Imagine a book on your nightstand about why people make bad decisions. You bought it hoping to understand your own habits a little better. You have made it through the introduction. Now imagine asking an AI assistant to explain the book’s central argument, give you three examples that apply to your life and identify its weakest claims. Within minutes, you have something that feels useful. You ask a follow-up question. The explanation gets clearer. The book is still on the nightstand. But the reason to finish it is less obvious.
That is a serious challenge to the value of some books. It deserves a better answer than an automatic defense of reading or a confident prediction that artificial intelligence will replace it. Books do different jobs. They explain, investigate, entertain, persuade and preserve experience. AI’s ability to perform one of those jobs does not tell us whether it can replace the others. But neither should our affection for books prevent us from acknowledging when a faster alternative serves a reader well.
The most vulnerable book, in this sense, may be one whose main contribution is packaging familiar information. Consider a manager whose weekly meetings keep running long without producing decisions. She could buy a book on effective meetings. Or she could describe the problem to an AI assistant, ask for a few approaches and work with it to draft an agenda for Tuesday. When a suggestion seems impractical, she can explain why and ask for another. If her goal is a better meeting next week, that exchange may provide much of what she wanted from the book. The relevant comparison is what each approach helps her accomplish, not how many pages she reads.
A book must anticipate where an audience might struggle. An AI assistant can respond to the particular point at which one person gets stuck. Ask for a simpler explanation, a different example or an objection, and the conversation can change direction. For a book built around a few useful ideas and many repetitive illustrations, that creates a plausible substitute. This is an argument about usefulness, however, not evidence that AI has already reduced such books’ sales or prices. Establishing that would require separate evidence about buying behavior and its causes.
There is also a complication: Getting an explanation and acquiring an understanding are not always the same achievement. Return to the reader with the book about bad decisions. Suppose the AI explains the tendency to favor information that supports what we already believe. The idea seems obvious. Yet the next morning, the reader dismisses criticism of a favorite business proposal and eagerly forwards the one comment that supports it. Being able to recognize a principle in an explanation is different from noticing it at work in yourself.
In a 2025 study published in PNAS, researchers tested AI assistance in high-school mathematics. Students with access to a relatively unrestricted GPT interface performed better while receiving help, but performed worse than the comparison group when that help was taken away. An alternative designed with tutoring safeguards largely mitigated the problem. The finding does not establish that reading books is better than using AI. It suggests something narrower and more useful: A tool that helps people complete a task can fail to build their ability to do it independently.
The evidence also cuts against a blanket dismissal of AI as a shortcut. In a separate randomized study, published in Scientific Reports in 2025, 194 college physics students learned through a purpose-built AI tutor or an active-learning class. The researchers found better immediate learning outcomes, in less time, with the AI tutor. That study did not compare a chatbot with a book, and it did not establish that the advantage would persist over years. Together, the studies suggest that the design of the interaction matters. AI can guide someone through thinking, or make it easier to avoid doing the thinking themselves.
Books offer no automatic protection against the second problem. A reader can finish 300 pages, underline several passages and remember very little. Time spent is not proof of learning. Difficulty is not always depth. Still, a well-constructed book can offer something that a sequence of questions may miss: an intellectual route chosen by someone who knows the territory.
Imagine asking why an empire collapsed. A summary of a history book might give you the author’s answer: Its institutions became too weak to withstand a crisis. That is an intelligible conclusion. But how did the author distinguish institutional weakness from military defeat, economic trouble or decisions made by particular leaders? In the full book, you might encounter records that contradict one another, explanations that initially look persuasive and evidence that changes their meaning. Those chapters give you a chance to examine the author’s judgment. They may also leave you less certain than the summary did, for good reason.
A summary can report the conclusion. It may not preserve enough of the reasoning for a reader to judge whether the conclusion was earned. AI can also organize a sustained course of inquiry. The distinction is not an absolute limit of the technology. It is a difference between ways of using it. Asking for the takeaway from a difficult argument is a different activity from working through that argument, whether on a page or in a conversation.
Original reporting presents another distinction. A historian who uncovers neglected correspondence or a journalist who persuades people to describe events they have never discussed publicly is producing material that did not previously exist in that form. Summarizing the resulting book can make its findings more accessible. It does not perform the work that made those findings available. Indeed, a reader could find the summary sufficient while still depending on the original book’s existence. Personal convenience and the value of producing knowledge are separate questions.
With fiction and memoir, the limits of the summary become easier to see. Imagine a novel about two lifelong friends. An AI summary tells you that one eventually betrays the other. You know the central event. But in the novel, you might spend a hundred pages watching them protect each other, laughing at their private jokes and overlooking the small resentments that later become consequential. When the betrayal comes, you remember the promises that preceded it. The summary gives you the event. The novel gives the event its weight. For a reader who came for that experience, the time spent getting there is part of the value.
That does not make literature immune to competition from AI-generated entertainment. Nor does it prove that human authorship will always command a premium. Those are predictions about readers’ preferences. The more modest point is that replacing a story’s information does not automatically replace the reasons someone enjoys reading it.
AI might also make a demanding book more approachable. Our reader could pick up the book about decisions, finish a chapter and ask the assistant to challenge their understanding: What would count as evidence against that business proposal? What criticism have they dismissed too quickly? Used this way, the technology becomes a companion to reading. Whether that improves understanding depends on the quality of the assistance and what the reader does with it.
The sensible response is to become more deliberate about what we want from a book. Sometimes we need a fact. Sometimes we need a framework. Sometimes we want to follow a particular mind through a problem, or inhabit a story that cannot be reduced to its lesson. AI makes the first two easier to obtain without reading an entire volume. It also gives authors a stronger reason to offer something beyond information that can be conveniently rearranged.
The book on the nightstand may deserve to remain unfinished. Or the conversation about it may reveal why it deserves closer attention. The useful question is what disappears when we skip it. If the answer is mostly repetition, little may be lost. If the answer is the evidence, the uncertainty, the voice or the experience that makes the conclusion matter, getting the point may be only the beginning.
Evidence & Source Transparency
Evidence First shows its work. The article ends above; this section is included so readers can inspect the main sources behind the factual claims.
The list below does not source every sentence. It focuses on the factual claims most important to the argument.
1. AI assistance and independent learning
Claim or topic:
Students using a relatively unrestricted GPT interface performed better with assistance but worse than the comparison group after assistance was removed. A version with tutoring safeguards largely mitigated that problem.
Source:
Bastani and colleagues, “Generative AI without guardrails can harm learning: Evidence from high school mathematics,” PNAS (2025).
Source type:
Academic research; randomized controlled study.
What it supports:
In the mathematics setting studied, better performance while using AI did not necessarily translate into better independent performance. The design of the assistance mattered.
Important caveat:
The study did not compare AI use with reading books. It does not establish that AI generally harms learning or that books produce better understanding.
2. Learning with a purpose-built AI tutor
Claim or topic:
In a randomized study involving 194 college physics students, a purpose-built AI tutor produced better immediate learning outcomes, in less time, than an active-learning class.
Source:
Kestin and colleagues, “AI tutoring outperforms in-class active learning: an RCT introducing a novel research-based design in an authentic educational setting,” Scientific Reports (2025).
Source type:
Academic research; randomized controlled study.
What it supports:
A deliberately designed AI tutor can support effective learning in a specific educational setting.
Important caveat:
The comparison involved a particular tutor and classroom setting, not ordinary chatbot use versus reading a book. It did not establish that the learning advantage would persist over years.
3. AI as a substitute for some practical reading
Claim or topic:
An interactive AI exchange may provide much of what a reader wants from a book that packages familiar information, such as advice on running meetings.
Source:
Source needed for a direct empirical comparison of AI assistance with full-book reading for these purposes.
Source type:
Author’s analysis; direct supporting evidence not cited.
What it supports:
The article presents this as a plausible substitution, illustrated through a hypothetical manager. Neither cited study directly tests that example or measures the value of business or self-help books.
Important caveat:
The article does not establish that AI has reduced book sales or prices. Its argument concerns possible usefulness to individual readers, not a demonstrated market effect.
4. What summaries may leave out
Claim or topic:
A summary may omit reasoning needed to evaluate an author’s conclusion or the narrative experience that gives a fictional event emotional weight. Summarizing original reporting also differs from producing it.
Source:
The article’s analysis and hypothetical examples. No external source is cited for these comparisons.
Source type:
Interpretation and illustrative reasoning.
What it supports:
The history-book and friendship examples explain distinctions between receiving a conclusion, examining its evidence and experiencing a story. They are illustrations, not research findings.
Important caveat:
The article does not empirically establish how much readers lose through summaries or whether full-book reading produces better outcomes. Source needed for any measured claim about those effects. The cited educational studies do not answer those questions.
How to read this evidence
This article is the author’s analysis. The sources above are provided so readers can see where the factual claims come from and judge the evidence for themselves. Some sources support direct facts, while others provide context, estimates, or background evidence.
Production transparency
Evidence First uses artificial intelligence extensively for research, analysis, drafting, and editing. AI may generate substantial portions of the written article. Human editorial judgment determines the questions investigated, evaluates the evidence and competing explanations, reviews important factual claims and sources, determines what conclusions the evidence supports, and approves the article for publication. AI-generated statements are not treated as evidence; conclusions must be supported by the cited sources.
Corrections and updates
If a factual error is identified, this post will be corrected in the web version with a dated note explaining the change. Because email versions cannot be edited after sending, the web version should be treated as the current version.