What cognitive psychology says about interface memory load

Perceivio·28 May 2026

Working memory is small and shared

Working memory is the part of thinking that holds something while you use it: a phone number read from one screen to type into another, a reference code kept in mind while a form scrolls.

It is small, it is shared with whatever else the reader is doing, and it does not improve with practice. Anything a reader has to remember across a step is work the interface chose to give them.

This is the part of cognitive psychology that pays for itself in interface work, and it is close to the reason disability advocates and usability engineers keep arriving at the same recommendations.

Recognition beats recall

A reader who has to recall a piece of information is working from memory. A reader who has to recognise it is working from the screen. Recognition is easier, and every choice about a form is a choice between the two.

• Show the options rather than requiring a code. A list of payment providers is recognition; a merchant number is recall.
• Keep the summary visible while the details are entered, instead of asking the reader to remember what they asked for.
• Show a step number and what is left, so the reader is not holding the shape of the whole process in mind.
• Autocomplete addresses, dates and card details, and never ask for the same field twice in one process, which is what WCAG 2.2 asks for at 3.3.7 as well.

Three kinds of cognitive load

Cognitive load theory comes from educational psychology, and it separates the load a task carries into three kinds. The distinction is useful well beyond the classroom.

• Intrinsic load is the task itself. Booking a flight with three connections is harder than booking a train ticket, and no design change removes that.
• Extraneous load is the load the interface adds: an unlabelled icon row, a validation message that names no field, a layout that moves while it loads.
• Germane effort is the useful work of learning the product, which is worth protecting rather than cutting.

The usable rule is that extraneous load is always a defect and intrinsic load is sometimes a fairness question. An interface cannot make the task simpler, but it has no excuse for adding to it.

Chunking and progressive disclosure

Group the fields that belong together and separate the groups with space and a heading. Ask one decision per step where the decisions are hard, and one screen where they are easy.

Progressive disclosure means keeping the rarely used options out of the way, which is a real gain until it hides something a reader needed. Put the common path in front of the reader and keep the rest one press away, in a labelled place.

What it buys

Fewer errors, more completions, and shorter calls to support. It also removes barriers for readers with attention, memory or learning disabilities without a separate project, because the same change serves everyone.

That is the useful thing about memory load as a subject. It is not a special case for disabled users, and it does not need a separate budget. It is the ordinary work of not asking a reader to remember what the screen could have told them.

How to test it

Watch five people who have not seen the product try one task, and count the errors and the backtracks rather than the time. Ask them to say what they expect to happen before they press anything, which surfaces the places where the interface and the reader disagree.

Then run the same task with a screen reader and at 200 per cent text, because the load added by a magnifier or by listening rather than skimming is load the reader cannot simply work harder to absorb. Our how we test page sets out the passes we run, and cognitive accessibility in a digital product covers the criteria that carry it.