When Waiting Wins
Most people assume feedback should arrive as fast as possible: correct an error immediately and it cannot take root. The intuition is reasonable, but the evidence often cuts against it. When learners receive corrective feedback after a delay — hours, or even a day — they tend to remember the correction better than when it arrives within seconds of the response.
The finding has a long history. Early laboratory work on humans showed that immediate feedback sometimes produced better performance on a same-session re-test, while delayed feedback produced better performance when the test came days later. The two measures were pulling in opposite directions, which is why short experimental windows missed the effect for so long.
The mechanism researchers find most convincing draws on the same logic that underlies desirable difficulties: a slight struggle at encoding creates a sturdier trace.
When feedback is delayed, the learner has often forgotten the error and must actively retrieve the original attempt before integrating the correction — a small but real act of retrieval that strengthens the updated memory. The error itself is also less likely to be re-encoded strongly, because less time has been spent rehearsing it.
Immediate feedback feels more helpful, and by some short-term measures it is; the learner knows faster whether they were right.
Robert Bjork at the University of California, Los Angeles, whose lab has done more than any other to map the interaction between difficulty and retention, has described the phenomenon as consistent with his broader framework: conditions that feel worse during acquisition frequently produce better long-term retention. Delayed feedback feels frustrating but tends to produce the more durable correction.
The effect is not universal. For procedural skills — where an error could compound into a worse error if left uncorrected — immediate feedback retains a clear practical case. And where learners are very early in acquisition, sitting with uncertainty can sometimes compound confusion rather than resolve it. The delay-advantage is clearest for factual and conceptual material tested with a meaningful lag before the delayed feedback arrives, typically somewhere between twenty minutes and twenty-four hours in the literature. Beyond that window the evidence thins.
What survives scrutiny: the intuition that faster is always better is wrong, and not in a minor way. The optimal timing of feedback depends on what you want to measure and when.