with the emphasis on what's most repeated in the text and most commented on about it. It doesn't prepare you to read, it spares you reading. And Melumad and Yun measure precisely the cost of that saving.

On nobody rereading, they're right, and it isn't an argument against my thesis but against its reach. This text is for people who reread or want to. For everyone else, the summary is a terrific tool for deciding what to read.

A human summary and a statistical one

There's a distinction I find useful even if it's hard to measure. The summary from a friend who has read the book and knows you comes with a direction: «pay attention to chapter three», «don't let the opening scare you off», «the character who matters is the one who seems minor». It selects with you in mind, and that's why it leaves some work done and some work still to do.

An assistant's summary doesn't know you, beyond what you've told it. It selects by what's most prominent in the text, and in what's been written about the text. Sometimes that coincides with what matters and sometimes it doesn't. For a technical report you need to skim, perfect. For a book you want to change something in you, it's like being told the ending of a film… sorry, worse: it's being told the whole film and then told to enjoy it.

That said, the line isn't drawn by the tool but by the question. You can ask an assistant for something else: not to tell you the plot, to tell you only what to look out for, what historical context helps, what a first reader usually misses. That looks a lot more like an Ausubel advance organizer than a summary, and it doesn't take away the reader's work of building the book. The limit is that the assistant doesn't know what matters to you unless you tell it, and we almost never do. We ask for «summarise it» because it's the shortest thing to type, and the shortest thing to type is usually the one that takes the most work off our hands.

A rule for the bedside table

The rule I take from all this is just one: if I suspect a book is going to deserve a reread, I don't ask for a summary beforehand. The first time it gets read whole, wrong hypotheses and murky passages included. Afterwards, if anything, I ask an assistant about references that slipped past me or how other people have read it. In that order, the tool widens; in the reverse, it replaces.

For everything else —instrumental books, reports, essays I'll read only once— summary without remorse.

Does rereading work again like this? My notebook is no use as proof, because since I've had the hypothesis I read looking to confirm it. I'll leave the test to you: a book that marked you, no summary, and then again in a few months. If you notice nothing, maybe Dunlosky was right and I was just in love with a habit.

Definitions

Deep reading: term popularised by Maryanne Wolf for the slow processes of comprehension —inference, analogy, critical thought, empathy— that demand time and sustained attention.

Advance organizer: concept from David Ausubel (1960). Introductory material, more general and abstract than the text, presented before reading it to help integrate the new information.

Fluency illusion: the feeling of having mastered some content because it feels familiar on rereading, even when you can't retrieve it without the text in front of you.

Directed summary: a summary made by someone who knows the reader and selects what's relevant to them. I distinguish it here from the generic summary a language model produces.

References

Wolf, M. (2018). Reader, Come Home: The Reading Brain in a Digital World. Harper. Framework on deep reading and its conditions.

Melumad, S. and Yun, J. H. (2025). Experimental evidence of the effects of large language models versus web search on depth of learning. PNAS Nexus 4(10), pgaf316. <https://doi.org/10.1093/pnasnexus/pgaf316>

Kosmyna, N. et al. (2025). Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task. arXiv:2506.08872. <https://arxiv.org/abs/2506.08872>

Dunlosky, J., Rawson, K. A., Marsh, E. J., Nathan, M. J. and Willingham, D. T. (2013). Improving Students' Learning With Effective Learning Techniques: Promising Directions From Cognitive and Educational Psychology. Psychological Science in the Public Interest 14(1), 4-58. <https://journals.sagepub.com/doi/abs/10.1177/1529100612453266>

Ausubel, D. P. (1960). The use of advance organizers in the learning and retention of meaningful verbal material. Journal of Educational Psychology 51(5), 267-272.

Delgado, P., Vargas, C., Ackerman, R. and Salmerón, L. (2018). Don't throw away your printed books: A meta-analysis on the effects of reading media on reading comprehension. Educational Research Review 25, 23-38. <https://doi.org/10.1016/j.edurev.2018.09.003>

Schwabe, A., Lind, F., Kosch, L. and Boomgaarden, H. G. (2022). No Negative Effects of Reading on Screen on Comprehension of Narrative Texts Compared to Print: A Meta-analysis. Media Psychology 25(6), 779-796. <https://doi.org/10.1080/15213269.2022.2070216>

Further reading

Italo Calvino, Por qué leer los clásicos. Essay on the books we reread and what they give on each pass; the definition of a classic as a book that never finishes saying what it has to say.

Mortimer J. Adler and Charles Van Doren, How to Read a Book (revised edition, 1972). Classic manual on the levels of reading and analytical reading.

Alberto Manguel, Una historia de la lectura (1996). How people have read in different eras and what changes when the way of reading changes.

## You might also like - What happens to your brain when you use ChatGPT, according to MIT? - The illusion of learning - The loss of focus - Knowledge doesn't settle

## Elsewhere - Maryanne Wolf: Building Deep Reading Skills in a Digital World (UCLA Center for Dyslexia, Diverse Learners and Social Justice) - How to teach kids who flip between book and screen (MIT Technology Review)