They changed one verb and the same car ran 14 km/h faster

In 1974, forty-five students at the University of Washington watched seven short films of traffic accidents, the kind used in safe-driving courses. Those asked how fast the cars were going "when they smashed into each other" gave an average of 40.8 miles per hour; those asked "when they contacted each other" gave 31.8. In kilometres: 65.7 against 51.2, fourteen and a half of difference from exactly the same film, the same cars, the same impact.

The question was identical but for the verb, and there were five: smashed, collided, bumped, hit, contacted. That word was the experiment's only variable. And in a second group, a week later, the same verb made sixteen people remember broken glass where there was none.

It was signed by Elizabeth Loftus and John Palmer, a psychologist and a student at that university, in a five-page paper that is now among the most cited in its field. This is how they built the second half.

They took 150 fresh students and showed them a film of under a minute in which several cars piled into each other; the crash itself lasted about four seconds. Fifty were asked how fast the cars were going when they "smashed". Another fifty, when they "hit". The last fifty were asked nothing about speed. Then everyone went home.

A week later they were called back and slipped a trick question: did you see any broken glass? There was not a single piece of broken glass in the film. In the "smashed" group, sixteen out of fifty said yes. In the "hit" group, seven. In the group that had been asked nothing, six.

They did not hesitate. They did not say "I think so". They remembered it.

That a memory gets rebuilt every time it is used is known to anyone who has argued with a sibling about a holiday twenty years ago. What Loftus and Palmer measured is something else, and it is the uncomfortable part: who does the rebuilding. Not only the person remembering. Also, with equal rights, the person doing the asking. One verb of theirs, chosen without thinking, goes into the brickwork and stays there.

It is worth saying what the experiment does not prove, because classics get stretched. These were films, not real crashes; these were students in a lab with nothing at stake; and the effect is a group effect, not a personal one: nobody can plant a whole memory in you with one sentence. What moves is the edge of what you remember — the speed, the colour, whether there was glass — and questions nudge that edge towards wherever they point.

Where it did change everything was in how interviews are conducted. The cognitive interview, from the 1980s, and the PEACE model used by British police since the 1990s forbid precisely what Loftus and Palmer did on purpose: putting the fact you are fishing for inside the question. In practice the interview opens with a broad question and a long pause, and the closed questions are saved for the end, once the witness has told their version with nobody's help.

And it did not stay a laboratory matter. The Innocence Project, the American organisation that reviews convictions using DNA evidence, counts 367 DNA exonerations in the United States, and in 252 of them — 69 per cent — a witness had identified the wrong person. People who pointed at a stranger in a courtroom with total conviction, and who were not lying. That is how they remembered it.

And it works outside a police station, because you do it every time you ask about something you did not see. Ask your child whether they misbehaved at school and you will have handed them the script before they open their mouth. Offer a customer who cancelled "was it the price?" and some will say yes, because it is the easiest way out and you opened it for them. With your partner it is the worst of the three: the moment you ask whether the restaurant was expensive, you will never know what they thought before hearing you.

Tomorrow, when you actually want to know something, take out of the question the fact you are fishing for, and take out the verb above all: that is what carries your suspicion. Ask "how was school?" and then say nothing for ten seconds: the other person fills the gap, and they fill it with their own material. Save "did they hit you?", "was it the price?" and "was there broken glass?" for the end, when there is nothing left to manufacture.