Attention on a web page is front-loaded, not fold-bounded, and the difference matters because most homepage arguments are about the wrong boundary. Eye-tracking of more than 130,000 fixations from 120 participants found 42 percent of viewing time in the top 20 percent of the page and 74 percent within the first two screenfuls. The fold is not a wall. It is a slope.

The same body of research contains a second useful correction: the share of time spent above the fold fell from 80 percent in 2010 to 57 percent in 2018. People scroll more than they did. They still start at the top.

What follows separates the measured figures from the interpretations built on them, because the most repeated homepage numbers are the latter.

What was actually measured about attention

Three studies, each with a stated method, and one widely quoted number that belongs to none of them.

Dwell time, from browser logs. An analysis of log data from a browser plug-in covered 205,873 addresses, “each of which is accompanied by at least 10,000 dwell time observations.” The headline: “the dwell time is no more than 70 seconds on 80% of the 205,873 pages.”

And the shape of the curve is the interesting part. The distribution parameter was below one on 98.5 percent of pages, meaning “Web browsing exhibits a strong ‘negative aging’ phenomenon, that is, some ‘screening’ is carried out at the early stage of browsing a page, and the rate of subsequent abandonment decreases over time.” The authors’ summary: users “adopt a ‘screen-and-glean’ type browsing behavior.”

One sampling caveat worth stating. Only pages with at least 10,000 visits in two weeks qualified. This is a set of high-traffic pages, not the web at large.

The ten-second figure is an interpretation, not a measurement. A 2011 article reading that curve concluded that “the first 10 seconds of the page visit are critical for users’ decision to stay or leave”, and that “only after people have stayed on a page for about 30 seconds does the curve become relatively flat.” Useful, and attributable to that reading rather than to the underlying study.

Independent corroboration, with its own caveat. An analysis of 2 billion visits across 2,000 sites and 580,000 articles reported that “a stunning 55% spent fewer than 15 seconds actively on a page.” That is company-published, using a proprietary definition of active time, and it is not peer reviewed. It points the same way.

And one finding from it that should temper any content strategy. “There is no relationship whatsoever between the amount a piece of content is shared and the amount of attention an average reader will give that content.”

Measured distribution of page viewing time by vertical position and the change over eight yearsThe measured distribution of page viewing time by vertical position on a page, drawn from an eye tracking analysis of over one hundred and thirty thousand eye fixations recorded from one hundred and twenty participants across thousands of sites from a wide range of sectors and industries, on a screen resolution of one thousand nine hundred and twenty by one thousand and eighty pixels. Four proportions were measured. More than forty two percent of viewing time fell within the top twenty percent of the page. More than sixty five percent of viewing time fell within the top forty percent of the page. Fifty seven percent of page viewing time fell above the fold. Seventy four percent of viewing time fell within the first two screenfuls, up to two thousand one hundred and sixty pixels. A longitudinal comparison from the same source records that in 2010 eighty percent of viewing time was spent above the fold, so that the proportion fell from eighty percent to fifty seven percent over eight years, indicating that the fold constrains attention less than it formerly did while attention remains concentrated toward the top of the page. Separately, an analysis of browser log data covering two hundred and five thousand eight hundred and seventy three addresses, each carrying at least ten thousand dwell time observations, found that dwell time is no more than seventy seconds on eighty percent of those pages, and that the distribution shape parameter fell below one on ninety eight point five percent of them, which the authors interpret as strong negative aging in which screening is carried out early in the visit and the rate of subsequent abandonment decreases with time, summarised as a screen and glean browsing behaviour. A sampling caveat applies to that dataset, since only pages receiving at least ten thousand visits within a two week window qualified for inclusion, making it a set of high traffic pages rather than a sample of the web generally.Where the viewing time actually goesTop 20% of the page42%Top 40% of the page65%Above the fold57%First two screenfuls74%130,000+ fixations, 120 participants, 1920 by 1080 screen.The eight-year changeAbove the fold: 80% in 2010, 57% in 2018.People scroll more than they used to.And how long they stay”No more than 70 seconds on 80% of the205,873 pages.”The right mental model is a slope, not a wallAttention decays with vertical position rather than stopping at a line. Which means the question is not what fitsabove the fold, but what deserves the top fifth of the page.
Four measured proportions from one eye-tracking dataset, and the eight-year change in the headline figure. Source : Nielsen Norman Group, Scrolling and Attention, 2018, and Liu, White and Dumais, SIGIR 2010 (2018)
Three grades of evidence behind commonly cited homepage attention claimsThree distinct grades of evidence underlying commonly cited claims about attention on web pages, which are frequently conflated and which support different strengths of assertion. The first grade is measured data, exemplified by the analysis of browser log data covering two hundred and five thousand eight hundred and seventy three addresses each carrying at least ten thousand dwell time observations, which found that dwell time is no more than seventy seconds on eighty percent of those pages and that the shape parameter fell below one on ninety eight point five percent of them, and by the eye tracking analysis of over one hundred and thirty thousand fixations from one hundred and twenty participants which produced the proportions of viewing time by page position. Both have published methods, stated sample sizes and identifiable sampling limitations, and support direct assertions of fact. The second grade is interpretation of measured data, exemplified by the widely quoted claim that the first ten seconds of a page visit are critical to the decision to stay or leave, and that the curve becomes relatively flat only after about thirty seconds. These statements come from a 2011 reading of the 2010 dwell time curve rather than from the underlying paper, are reasonable inferences, and should be attributed to the person who made the inference rather than presented as findings of the original study. The third grade is self reported industry data, exemplified by an analysis of two billion visits across two thousand sites and five hundred and eighty thousand articles reporting that fifty five percent of visitors spent fewer than fifteen seconds actively on a page, and that twenty four percent of visitors scrolled at all on native advertising against seventy one percent on normal editorial content. This is company published rather than peer reviewed, relies on a proprietary definition of actively engaged time, and should be cited with both qualifications attached. All three grades are usable in an argument, but conflating them is how an interpretation of a curve becomes presented as research.Three grades of evidence, all usableMeasured”No more than 70 seconds on 80% of the 205,873 pages.” 130,000+ fixations from 120 participants.Published method, stated sample, known sampling limits. Assert directly.Interpreted”The first 10 seconds of the page visit are critical for users’ decision to stay or leave.”A 2011 reading of the 2010 curve. Reasonable. Attribute to the reader, not the study.Self-reported industry data”A stunning 55% spent fewer than 15 seconds actively on a page.” 2 billion visits, 2,000 sites.Company published, proprietary metric, not peer reviewed. Cite with both caveats.Conflating them is how a reading of a curve becomes “research shows”.
All three are usable. Confusing them is how a reading of a curve becomes 'research shows'. Source : Liu, White and Dumais 2010; Nielsen 2011; Chartbeat via Time, 2014 (2014)

The F-pattern is a symptom, not a law

This is the most misapplied finding in web design, and its own publisher issued the correction.

The original observation, from 2006. Recording “how 232 users looked at thousands of Web pages”, three movements were described: a horizontal sweep across the upper content area, a shorter second horizontal movement lower down, and a vertical scan down the left side.

The correction, from the same organisation in 2017. The pattern appears only when three conditions hold together: the page “includes text that has little or no formatting for the web. For example, it has a ‘wall of text’ but no bolding, bullets, or subheadings”; the user “is trying to be most efficient on that page”; and the user “is not so committed or interested that he is willing to read every word.”

And the sentence that reframes it entirely. “The F-pattern is the default pattern when there are no strong cues to attract the eyes towards meaningful information.”

With a verdict attached. “The F-shaped scanning pattern is bad for users and businesses: it means that users may skip important content simply because it appears on the right side of the page.”

Five other patterns are documented. Layer-cake, where “the eyes scan headings and subheadings and skip the normal text below”. Spotted, marking, bypassing, and commitment, the last being “fixating on almost everything on the page.”

Which changes what to do about it. Designing for the F-pattern accepts the failure. Formatting the page so that headings, bold text and lists carry the meaning produces the layer-cake pattern instead, which is the one you want.

One honesty note on the sample. The 2017 research reports per-example figures such as “more than 45 people” and “47 people”, without a total for the programme. Treat the pattern taxonomy as well-observed rather than as precisely quantified.

Conditions producing the F shaped reading pattern and the alternative patterns documented alongside itThe three conditions that together produce the reading pattern named after the letter F, and the five alternative scanning patterns documented alongside it by the same research organisation which originally described the F pattern in 2006 and issued a substantial clarification in 2017. The first condition is that a page or a section of a page includes text with little or no formatting for the web, the example given being a wall of text containing no bolding, bullet points or subheadings. The second is that the user is trying to be most efficient on that page. The third is that the user is not so committed or interested as to be willing to read every word. All three must hold. The organisation’s own reframing states that the F pattern is the default pattern when there are no strong cues to attract the eyes towards meaningful information, and its verdict is that the F shaped scanning pattern is bad for users and businesses because it means users may skip important content simply because it appears on the right side of the page. Five alternative patterns are documented. The layer cake pattern occurs where the eyes scan headings and subheadings and skip the normal text below them, and is the pattern produced by well formatted content. The spotted pattern involves skipping large amounts of text to look for something specific. The marking pattern involves keeping the eyes in one place while scrolling. The bypassing pattern involves deliberately skipping the first words of a line where multiple lines begin identically. The commitment pattern involves fixating on almost everything on the page. The design implication is that designing for the F pattern accepts the failure it represents, whereas formatting a page so that headings, bold text and lists carry the meaning produces the layer cake pattern instead. A sampling note applies to the 2017 work, which reports per example participant figures such as more than forty five people and forty seven people without stating a total for the research programme.Three conditions, and five other patternsThe F appears only when all three hold1. “A ‘wall of text’ but no bolding, bullets, or subheadings”2. “The user is trying to be most efficient on that page”3. “Not so committed or interested that he is willing to read every word”Which is why it is a symptom”The F-pattern is the default pattern when there are no strong cues to attract the eyes towards meaningful information.”The pattern you want insteadLayer-cake: “the eyes scan headings andsubheadings and skip the normal text below.”Produced by formatting, not by layout tricks.And four more, documentedSpotted, marking, bypassing, and commitment,the last being “fixating on almost everything”.Designing for the F accepts the failure. Formatting the page removes the condition that causes it.
The F appears when nothing on the page competes for the eye. Formatting produces the layer-cake instead. Source : Nielsen Norman Group, F-Shaped Pattern of Reading on the Web, 2017 (2017)

The B2B number everyone cites, and its date

There is one widely quoted study on B2B usability, and citing it responsibly means citing when it happened.

The method. Twelve focus groups, 55 one-to-one testing sessions and seven field studies, with 79 participants in total, testing 179 B2B websites.

The headline. “B2B sites earned a mere 58% success rate”, against “mainstream websites” at 66 percent.

And the finding that has aged best. “The most user-hostile element of most B2B sites is a complete lack of pricing information.” Users ranked pricing first among 28 information types, 29 percent above the second.

The date is 2006. That is worth saying out loud whenever the 58 percent figure appears, because it is usually presented as current.

A newer edition exists, with a larger site base. “Representative business users tested 293 B2B websites” across the US, UK and Singapore, producing 188 guidelines. The participant count for that edition is not published, and the report is paid.

So the honest position on the pricing question. It is the best-attested complaint in B2B usability research, and the research is twenty years old. Treat it as a strong prior rather than as a current measurement, and test it on your own traffic.

Methodology and date of the most frequently cited business to business usability researchThe methodology, results and publication date of the most frequently cited research on business to business website usability, together with the current edition of the same report. The original study combined three methods, comprising twelve focus groups, fifty five one to one usability testing sessions and seven field studies, with a combined study group of seventy nine participants of whom thirty nine were male and forty female, and with sessions conducted mainly in the United States including California, Washington and Arizona, and some in the United Kingdom. Participants tested one hundred and seventy nine business to business websites. The headline result was that business to business sites earned a fifty eight percent task success rate, compared with a substantially higher sixty six percent success rate recorded for mainstream websites. The finding most frequently quoted from the study concerns pricing, the report stating that the most user hostile element of most business to business sites is a complete lack of pricing information, and recording that users ranked pricing highest among twenty eight information types, by a margin of twenty nine percent above product availability in second place. The critical qualification is the publication date of the thirty first of May 2006, which is rarely stated when the fifty eight percent figure is quoted and which makes the research approximately twenty years old. A current third edition of the same report exists, running to four hundred and nineteen pages and containing one hundred and eighty eight design guidelines together with a checklist of one hundred and eighty one, based on representative business users testing two hundred and ninety three business to business websites across the United States, the United Kingdom and Singapore, using usability testing sessions, focus groups, site visits, longitudinal diary studies and heuristic design reviews. The participant count for that edition is not published on its public page and the report itself is paid, so the only publicly available participant figure for business to business usability research remains the seventy nine participants of the 2006 study.The 58 percent figure, and what is behind itThe method12 focus groups55 one-to-one testing sessions7 field studies79 participants, 179 B2B websitesThe results58%task success on B2B sites66%on mainstream sitesPricing ranked first of 28 information types.And the part that is usually left outIt was published on 31 May 2006. The figure is twenty years old and is routinely quoted as current.How to use it honestlyThe pricing complaint is the best-attested finding in B2B usability research, and the research is two decades old.Treat it as a strong prior worth testing on your own traffic, not as a current measurement of your audience.
Three focus groups, 55 sessions, 79 people, 179 sites. And it was published in 2006. Source : Nielsen Norman Group, B2B Usability, 31 May 2006, and the current third edition report (2006)

What that means for the page

Five decisions follow from the measured material, and none of them is about the fold. They are the ones we work through first when a B2B homepage is rebuilt around what it must say in its top fifth.

Spend the top fifth on what you do and who for. That is where 42 percent of viewing time falls. A rotating banner spends it on nothing.

Assume screening, not reading. The abandonment curve is steepest early, and the described behaviour is “screen-and-glean”. The first screenful is answering a filtering question, not delivering a message.

Format so that headings carry the meaning. The layer-cake pattern is available to any page whose subheadings say something. The F-pattern is what happens when they do not.

Put something about price on the page. Not necessarily a number. A range, a starting point, a “typical engagement” figure, or an explicit statement of why there is none. The complaint in the research is about absence, not about the number.

And design for the second screenful too. 74 percent of viewing time falls within the first two, which means the second is worth roughly as much as the fold argument you were about to have.

One thing to stop doing. Optimising for what fits above a line at a particular resolution. The measured distribution is a gradient, and the line has moved by 23 points in eight years.

Five homepage design decisions supported by the measured attention and usability researchFive design decisions for a business to business homepage which follow from measured research rather than from convention, none of which concerns the position of the fold. The first is to spend the top fifth of the page on what the organisation does and who it does it for, since more than forty two percent of measured viewing time falls within the top twenty percent of the page, and a rotating banner occupying that space delivers no fixed message. The second is to assume screening rather than reading, since the measured abandonment curve is steepest at the start of a visit and the behaviour described by the researchers is screening and gleaning, meaning the first screenful answers a filtering question rather than delivering a full message. The third is to format the page so that headings carry the meaning, since the layer cake scanning pattern in which the eyes scan headings and subheadings while skipping the body text is available to any page whose subheadings are substantive, whereas the F shaped pattern is what occurs in the absence of such cues. The fourth is to place some information about price on the page, not necessarily a specific figure but a range, a starting point, a typical engagement value or an explicit statement of why none is given, because the most cited business to business usability study identifies the complete lack of pricing information as the most user hostile element of such sites and records users ranking pricing first among twenty eight information types by a margin of twenty nine percent, while noting that this study dates from 2006. The fifth is to design for the second screenful as well as the first, since seventy four percent of measured viewing time falls within the first two screenfuls up to two thousand one hundred and sixty pixels. The corresponding practice to abandon is optimising for what fits above a line at a particular screen resolution, since the measured distribution is a gradient and the above the fold share fell from eighty percent to fifty seven percent across eight years.Five decisions, none about the fold1. Spend the top fifth on what you do and who for42% of viewing time lands there. A rotating banner spends it on nothing.2. Assume screening, not readingThe first screenful answers a filtering question. It does not deliver the message.3. Make the headings carry the meaningThat produces the layer-cake pattern. The F-pattern is what happens without it.4. Put something about price on the pageA range, a starting point, or a reason there is none. The complaint is about absence.5. Design for the second screenful too74% of viewing time falls in the first two, which is most of the fold argument settled.Stop optimising for a line at one resolution. The distribution is a gradient, and it moved 23 points in eight years.
None of them is about the fold. Four are about the top fifth of the page and what the headings say. Source : Method, over the attention, reading pattern and B2B usability research cited above (2026)

What to do with this

Open your homepage and measure the top 20 percent of the page. Write down what occupies it. If the answer is a carousel, a stock photograph or a slogan that could belong to any company in your category, you have spent the most valuable space you have on nothing.

Then read only your headings, in order, with the body text hidden. If that sequence does not explain what you do, who for, and what to do next, your page will produce the F-pattern regardless of how it is laid out.

Add whatever you can honestly say about price. If there is a genuine reason you cannot publish a number, publish the reason. The research complaint is about the silence rather than the figure.

And check the second screenful as carefully as the first, because three quarters of attention falls in the two together and only one of them usually gets reviewed.

The related pieces are B2B website usability that is testable and landing page structures by objective.