Retention is brief and vulnerable to interference

What it is

Working memory content isn't just limited in how much it can hold at once — it also fades fast, and is unusually easy to knock out by "something else happening in between," even when that something is entirely unrelated to what was being held. A user remembers a verification code, switches apps to enter it, an unrelated notification pops up mid-switch, and by the time they get back the digits are gone. That isn't simply because time passed — it's because handling the notification consumed the very mechanism that was maintaining the code. This gets attributed to "too much time went by," but the real damage usually comes from interference, not the clock.

Why it happens

Interference is more destructive than plain waiting because maintaining content in working memory depends on ongoing processing effort, not on filing content away in a box and being done with it. The moment attention shifts to process new information or perform another cognitive task, the processing resources that were maintaining the earlier content get reassigned, and that content decays rapidly or is displaced outright. This differs from the mere passage of time: for the same elapsed interval, forgetting is markedly lighter if nothing cognitively demanding happens during the wait than if the person is asked to do even simple mental arithmetic or reading in the meantime. This is also why an interruption's damage to working memory tends to track how much cognitive processing the interrupting task itself demands, not how long the interruption lasted.

Studying it

The classic paradigm has participants memorize a set of items, then inserts an interference task requiring active cognitive processing (serial subtraction, rapid judgment tasks), then tests recall accuracy against a control condition of the same duration with no interference task inserted. If forgetting is markedly worse under the interference condition, the damage is attributed mainly to interference rather than pure time decay. Researchers also systematically vary the interference task's cognitive load to see whether forgetting scales with the interfering task's difficulty, which confirms an interference effect as distinct from a time effect.

Where it stops holding

How destructive interference is depends on how similar the interfering task is to the original content and how much processing depth it demands: a brief pause requiring no active processing (quiet waiting) causes far less forgetting than an equally long inserted task requiring active thought. This also means "the user was interrupted for a few seconds" alone is underspecified — what matters is what the user was doing during that interruption and how much of the same cognitive resource it consumed.

Applying it

• For any flow that requires a user to briefly hold a piece of information before entering, confirming, or checking it against something, evaluate whether the flow inserts other steps requiring active thought in between (a confirmation dialog, a secondary choice, a jump to another screen requiring judgment) — such inserted steps destroy held content far more reliably than plain waiting does.
• For verification codes, one-time passwords, or values that must be briefly held in mind, let users complete entry on the same screen wherever possible, avoiding an app or page switch mid-task; where a jump is unavoidable, keep the original information available to check back against rather than requiring it to be carried entirely from memory.
• How to check: compare two test conditions — one that goes straight from the memorization step to the next step, and one that inserts a secondary step requiring judgment or input in between; compare input error rates or how often users look back to check. A larger gap flags the flow as more interference-sensitive and a higher priority for redesign.

Related

• Same group: A6.02.1 The number of items working memory can hold at once is limited · A6.02.3 Requiring users to remember content across pages is itself a design flaw · A6.02.4 The classic capacity estimate is closer to four chunks than the earlier figure of seven · A6.02.5 Without rehearsal, working memory content decays naturally within roughly ten to twenty seconds · A6.02.6 Rehearsal extends retention but draws on the same limited processing resource · A6.02.7 Capacity varies with item complexity, so simple and complex items can't be compared on one number · A6.02.8 Capacity limits and time-based decay are two independent mechanisms, and easing one doesn't substitute for the other
• Nearby: A5.08 Interruption cost and task resumption · A6.16 Proactive and retroactive interference
• Search terms: interference-based forgetting · proactive interference · retroactive interference · working memory disruption