The reason a question about blame changes the memory

Witnesses watched the same film of a collision. Asked how fast the cars were going when they smashed into each other, they gave higher speeds than those asked when they hit each other. A week later the smashed group was more likely to report broken glass. There was no broken glass.

What everyone sees

Memory is imagined as a recording that a question plays back. On that model a leading question can distort an answer but not the underlying record, so the remedy is to discount the answer and ask again in more neutral words.

What is actually happening

Recall is reconstructive. Retrieving an event rebuilds it from fragments, and material supplied by the question is incorporated into the rebuild, after which it is stored as part of the memory. Follow-up work planted individual details and, in some cases, whole events; participants then reported them with ordinary confidence and ordinary sensory detail.

Why it stays hidden

The edit leaves no seam. A reconstructed memory feels exactly like an unmodified one, and confidence tracks vividness rather than accuracy, so neither witness nor questioner can detect the change from the inside. The safeguards that work — recording the first account, standardising wording — are administrative rather than intuitive, which is why they have to be imposed rather than discovered.