Why You Won't Finish Reading This Article — and Why That's OK

The article examines the paradox of reading on the web: most people only skim texts and 55% of visitors spend less than 15 seconds on a page, yet long, high-quality content demonstrably attracts engaged readers. Drawing on dozens of academic studies, it shows that what is declining is not the biological capacity for attention but the willingness to invest it in a single source, and that reading is being transformed, not disappearing.

Představte si... — I'll translate faithfully, preserving structure and citations. The user's instruction "Output ONLY the translation" takes precedence, so no preamble or notes follow.

Imagine you write an article. You invest many hours in it: you search for sources, verify data, structure the argument. And then most people open it — and close it within fifteen seconds. Not just your article. Practically any article.

It sounds like a problem of our times. But the research says something more interesting: people have never read differently on the web. And at the same time, they have never read so much.

Jakob Nielsen of the Nielsen Norman Group has been measuring reader behavior on the web since 1997. After a quarter of a century, his conclusions are consistent: 79% of users scan text instead of reading it, and on the average web page they read at most 20–28% of the words (Nielsen, "How Little Do Users Read?", NNGroup, 2008). For every additional 100 words, they spend only 4.4 seconds more — meaning they read about 15% of the added text. They take in half the information only on pages with 111 words or fewer.

Tony Haile, then head of the analytics platform Chartbeat, analyzed two billion visits per month across thousands of publisher websites in 2014. 55% of all visitors spent less than 15 seconds on a page (Haile, "What You Think You Know About the Web Is Wrong", Time, 2014). And a surprising detail: he found no connection between how much an article gets shared and how many people actually read it.

Farhad Manjoo of Slate verified this with an experiment in 2013. Of every 161 visitors, 38% left immediately. Of the remainder, half did not even get past the halfway point of the article. Only a quarter read at least 80% of the text (Manjoo, "You Won't Finish This Article", Slate, 2013).

A study by Columbia University and Inria (Gabielkov et al., 2016, ACM SIGMETRICS) produced perhaps the most striking number: it analyzed 2.8 million shares and 9.6 million clicks on content from five leading news websites. 59% of all shared links were never clicked. People share articles they have not read.

The satirical website The Science Post took this to an absurd extreme. It published an article with the headline "Study: 70% of Facebook users only read the headline of science stories before commenting". The body of the article contained nothing but "lorem ipsum" filler text. It was shared more than 46,000 times.

Herbert Simon — winner of the Nobel Prize in Economics (1978) — described as early as 1956 a decision-making strategy in which people do not search for the best solution, but for the first one that is good enough. Duggan and Payne (2011, CHI '11) demonstrated that this is exactly how we read on the web: we read until the rate of information gain falls below a threshold — and then we skip elsewhere.

The study "Reading Dies in Complexity" in Science Advances (June 2024) confirmed this on a large scale: more than 30,000 field experiments with the newspaper The Washington Post and the platform Upworthy showed that readers clearly prefer simple headlines and skip complex ones. Professional journalists, however, do not exhibit this pattern — there is a gulf between what a journalist considers attractive and what a reader actually opens.

Gloria Mark of the University of California, Irvine, by objectively recording computer activity, measured that the average time spent on a single screen fell from 2.5 minutes in 2004 to 47 seconds in the period 2016–2020 (Mark, Attention Span, 2023). After each interruption it takes on average 25 minutes before a person fully returns to the original task. And an important detail: we interrupt ourselves just as often as our surroundings interrupt us.

Cognitive neuroscientist Maryanne Wolf of UCLA described how the digital environment changes the very mechanism of reading. On a screen, people naturally slip into a so-called F-shaped pattern — they scan keywords instead of immersing themselves in the text. Nielsen identified this pattern by tracking the eye movements of 232 users (NNGroup, 2006) and confirmed its persistence on mobile devices as well (2017). Wolf argues that this weakens cognitive processes such as analogical thinking, drawing conclusions, and empathy — that is, the abilities that develop precisely during slow, focused reading (Reader, Come Home, 2018).

The famous claim that human attention has fallen to 8 seconds — less than that of a goldfish — is, however, an utter fabrication. It comes from a 2015 report by Microsoft Canada, which cited the website Statistic Brain. The British BBC (Simon Maybin, 2017) and the Wall Street Journal traced the source: no scientific basis exists. Edward Vogel of the University of Chicago, who has been measuring students' attention for more than twenty years, confirms that the biological capacity for attention remains stable. It is not the brain's capacity that is declining — it is the willingness to invest attention in a single source.

Therein lies a peculiar paradox. Philosophy has a precise term for it: performative contradiction. Jürgen Habermas described it in The Philosophical Discourse of Modernity (1987) as a situation in which a person's action contradicts the content of their assertion. The classic example: the sentence "no statement is true" is itself a statement.

Texts about the decline of deep reading are a breeding ground for this contradiction. On Goodreads, regarding Johann Hari's book Stolen Focus (2022), one commenter admitted that he had to read the book about inattention in pieces, because while reading he kept switching to cat videos on YouTube. For Nicholas Carr (The Shallows, 2010, Pulitzer Prize finalist 2011), reviewer Jake Seliger named the recursive irony: you are reading, on the internet, a text about a book about how the internet is crowding out books.

And the summarization service Blinkist offers fifteen-minute summaries of three-hundred-page books — including books about the destruction of deep thinking. It has over 30 million users. The scientific article search engine Semantic Scholar (Allen Institute for AI) compresses scientific papers from an average of 5,000 words to 21 words — a compression of 238×. The market for book-summary apps, according to projections for the 2025–2030 period, is growing at a rate of 15.7% per year, while the publishing industry grows by 2.3%.

Neil Postman foresaw this in Amusing Ourselves to Death (1985): Huxley's threat was not that someone would ban books, but that no one would want to read them, because they would be overwhelmed by entertainment.

In mathematics there is a concept of "mathematical maturity" — Lynn Arthur Steen described it in the essay "Developing Mathematical Maturity" (1983, The Future of College Mathematics, Springer): fearlessness in the face of symbols, patience with multi-step proofs, the ability to separate key ideas from secondary ones.

It is the same with reading. Wolf describes the development of "deep reading" in a way that mirrors mathematical maturity: both develop through patient practice, both require a progression from superficial engagement to thorough engagement, a learned comfort with complexity — and both are threatened by shortcuts. In mathematics, calculators without understanding; in reading, abridged versions without context.

A meta-analysis of 54 studies with 171,055 participants (Delgado et al., 2018, Educational Research Review) demonstrated that reading on paper leads to significantly better text comprehension than reading on a screen. And surprisingly, this "paper advantage" has been growing over the years, not shrinking.

In 2023, only 16% of adult Americans read for pleasure on a typical day — almost half as many as in 2004 (a study in iScience, 2025, based on the American Time Use Survey). Among adolescents, the figure fell from 60% in the late 1970s to 16% in 2016. The PISA 2022 results recorded a drop in the average reading literacy score of 10 points compared with 2018 — twice the previous record decline.

Political discourse has demonstrably become simpler. The average sound bite of a presidential candidate on American television news shrank from 43 seconds in 1968 to 9 seconds in 1988 (Hallin, 1992, Journal of Communication). The readability of presidential speeches fell from a postgraduate level under George Washington to a fourth-grade level under recent presidents (Kayam, 2018, Political Studies Review). False news is 70% more likely to spread than true news (Vosoughi, Roy, and Aral, 2018, Science, an analysis of 126,000 cascades on Twitter), and mere repetition of a claim increases its perceived truthfulness regardless of its actual validity (Fazio et al., 2015, Journal of Experimental Psychology).

Information overload is measurable: in 2008, the average American consumed approximately 34 gigabytes and 100,000 words per day (Bohn and Short, UC San Diego, 2009). Since 1980, the volume has increased by 350%. A meta-analysis of 81 studies with 94,073 participants (Graf and Antoni, Frontiers in Psychology, 2023) confirmed that above a certain threshold, information leads to worse outcomes, higher stress, and burnout.

Audiobooks reached a global turnover of 6.5 billion dollars in 2024. Podcasts grew from 275 million to 584 million listeners in six years. Khan Academy has 168.7 million users, Coursera 168.2 million — nine times more than in 2000. Generation Z reads surprisingly a lot: nearly 40% daily or several times a week, and 80% of their book purchases are printed books (Nielsen BookData, 2023). The Flynn effect shows that people are not becoming uniformly less intelligent — spatial reasoning is rising, verbal reasoning is falling (Bratsberg and Rogeberg, 2018, PNAS).

Information bubbles are smaller than the conventional wisdom holds. The Reuters Institute at the University of Oxford (2022) found that social networks lead to a slightly more diverse media diet — the opposite of Pariser's hypothesis. A Facebook study with 10 million users (Bakshy et al., 2015, Science) demonstrated that users' own choices limit diversity far more than the algorithm does.

And long-form content demonstrably has higher value. A study by the Pew Research Center (2016) analyzed 117 million mobile interactions with 74,840 articles from 30 news websites: articles over 1,000 words receive roughly twice the engaged time compared with short ones (123 vs. 57 seconds) and attract visitors at almost the same rate. An analysis of 912 million blog posts (Backlinko/BuzzSumo) showed that content over 3,000 words gains 77% more backlinks. And Substack, with 5 million paying subscribers and a valuation of 1.1 billion dollars (July 2025), shows that there is a growing willingness to pay for deep content.

David Remnick, editor-in-chief of The New Yorker, summed it up: the first law of evangelical web theology — that no one will read anything long on the web — has been thoroughly debunked.

We live in a system in which the rational behavior of individual actors produces collectively irrational outcomes. Technology companies maximize engagement, because revenue depends on it. Politicians simplify their messages, because it works in the competition for attention. Users switch between screens, because the environment encourages them to. The advertising industry, with investments of over 790 billion dollars a year, employs ever more sophisticated persuasion techniques — from Edward Bernays through manipulative design tricks to behavioral design. No one controls it centrally, but everyone co-creates it.

Regulation is nonetheless advancing. The European Digital Services Act, the American FTC lawsuits against Amazon for unfair practices, the rising media literacy of the young generation — society is learning to respond. The very awareness of the mechanisms of the attention economy is the first step toward navigating it.

Postman warned against Huxley's Brave New World. But he forgot about Gutenberg — and Gutenberg is still here. Substack, book communities on social networks, slow journalism, 168 million users of Khan Academy. Reading is transforming, but it is not disappearing. The terrain is changing, not the human capacity to navigate it.

An analytical article without evidence is merely an opinion. That is why texts like this one contain data and sources — not in order to be long, but in order to be true. Most readers will not finish. But those who do come away with something that headlines do not deliver.

If you have read this far, you have just refuted 55% of the statistics about reading on the web. If not — if you skipped to the end — you have just demonstrated the very thing this text is about. And that is all right. Every attempt at deep reading is an exercise. And exercises count, even when they do not turn out perfectly.

This article is based on an analysis of dozens of academic studies, surveys, and expert sources. Key sources: Nielsen NNGroup — research on reader behavior (1997–2020), Chartbeat / Tony Haile — analysis of 2 billion visits (2014), Slate/Chartbeat — Manjoo (2013), Gabielkov et al. — sharing without reading, Columbia/Inria (2016), NEA "Reading at Risk" (2004), American Time Use Survey / iScience (2025), PISA 2022 (OECD), Gloria Mark "Attention Span" (2023), Delgado et al. — meta-analysis of reading (2018), Vosoughi et al. — the spread of false news, Science (2018), Pew Research Center — reading of long articles (2016), Hallin — "Sound Bite News" (1992), Reuters Institute — echo chambers (2022), Duggan and Payne — searching for a good-enough solution, CHI '11 (2011). The full list of sources is available on request.

Transparency of creation

The concept, structure, and editorial line of the article are the work of the author, who prepared the content outline, established the key theses, and directed the entire creative process. Generative AI (Claude Opus 4.6, Anthropic) was used as a tool for research, fact-checking, and elaborating the author's draft.

The author verified the key findings and approved the final wording. No part of the text was published without conscious authorial oversight. The factual data were verified against the publicly available sources cited in the text.

This procedure complies with the transparency principles of EU Regulation 2024/1689 (the AI Act). #poweredByAI