Recognition rather than Recall

011 min

A task that depends on recall

Picture a designer watching a usability session. A finance assistant is filing a travel expense, and the form has a field called cost centre. The field wants a code, and the assistant does not know it. They open a spreadsheet in another window, find the code, switch back and type it. Two expenses later they do the same lookup again.

The form works as built. But it asks the person to bring a piece of information out of memory, and nothing on the screen helps them find it. Each lookup also pulls their attention away from the expense they were filing. This is the cost that one of Jakob Nielsen's ten usability heuristics, the one named Recognition rather than Recall, was written to prevent. The next question is where the rule came from and how much weight it carries.

022 min

Where Recognition rather than Recall comes from

Jakob Nielsen and Rolf Molich published a first set of usability rules in 1990. They built it for a method called heuristic evaluation, in which a few reviewers inspect an interface and judge it against a short list of rules. In 1994 Nielsen ran a factor analysis on 249 usability problems, a statistical method that looks for shared patterns behind many observations, and used the result to revise the list. The memory rule survived that revision. The Nielsen Norman Group, the consultancy Nielsen co-founded, says the ten rules have stayed the same since 1994, and the sixth now reads:

No single experiment produced this rule. It comes from many reviews of real interfaces, together with memory research that arrived later in the explanation. That makes it practitioner convention, not a law with a formula. A convention is a good guide, and the later sections show where it stops being one.

What the rule does not say

Some people read it as "never ask anyone to remember anything". The wording is narrower. It targets information a person must carry from one part of the interface to another, such as a code read on one screen and needed on the next. A design can expect people to learn things over time, for example where the save command is after a month of use. It should not expect them to keep a value in mind between screens.

033 min

Why recognition is easier than recall

Memory researchers describe what a person knows as units, and each unit has an activation level that sets how easily it comes to mind. Three things raise activation: how often the unit was used, how recently it was used, and whether something in the current surroundings is linked to it. Hearing the word Paris brings France and the Eiffel Tower to mind for that third reason. The cue spreads activation to the units tied to it.

Recognition and recall differ in how many cues come with the question, and recall comes with fewer. Asked who wrote Moby Dick, a person must find the answer with almost no help. Asked whether Herman Melville wrote it, they have the answer inside the question. The same gap explains why a multiple-choice test feels easier than an open question.

One long-standing theory says recall has two stages, a search of memory and then a decision about what was found, while recognition has only the decision stage. A second idea, called encoding specificity, says retrieval works best when the cues at the time of remembering match those present when the memory formed. Brain imaging studies find that some regions are more active during recall than during recognition. Neither idea is settled. Recall has even beaten recognition in some cases, such as words a person fails to recognise but can recall later.

Daily life mixes the two. A person who wants to return to a site may recall only a few words about it, type them into a search engine and then recognise the site in the results. In 2008 Eytan Adar, Jaime Teevan and Susan Dumais reported that searching again and picking the page out of the results is the preferred way to get back to a page on the web. Typing the query is still recall, and most people are poor at it. That is why search suggestions help: they turn part of query writing into a choice from a list.

042 min

How recognition replaced typed commands

Before graphical screens, a person had to know the command. On a Unix system, renaming a file meant typing mv followed by the old and new names, so the person had to recall both the command and the order of the names. A rarely used feature such as strikethrough in a word processor was harder, because its name had to come from memory too. Menus changed that. The computer lists the commands and the person picks one, so the word Strikethrough is recognised instead of recalled.

The designers of the Xerox Star, released in 1981, chose menus and pointing because older systems made people remember and type a great deal. They wrote in 1989 that this typing and remembering hurts learning, especially for people who use a machine only now and then. The Star team treated the idea as a design principle: see and point instead of remember and type.

Gestures bring the old problem back. A swipe has to be remembered because nothing on screen says it is possible. Tips, progressive disclosure and visual hints give the memory something to work from. Tutorials fail for a related reason. An app that shows a tutorial once expects people to memorise it, but they rarely return to it. Tips shown on the page where the action is possible work better, and so does onboarding that presents one step at a time, when the person needs it.

052 min

Applying Recognition rather than Recall

Back in the usability session, the designer can fix the expense form in a few moves. The same moves work on most screens. The expected gains are faster tasks, quicker learning and fewer errors, because people spend less effort retrieving things.

1. 1

   Offer a list instead of an empty field.

   Show the person's recent cost centres and add a search box that filters as they type. They recall at most a few letters, then recognise the rest.

2. 2

   Reuse earlier answers.

   Fill in or offer any value the person already gave in the same process, and repeat their choices on the review screen.

3. 3

   Keep labels visible.

   Put a label beside every control. Placeholder text vanishes once someone types, so it cannot do that job.

4. 4

   Pair icons with words.

   An icon is recognised only if it is familiar, so add a label, at least at first. A tooltip can explain an icon on hover, but a touch screen has no hover, so a visible label is safer. Familiar conventions, such as a trash can for delete, work for the same reason.

5. 5

   Show shortcuts inside menus.

   The menu lets a novice recognise the command, and the shortcut shown beside it lets them move to recall.

6. 6

   Group and name items with words people already use.

   Card sorting, where people sort labelled cards into groups, shows which words they expect.

7. 7

   Keep history and favourites.

   Recent searches, recently viewed items and saved lists let people resume without remembering names.

8. 8

   Show where the person is.

   A progress indicator or breadcrumb trail removes the need to remember which step they are on.

9. 9

   Choose the input to fit the value.

   A calendar shows the days around today at a glance, but it is slow for a birth year, so also accept typing.

10. 10

    Give defaults and filters.

    A sensible default or a category filter turns an open choice into a short one.

11. 11

    Make the likely next step stand out.

    Colour and type size act as cues that draw the eye to it.

To find the places that need these changes, walk through each task and mark every step where the person must produce something that is not on the screen. Each mark is a recall demand. Reviewers doing a heuristic evaluation can run this check in an hour.

062 min

Recognition rather than Recall in real products

The applied cases are easy to check. Google and Bing keep search histories so people can retrace a query. Amazon's personalised homepage lists recently viewed and purchased items, so a shopper who left a purchase unfinished sees it again. Spotify shows a Recently Played list, and TurboTax asks multiple-choice questions where a blank form would ask for recall.

A violated case is also on record. The Nielsen Norman Group's 2024 article on recognition and recall shows the Volvo mobile app, where icons carry no labels and are hard to interpret. The group's advice is a label beside at least some of the icons for the first uses. The command line is the older example: it works for people who know the commands and gives everyone else nothing to go on.

Medical devices need the rule even more. A nurse setting an infusion pump under time pressure should see the available options, not rely on a memory of last week's training.

Text boxes for AI tools

A prompt box for an image generator is a recall task. Tarun Mugunthan of the Nielsen Norman Group studied more than 100 Midjourney prompts in 2023. He found that people often could not recall the words for the style or detail they wanted. His proposed fix was a hybrid interface, with a text box plus thumbnails of visual styles to select. A person can pick the look they recognise without knowing its name, and over time they learn the names.

071 min

Recognition rather than Recall in log-in

Logging in is the classic recall task. A person must produce a username and a password with almost no cues, and the site name is the only hint. People cope in predictable ways. Some reuse the same credentials everywhere, some build a password from the site's name, and some keep a list on paper or in a password manager.

Researchers asked whether recognition could replace the typed secret. In 2000 Rachna Dhamija and Adrian Perrig at the University of California, Berkeley built a log-in system named Déjà Vu in which the user proves who they are by picking out images they have seen before. In the study, about 90 percent of participants passed with the image method and about 70 percent with passwords and PINs. The authors suggested it for devices where typing is hard and for passwords used rarely.

The idea has not replaced passwords, but accessibility rules now ask for the same thing. The next section covers them.

081 min

Recognition rather than Recall and accessibility

Two success criteria in the Web Content Accessibility Guidelines (WCAG) 2.2 turn this heuristic into a testable requirement. WCAG 2.2 success criterion 3.3.7, Redundant Entry, at Level A, requires information the user already gave in a process to be filled in automatically or offered for selection when it is needed again. This is the expense form's second fix, written as a rule. WCAG 2.2 success criterion 3.3.8, Accessible Authentication (Minimum), says that signing in must not require a memory or puzzle test, such as recalling a password, unless the site also offers an alternative method, an assisting mechanism or a task that only asks the person to recognise objects.

The W3C explains why. Some people with cognitive disabilities cannot memorise a username and password, solve a puzzle or retype a one-time code. The object recognition exception is notable here: a test that only asks the person to recognise objects is allowed, which is the idea behind Déjà Vu.

092 min

Where Recognition rather than Recall backfires

Showing everything is not the same as helping. The rule fails in five known ways.

101 min

Recognition rather than Recall vs. nearby ideas

Three neighbours are often mixed up with it.

111 min

Frequently asked questions

?3 questions

Questions people ask

§10 sources

Sources