Cognitive accessibility: designing for attention and mental load

In 2004, a person stayed on a screen for an average of two and a half minutes before switching to something else. Since 2016, the same measurement gives 47 seconds. That is what Gloria Mark, a researcher at the University of California (opens in a new tab), documented over twenty years by logging the actual activity of knowledge workers on their computers rather than what they reported. And during those 47 seconds, working memory, the one that keeps a piece of information in mind just long enough to use it, holds about four items at once, four menu entries or four form fields, according to Nelson Cowan's synthesis (opens in a new tab) that corrected Miller's famous "seven plus or minus two".

That is the budget a website has to make itself understood. An intro animation, a menu with twelve entries, a form that asks again for an address already typed, a banner that moves while someone is reading, each of these takes a share. When the budget runs out, the visitor does not complain. They close the tab.

Web accessibility was built around contrast, keyboards and screen readers. Those are necessary topics, covered here for interactions (opens in a new tab) and there for their economic weight (opens in a new tab). Cognitive accessibility, the part that concerns attention, memory and comprehension, remains the least tooled and the least tested. It is nevertheless the one that affects the most people, including visitors with no recognised disability at all.

Population figures give a first scale. The French national health institute's collective expertise on learning disorders puts the prevalence of dyslexia in France between 6 and 8% (opens in a new tab). Two epidemiological studies cited in the same chapter note that 7% of 18 to 29 year olds have serious or significant reading difficulties. Add the people living with an attention disorder, a brain injury, early dementia or simply the effect of age on processing speed. None of them show up in a contrast test.

What these people experience on an interface has been described precisely. In a focus-group study published in PLOS One in 2024 (opens in a new tab), neurodivergent and neurotypical students compared their experience of online platforms. The difficulties they raised almost never concern the content itself but its presentation. Material that is too dense, navigation whose structure cannot be guessed, a list of resources handed over with no indication of where to start. The authors conclude that this extraneous load depletes working memory before the actual difficulty of the task is even reached.

The second scale is wider still. Microsoft formalised in its inclusive design guide (opens in a new tab) the idea that a given limitation exists in permanent, temporary and situational forms. For cognition, the permanent form is the diagnosed disorder. The temporary form is the sleepless night, the concussion, the medical treatment, the bereavement. The situational form is the parent filling in a form with one hand while the other holds a child, the sales rep checking your pricing page between two meetings, the patient looking for a practice in a noisy street. Over the course of a day, almost everyone goes through at least one of these. Designing for limited attention means designing for the actual state of most visits.

The theoretical framework dates from 1988. John Sweller (opens in a new tab) showed that the processing capacity taken up by the form of a problem is no longer available for the problem itself. He distinguishes intrinsic load, inherent to what has to be understood, from extraneous load, produced by the way it is presented. On a website, intrinsic load is the price of a service, the difference between two offers, the steps to follow. Extraneous load is everything else. It is the only one design has any grip on.

Three long-standing measurements remain the best benchmarks for estimating it. The first is Cowan's four-item capacity, cited above. A navigation bar with six top-level entries already exceeds what most visitors can hold in mind at once. The second comes from Jakob Nielsen's analysis (opens in a new tab) of data from Weinreich and colleagues, who logged 45,237 page views in natural conditions. Time spent on a page grows by 4.4 seconds for every 100 words added, which leaves time to read at most 28% of the words on an average page, more realistically 20%. The text a visitor does not read still costs attention, since it has to be scanned to be dismissed.

The third measurement deserves to be quoted with its caveat. Iyengar and Lepper's 2000 jam experiment, where a display of six jars sold ten times better than a display of twenty-four, fed fifteen years of talk about the "paradox of choice". The meta-analysis by Scheibehenne, Greifeneder and Todd (opens in a new tab) across 50 experiments brought the average effect down to zero. What survived is more useful than the myth. Choice overload appears when the options look alike, when the stakes feel high and when the person choosing is not an expert in the field. That is exactly the situation of a prospect on a services page. They do not know your trade, they are comparing offers they can barely tell apart and they are committing money.

The WebAIM Million report, which analyses the million most visited home pages every year, gives a measure of the general direction. In February 2026, a home page contained an average of 1,437 elements (opens in a new tab), 22.5% more than the previous year and nearly double the 2019 figure. In the same report, 95.9% of pages have at least one detectable WCAG error, up for the first time in seven years. 15.2% contain links whose label does not say where they lead. Pages are getting heavier to understand, not just heavier to load.

The table below groups the most frequent drains and what the data says about each.

Each source of extraneous load has a documented cost and a known remedy

Two rows of this table deserve a note. The carousel figures come from the statistics published by Erik Runyon (opens in a new tab), who instrumented the home pages of several university sites over three months. They date from 2013 and nothing since has shown the opposite. An automatic carousel imposes a rhythm on the visitor, moves content while they are reading it and hides what comes next. For someone whose attention is easily drawn away, it is the worst of the three. The form figures come from Baymard's synthesis (opens in a new tab) on checkout flows, where the gap between the number of fields shown and the number actually needed is the first avoidable cause of abandonment. The last metre of the journey (opens in a new tab) has its own article.

Version 2.2 of WCAG, published in October 2023, introduced for the first time criteria that explicitly target memory and attention rather than perception. Three of them apply directly to a showcase or services website.

These three criteria share one thing. They do not ask for a simpler website. They ask the website to take over what the visitor should not have to hold in memory. That is a working definition of cognitive accessibility that holds well beyond the legal scope.

The fuller reference framework is "Making Content Usable for People with Cognitive and Learning Disabilities" (opens in a new tab), a W3C note published in 2021 by the COGA task force. It organises the question into eight objectives, including "help users focus", "ensure processes do not rely on memory" and "help users avoid mistakes and know how to correct them". Its design patterns are concrete. Limit interruptions, keep critical paths short, announce fees at the start of a task, never move a control unexpectedly, let users go back, avoid data loss when a session expires. None of these patterns is technical. All of them are design decisions.

What comes next is being drafted. The WCAG 3.0 working draft (opens in a new tab), published on 3 March 2026, contains requirements titled "avoid exclusionary cognitive tasks", "adequate time", "clear language" and "process and task completion". The document will carry no normative weight for several years. The direction is set all the same. Cognition is treated there as a functional need in its own right, on the same level as sight or motor control.

Since 28 June 2025, the French law of 9 March 2023 transposing the European Accessibility Act extends the obligation to private consumer services in five sectors, including e-commerce and banking. Temesis's summary (opens in a new tab) details the arrangement. Online services are checked by the consumer protection authority (DGCCRF), the fine is 7,500 euros per breach for a company, 15,000 for repeat offences, with a possible penalty of up to 3,000 euros per day. Companies with fewer than ten employees and under two million euros in turnover are exempt.

The French reference standard, the RGAA, is at version 4.1.2. The government's accessibility site (opens in a new tab) announces version 5 for the end of 2026, adding that work in progress must under no circumstances be suspended in the meantime. For a practice, a tradesperson or a services SME, the legal obligation most often does not apply yet. The criteria, however, describe what makes a prospect finish or abandon their quote request. The regulatory calendar is one more reason, not the main one.

The lazy reading of everything above leads to a website in black text on a white background, with no motion and no relief. That is a reasoning error. Sweller does not say all load is bad. He says extraneous load, the kind that does not serve comprehension, competes with the useful kind. An interaction that helps orientation reduces load. An interaction that decorates increases it. The question to ask of every animation, transition and effect is not "is this too much?" but "what does this teach the visitor?".

A transition that shows where a panel comes from and where it goes back to teaches the structure of the page. A hover that reveals what a link will open teaches the destination before the click. A progressive reveal that shows one idea at a time teaches the hierarchy of the argument. These movements work for working memory instead of competing with it. Measuring whether an animation does anything (opens in a new tab) is feasible with everyday tools.

A distinctive website and a cognitively accessible one meet on a single principle. A signature interaction is stronger when it stands alone. Three effects fighting for the same screen cancel each other out, for memorability as much as for attention. The website that picks one striking interaction per screen and leaves the rest still is both more recognisable and lighter to move through. Restraint is not a constraint imposed on distinctiveness. It is what makes it legible.

A few principles follow directly from the data cited.

No automated tool measures the cognitive load of a page. Validators detect contrast and missing attributes, not attention consumed. What follows is done by hand on the three most visited pages and on the contact page.

This check produces a short list of fixes, usually fewer than ten, most of which are about writing and structure rather than code. That is what makes cognitive accessibility unusual among accessibility projects. It costs little to fix and it shows up on the indicators everyone already tracks, bounce rate, form completion rate, number of enquiries received.

Seven in ten disabled customers say they click away from a website they find difficult to use, according to the Click-Away Pound surveys relayed by Scope (opens in a new tab), for 17.1 billion pounds of revenue abandoned every year in the UK. Most of them report nothing. They leave. The tired, rushed or distracted visitor who gives up for the same reasons appears in no statistic. They are simply missing from your enquiries.

A website has 47 seconds and four items to make itself understood. Everything distinctive about it has to fit inside that budget, or nobody will see it.