Forgetting is fastest right after learning

What it is

Forgetting of newly learned material doesn't proceed at a constant rate over time: most of the loss happens within the first minutes to a couple of days after learning, after which the rate of loss drops off sharply and whatever survives that window tends to stick around much longer. This decelerating loss-rate-over-time pattern is the forgetting curve, first measured by Hermann Ebbinghaus using self-administered trials with nonsense syllables. It describes how information in long-term memory fades over hours, days, and weeks — a different phenomenon from the seconds-to-tens-of-seconds decay of working memory. The forgetting curve answers "how much is left a day, a week, a month after learning," not "can this still be used a few seconds after being held in mind."

Why it happens

A newly encoded memory trace is fragile when first formed and needs time to be stabilized through consolidation into a more durable representation. During that fragile window, interference and competition from newly incoming information easily wipe the trace out, which is why the forgetting rate is highest in that period. Once a trace survives the fragile window without being overwritten, it becomes comparatively more robust, and the per-unit-time probability of losing it drops substantially. That's why the curve is negatively accelerated — often fit with a power function or exponential decay — rather than linear: it isn't that memory leaks out steadily, it's that the fragile early period filters out a batch of traces, and whatever survives that filtering is more resilient.

Studying it

Ebbinghaus's original method was savings: learn nonsense material to a set criterion, relearn it after varying delays, and use the reduction in relearning effort ("savings") to infer how much was retained. Modern work more often uses direct testing: administer recall or recognition tests after different retention intervals and plot accuracy against elapsed time; some studies track retention over years or decades — for instance following adults who once learned a foreign language — to check whether the curve genuinely flattens toward a nonzero floor rather than approaching zero. A shared requirement across these paradigms is controlling for any rehearsal or re-exposure during the interval — a single incidental reactivation during the delay contaminates what's supposed to be a pure forgetting measurement.

Where it stops holding

Ebbinghaus deliberately used nonsense syllables to strip out prior knowledge and associative meaning, so the curve would reflect memory decay alone; material that is meaningful, well understood, or anchored to existing knowledge structures forgets noticeably more slowly and the curve flattens differently, so the specific numbers from nonsense-material studies don't transfer directly. The curve also doesn't apply uniformly across memory types: once a motor or procedural sequence has been repeated enough to become automatized, its rate of loss falls well below what this curve predicts — a different memory mechanism is doing the work by that point.

Applying it

• Don't judge how well a tutorial, onboarding flow, or piece of instructional copy "took" by testing users the moment it ends — that moment sits before the steepest part of the forgetting window, so it almost always overstates real retention; things that look learned can be half-gone by the next day.
• When evaluating one-time teaching moments, schedule the retention check after a real gap (the next day, or a week later) rather than immediately at the end of instruction, so the number reflects what users actually carry forward.
• Verification: test the same cohort's task completion right after instruction, again the next day, and again a week out, and compare the drop-off across the three points to see whether the current onboarding only looks effective in the moment.

Related

• Same group: A6.06.2 Operating knowledge for infrequently used features is inevitably forgotten
• Nearby: A6.02 Working memory capacity · A6.09 Procedural memory and automatization · A6.15 Forgetting curve and spaced repetition
• Search terms: forgetting curve · Ebbinghaus · memory consolidation · retention interval