Reading

What the data actually says about the oldest art of consumption: whether reading is rising or falling, what is happening to book sales, how unequally reading is distributed, and what the research genuinely proves — as opposed to what gets claimed — about reading and cognition.

The facts at a glance

Is reading up or down?

Down, and the best dataset on it is unusually good. A 2025 study in iScience by researchers at the University of Florida and University College London analyzed twenty years of the American Time Use Survey — over 236,000 respondents between 2003 and 2023 — and found that the share of Americans reading for pleasure on an average day fell from about 28% in 2004 to about 16% in 2023, a decline of roughly 3% per year and more than 40% overall. The drop was not uniform: gaps widened by race, education, and income, meaning reading declined fastest among groups that already read least.

Survey self-reports agree. Gallup's long-running question found Americans read 12.6 books in 2021, the lowest since the series began in 1990, down from 15.6 in 2016 and 18.5 in 1999. Notably, the share reading zero books held steady at about 17% — the decline came from the formerly heavy readers: college graduates dropped from an average of about 21 books a year across 2002–2016 to 14.6.

The next generation reads less still. The UK National Literacy Trust's 2025 survey of nearly 115,000 children found just 32.7% of 8-to-18-year-olds enjoy reading in their free time — the lowest in the survey's twenty years, down 36% from 2005. In France, the CNL–Ipsos barometer of April 2025 found books read per year fell from 22 in 2023 to 18 in 2025, daily reading at its lowest in a decade, and 7-to-19-year-olds reading about 19 minutes a day against roughly ten times that in front of screens — with 16-to-19-year-old boys at 7 minutes a day.

Book sales: the paradox

Here is the puzzle: reading time collapsed, but book sales didn't. US print unit sales ran about 694 million in 2019, spiked to nearly 840 million in the 2021 pandemic surge, and have settled in the high-700-millions since — essentially flat year over year, and still comfortably above pre-pandemic levels, according to Circana BookScan. Audiobooks have meanwhile been the industry's fastest-growing format for a decade, with year-after-year double-digit revenue growth for much of that period.

The resolution of the paradox is that buying and reading are different acts performed by different populations. Sales are sustained by a small, stable base of heavy buyers — and by the familiar gap between books purchased and books finished — while the marginal, occasional reader who drove the higher participation numbers of the early 2000s has largely reallocated that time to screens, the dynamic covered in the attention economy.

Is it concentrated in a few people?

Extremely, on both sides of the market. On the demand side, reading follows the same power-law shape as most cultural consumption: the roughly 4% of Americans who read 50 or more books a year account for close to half of all books read, and the top fifth of readers for the large majority. Gallup's averages are pulled up by this tail — the median American reads far fewer than the 12.6-book mean. Industry analyses consistently find women make around 80% of fiction purchases.

On the supply side, concentration is starker. Data disclosed around the 2022 Penguin Random House–Simon & Schuster antitrust trial, covering 45,571 frontlist titles from the top ten US publishers over 52 weeks, showed about 66% sold fewer than 1,000 copies and fewer than 2% sold more than 50,000. The viral version of this claim — “half of all books sell fewer than a dozen copies” — is wrong (the real figure for under-12-copy frontlist titles was about 15%), but the corrected picture is still one where a thin sliver of blockbusters finances everything else.

What reading provably does to cognition

The honest answer requires sorting the evidence by strength. The best-established finding is the “upward spiral”: Mol and Bus's 2011 meta-analysis of 99 studies found moderate-to-strong correlations between print exposure and vocabulary, comprehension, and technical reading skill from infancy through university, with better readers reading more and thereby improving faster — the Matthew effect Keith Stanovich described, where early advantages compound.

The experimental evidence on fiction and social cognition is real but small. Dodell-Feder and Tamir's 2018 meta-analysis of experimental studies found reading fiction produces a small but reliable improvement in social-cognitive performance (g = 0.15); Mumper and Gerrig's 2017 meta-analysis of correlational studies found lifetime fiction readers score modestly higher on empathy and theory-of-mind measures than nonfiction readers. The famous 2013 Kidd and Castano result — that a few minutes of literary fiction measurably improves theory of mind — largely failed direct replication in 2016 and should be treated as fragile.

The longitudinal evidence is striking but correlational. A Yale study following 3,635 adults over 50 for twelve years (Bavishi, Slade, and Levy, 2016) found book readers had a 20% lower mortality risk and a roughly 23-month survival advantage after adjusting for age, education, wealth, health, and depression — with a dose-response pattern, and a stronger effect for books than for newspapers or magazines. Wilson and colleagues' 2013 study in Neurology, which followed older adults to autopsy, found lifelong cognitive activity like reading was associated with slower late-life cognitive decline even after accounting for the plaques and tangles found in their brains. And at the mechanistic level, Dehaene and colleagues showed in Science (2010) that learning to read physically reorganizes the visual cortex — literacy is one of the clearest demonstrations that a cultural practice rewires the brain.

Critique — how to hold these numbers

• Time-use surveys undercount new formats. The American Time Use Survey records a primary activity: an audiobook during a commute or reading woven into phone use is easy to miss, so part of the measured decline may be substitution the instrument can't see. The size of the collapse, however, is too large to be explained away by this alone.
• Sales are not reading. Unit sales count transactions, not attention — gift purchases, unread bestsellers, and the tsundoku pile all inflate the sales figures relative to actual reading, which is exactly why flat sales can coexist with collapsing reading time.
• Correlation still isn't causation. The longevity and cognitive-decline findings adjust for many confounders but cannot randomize who reads; healthier, wealthier, more educated people read more to begin with. The only truly experimental results — fiction and social cognition — show small effects, and the flashiest single-session finding failed replication.
• Averages hide the distribution. With consumption this concentrated, a mean like “12.6 books a year” describes almost nobody: the modal American reads close to zero and a small tail reads enormously, so aggregate statistics move when the tail moves.

Impact

Put together, the data describes a practice becoming a minority pursuit: participation falling for twenty years straight, a market held up by a small base of heavy buyers, and benefits — vocabulary, comprehension, some social cognition, plausibly slower cognitive aging — that accrue precisely to the shrinking group that still reads. Since the strongest finding in the literature is the upward spiral, in which reading skill and reading volume feed each other, a society that reads less doesn't decline uniformly: it splits into compounding readers and compounding non-readers. The forces pulling on the other end of that spiral are the subject of Brain rot and Doomscrolling.