Cognitive load

Cognitive load is the amount of mental work a task demands. In UX, Nielsen Norman Group defines it as "the amount of mental resources that is required to operate the system." Because working memory is limited, every bit of effort spent figuring out the interface is effort not spent on the actual task.

Where it comes from

Cognitive load theory comes from educational psychology. John Sweller's 1988 paper in Cognitive Science argued that conventional problem solving uses a relatively large amount of cognitive processing capacity, which is then unavailable for learning. Later work by Sweller, van Merrienboer and Paas (1998) described a limited capacity working memory alongside an effectively unlimited long-term memory, and is where the distinction between types of load became central.

Two types matter most for design:

• Intrinsic load is the effort the task itself demands. Comparing three mortgage offers is inherently harder than checking the weather. In NN/g's words, it is the effort of absorbing new information and keeping track of your own goals.
• Extraneous load is effort caused by how information is presented. NN/g describes it as processing that takes up mental resources but doesn't actually help users understand the content. Confusing layouts, jargon and unnecessary choices all add it.

Designers can rarely remove intrinsic load, but they can remove most extraneous load. (Later versions of the theory also name a third type, germane load, linked to building long-term understanding. It matters more for teaching than for everyday UI work.)

Why it matters

When load is too high, people slow down, miss details, make errors and give up. This hits hardest when users are new, distracted, on a small screen or under stress.

How to apply it

NN/g's three tips are a good starting point: avoid visual clutter, build on existing mental models, and offload tasks.

Do:

• Remove decoration, redundant links and copy that does not help the task.
• Use familiar layouts and labels so people do not have to learn your conventions.
• Show information instead of asking people to remember it. Prefill what you know, and use defaults.
• Break complex tasks into steps, or hide rare options with progressive disclosure.

Don't:

• Ask users to carry information from one screen to the next.
• Present every option at once when most people need two.
• Mistake "fewer elements" for "less load." Hiding essential information makes people hunt for it, which also costs effort.

Common mistakes

• Treating cognitive load as a number to minimize at all costs. Some tasks are intrinsically demanding. The goal is to cut waste, not to oversimplify.
• Citing "7 plus or minus 2" as a hard rule for menus. Working memory limits are real, but people scan visible menus rather than memorize them.
• Ignoring the reader's state. The same page can be easy at a desk and overwhelming on a phone in a hurry.

In AI products

Generative AI can shift load rather than remove it. A long, dense answer moves effort from writing to reading and checking. Lead with the answer, keep paragraphs short, use structure only when it helps, and offer "show more" for detail. Streaming responses help with waiting but can add load if text reflows while people try to read. Open-ended prompt boxes also add load, because users must work out what to ask and how to phrase it. Suggestions, templates and sensible defaults reduce that cost.