Memory is rebuilt at every retrieval, not replayed from a fixed archive

What it is

Recalling an experience is not reading a fixed record back verbatim — the brain assembles a version on the spot, guided by present cues, current goals, and existing knowledge. This is reconstructive memory. It breaks the "store-then-retrieve" database metaphor entirely: memory works more like repainting a picture each time — the outline stays roughly the same, but details can shift from one telling to the next. Bartlett's classic serial-reproduction studies showed this directly: people recalling a story from an unfamiliar culture unconsciously swap its details for more familiar equivalents from their own.

Why it happens

On-the-spot reconstruction runs on schemas — general knowledge structures about what a given kind of event or object is usually like. At retrieval the brain doesn't play back the original encoding word for word; it uses schemas to fill gaps left by details that were never fully attended to, or that have since decayed. Content consistent with the schema gets filled in automatically; content that conflicts with it tends to be simplified or dropped outright. This is why memory errors are rarely random noise — they drift systematically toward what's "more plausible" or "more typical," because reconstruction's goal is a coherent, usable narrative, not a faithful copy of the original input.

Studying it

The earliest evidence came from serial-reproduction designs: one participant retells material to the next, and after several relays the final version is compared against the original — the divergence clusters in the direction of schema assimilation. A second line of evidence is the false-memory paradigm (e.g., the DRM paradigm): presenting a list of semantically related words leads participants to later "recall," with high subjective confidence, a related word that was never actually presented — showing memory content can be fabricated outright. Both paradigms depend on the researcher being able to independently verify what the original material actually was; without that reference point there is no way to tell reconstruction from accurate recall.

Where it stops holding

Reconstructive drift doesn't mean memory is uniformly unreliable: for experiences that are highly familiar, frequently retrieved, or emotionally intense, the gap between the reconstructed version and the original tends to be small; drift shows up most under free recall, long delays, and low retelling frequency. Lab paradigms deliberately engineer schema conflicts or semantic-association traps to make the effect visible, and everyday memory tasks don't always hit those same traps — so error rates measured in the lab shouldn't be read as the error rate for users' ordinary recall.

Related

• Same group: A6.20.2 Retrieved memories absorb new information, shifting later recall · A6.20.3 Misleading information can be unknowingly incorporated into an original memory after the fact · A6.20.4 A user's account of what they previously did cannot serve as an objective behavioral record
• Nearby: A6.06 Forgetting curve · A6.13 Encoding in long-term memory
• Search terms: reconstructive memory · schema · serial reproduction · false memory