Every few months, someone declares that nobody can read a book anymore. The evidence is less dramatic. The more useful question is whether we are losing the habit of doing the mental work that books demand.
AI did not create the problem. It makes the trade-off sharper. A good model can summarize a report, produce a first draft, and retrieve an argument in seconds. That is useful. It is also an invitation to stop before you have formed a view of your own.
Are people actually abandoning books?#
Not on the available evidence. Pew Research Center found that 75% of U.S. adults reported reading at least part of a book in the previous 12 months in its October 2025 survey. Print remains the most common format, with e-books and audiobooks adding rather than simply replacing reading for many people. Pew’s results are here.
That does not prove reading is healthy in every demographic or that long-form concentration is improving. It does puncture the lazy claim that books have become irrelevant.
Book sales are an even weaker proxy for attention. They measure purchases and format mix, not whether a reader finished a book, understood it, or changed a decision because of it. They can still tell us whether demand for books is resilient. They cannot tell us whether someone can hold an argument in their head long enough to challenge it.
The better measure is behavior. The U.S. Bureau of Labor Statistics reported that people aged 15 to 19 spent nine minutes a day reading for personal interest in 2024, while people aged 75 and over spent 46 minutes. The younger group spent 1.3 hours on games and leisure computer use. That is a genuine signal of competition for attention, but it is not proof that AI caused it. The BLS time-use data is worth reading directly.
What does AI change?#
AI changes the cost of producing an answer. It does not remove the cost of understanding the question.
That distinction matters in executive work. You can ask a model to summarize a market, compare competitors, or draft an investment memo. You cannot safely outsource the judgment about what is missing, which incentives matter, or what would break the conclusion.
That is the difference between discovery and knowledge. AI can reveal a field faster than any person can, but it cannot decide which finding matters to your business or carry the accountability for acting on it.
The early research supports caution, not panic. A 2025 CHI study of 319 knowledge workers found that people often reported less effort on cognitive tasks when using generative AI. The authors were explicit that their study did not establish causation. Their more useful observation was that the work shifts: from gathering information to verifying outputs, integrating responses, and supervising the tool. Read the paper.
That is the management problem. If AI removes the lower-value effort but we keep the verification and synthesis, it can make teams stronger. If it removes the struggle before anyone develops a point of view, it can make teams faster and shallower at the same time.
Forming that point of view requires choosing the right level of abstraction, not processing every available detail. That is the practical value of strategic forgetting and world models for leaders using AI.
Why is attention still a leadership issue?#
Reading is not morally superior to every other format. A strong audiobook, a serious conversation, or a well-built model can deepen understanding. The advantage of a book is structural: it forces you to stay with an argument for longer than a feed, a clip, or a generated summary usually does.
That matters when you run a company. Most strategic errors are not caused by a lack of information. They happen because the team has not separated signal from noise, challenged the default assumption, or stayed with a difficult question long enough to see the second-order effect.
AI can make this worse when it becomes the first reader of everything. The model gives you a clean answer before you have encountered the source material. You inherit its framing, its omissions, and its confidence level without noticing.
How should leaders use AI without outsourcing judgment?#
I would set three operating rules.
- Read before you prompt on the work that matters. For a board decision, investment thesis, regulatory change, or strategic partnership, start with the primary document. Use AI after you have a provisional view.
- Make verification visible. Require a source list, a counterargument, and a statement of uncertainty in AI-assisted work. “The model said so” is not analysis.
- Protect uninterrupted depth. Give people time to read and write without the feed, chat, or model in the loop. This is not nostalgia. It is how you preserve the ability to form an original position.
That discipline needs an operating design, not a calendar slogan. Architecture of Attention makes the case for treating information synthesis and focused analysis as deliberate capabilities.
The goal is not to make teams slower. It is to make sure speed does not hide a collapse in comprehension.
What should we research next?#
The claim that AI is ruining attention is ahead of the evidence. Better research would track the same people over time, distinguish passive consumption from deliberate AI use, and measure comprehension and decision quality rather than self-reported screen time.
We also need better organizational data. Do teams that use AI heavily produce stronger decisions after six or twelve months? Do junior operators build domain judgment faster when AI drafts the first answer, or slower? Which review practices prevent overreliance? Those are the questions worth answering.
Books are not dying. But the discipline that books represent can atrophy if we let every difficult passage become somebody else’s summary. Use AI to widen the field. Do not let it do your thinking for you.
Sources#
- Bureau of Labor Statistics, American Time Use Survey: 2024 Results
- Pew Research Center, Americans still opt for print books over digital or audio versions
- CHI 2025, The Impact of Generative AI on Critical Thinking
- Bonifacci et al., The relationship between mind wandering and reading comprehension: A meta-analysis
Frequently Asked Questions
? Is AI causing people to lose their attention spans?
? Are book sales a good measure of attention?
? What is a safe way for executives to use AI for research?
Read Next
