Is long-form writing dying? Newsletter writers and some commentators have spent the past two years arguing the opposite, and the loudest version of that argument is a revival. The problem is that the strongest evidence on reading behaviour is more than a decade old, and most of the newer claims are forecasts or commentary.
The revival talk is loud, but the newest hard data is older
The current conversation centres on newsletters. A Columbia Journalism Review piece, written after doubts in 2022 about "peak newsletter", argues that newsletter momentum has returned. The excerpt we reviewed contains no new hard data, so treat it as informed commentary.
Nieman Lab's December 2024 article, 'The longform renaissance', reports that Substack passed 2 million paid subscribers in 2023 and that readers spend about 12 minutes per newsletter. We could not trace either figure to a primary Substack document, so read them as reported rather than confirmed. The same piece frames long-form as a possible renaissance, which is a forecast, not a measurement.
An Al Jazeera Institute piece links the revival to AI, describing it as both a possible catalyst and a disruptor. It offers no measurements. We could not verify any source showing that long-form readership grew or shrank overall in 2025 or 2026.
What is new and what is recycled
The genuinely recent signals concern supply and curation, not reader numbers. Longreads and The Sunday Long Read continue to curate long pieces, and The Sunday Long Read lists issues dated into October 2026. That shows editors are still choosing and sharing long work. It does not show how many people finish it.
Most of the hard figures people quote are older. The Pew analysis uses September 2015 data and was published in 2016. Statista's chart of time on publisher pages, drawn from Taboola data via eMarketer, covers 2017. The Nature-family study on word entropy is from 2024, but its attention-economy reading is the authors' interpretation of word-frequency trends, not a direct measure of attention.
What the length data actually shows
The best-documented evidence on length comes from Pew Research Center. Its analysis of 117 million cellphone interactions with 74,840 articles from 30 US news sites, from September 2015, found that articles of 1,000 words or more averaged about 123 seconds of engaged time, against 57 seconds for shorter pieces. The longer articles drew roughly the same number of visitors.
"Engaged time" is the important caveat. Pew measured scrolling, clicking and tapping. It did not measure comprehension or recall. A piece that keeps a thumb moving is not the same as one that is understood.
The advantage is not uniform. Pew also found that time with science and technology stories fell as length rose, so a rule that longer always holds attention would overstate the data.
Trade press coverage of Chartbeat data, reported by Digiday, points to a ceiling. The analysis drew on 200 randomly selected publisher sites from a base of 15,000 Chartbeat customers. It found that a standard-length article, about 4,000 pixels, gets about 40 percent less time than an article twice as long, but that the benefit tapers beyond that length. Chartbeat's rough conversion to 700 to 800 words is approximate, and we could not verify the full methodology.
Why averaged time-spent figures mislead
Averages can mislead too. Baekdal, an analyst's newsletter, argues that publishers' average time spent divides total pageviews into total time, which can make attention look worse than it is. Its numerical examples are illustrative rather than measured, but the arithmetic point is worth checking against any figure you are given.
The Statista chart shows the clearest decline in the record. It reports average time on a single publisher page falling from 2.09 minutes in Q1 2017 to 1.90 minutes in Q4 2017, citing Taboola data via eMarketer. The figure carries a caveat: users may be consuming more content inside platforms such as Facebook. We did not retrieve the primary report, so the chart does not measure total attention.
The most widely shared attention-span figures trace to low-quality pages or secondhand citations. We would not repeat them without a primary source.
What the evidence cannot yet tell us
Several gaps matter more than the headlines. The sources we reviewed include little primary reading-time data published after 2016. Nothing reliable measures how AI summaries and search answers affect long-form reading or traffic. And few sources compare print, book and newsletter audiences on the same metrics, so claims that one format suits long reading better are hard to test.
The Nature-family study adds a different kind of evidence. It argues that American English word entropy has risen since about 1900, and that short-form media show higher entropy than long-form, with social feeds higher still. That describes the language of different formats. It is not a measure of how long people stay with any piece.
What writers can reasonably change
The following is editorial judgement built on limited evidence, not proven results. Put the payoff in the first screen, because Pew's data cannot tell you whether a skimmer finishes. Use subheads that let readers scan and return. Match length to topic, bearing in mind Pew's finding on science and technology. Compare your engaged time against your own baseline rather than a published average.
The advice to produce fewer, better pieces comes from a 2013 opinion essay by Damien Walter. Weigh it as opinion. Your own analytics, read against the caveats above, will tell you more about your audience than any headline about attention spans.