The first time he says "you always do this," she argues. The tenth time, she argues less. Somewhere past the fiftieth, in the middle of an unrelated conversation, she catches herself thinking: I always do this. She has no new evidence. She has heard the sentence.
In 1977 Lynn Hasher, David Goldstein and Thomas Toppino published a study that has become a fixture of cognitive psychology. Over three sessions spaced two weeks apart, participants rated the truth of dozens of plausible-sounding statements of general knowledge, some true and some false. Some statements were repeated across sessions; others appeared once. Repeated statements were rated as more probably true each time they came round, whether or not they were actually true. Nothing about the statements had changed except that the participants had seen them before.
The authors called it the frequency-validity effect. It is now usually called the illusory truth effect, and it has been reproduced many times. Alice Dechêne and colleagues at the University of Basel pooled the literature in a 2010 meta-analysis and found a reliable effect of moderate size, present across different kinds of statements and methods.
The leading explanation is processing fluency. A statement seen before is easier to read and understand the second time, and the mind uses that ease as a cue. Rolf Reber and Norbert Schwarz showed in 1999 that simply making a statement easier to read, by printing it in higher-contrast colour, made people more likely to rate it as true. Repetition is one route to fluency; clarity, rhyme and a confident voice are others.
Ian Begg, Ann Anas and Suzanne Farinacci made a decisive contribution in 1992. They had participants hear statements attributed to sources that were described as always truthful or always lying. Weeks later, participants still rated the repeated statements as truer, including the ones they had been told came from liars. The memory of having heard a statement outlasted the memory of who said it. Familiarity survived; the warning did not.
Christian Unkelbach at Heidelberg showed in 2007 that the link can be retrained: when participants learned in the lab that fluent statements were usually false, the effect reversed. That is evidence for the mechanism, not a practical remedy. Outside the lab, nobody retrains the link.
The most unsettling result is recent. Lisa Fazio, Nadia Brashier, Keith Payne and Elizabeth Marsh at Duke and Vanderbilt published a study in 2015 with the plain title "Knowledge does not protect against illusory truth." Participants read statements such as "The Atlantic Ocean is the largest ocean on Earth", which almost everyone knows to be false. Repetition still raised the rated truth of such statements. People who could have answered correctly if asked did not consult what they knew. They consulted how the sentence felt.
Gordon Pennycook, Tyrone Cannon and David Rand extended this to fabricated news headlines in 2018. A single prior exposure to a false headline increased how accurate people rated it, the increase persisted a week later, and it held even when the headline had carried a warning that fact-checkers disputed it, and even when the headline contradicted the reader's own politics. The effect was not large. It was there in every condition they tried.
It is important not to overstate this. The measured effect is a shift in rated truth, typically on a scale, not a conversion. Repetition makes a claim seem somewhat more plausible; it does not, on the evidence, make people believe things they find absurd. Nadia Brashier and Elizabeth Marsh's 2020 review lays out the boundary conditions: the effect is strongest for claims that are plausible to begin with and about which the person has weak or no prior belief. Whether repeated exposure compounds without limit, or saturates after a few repetitions, is still being worked out.
The abuser who says "you always do this" is not usually thinking about Hasher, Goldstein and Toppino. He does not need to. Repetition is cheap, and its target does not experience it as persuasion. Each restatement lands as something slightly more familiar, and familiarity is filed, over time, under true. "Everyone thinks you're difficult." "You know you're not good with money." "You'd be lost without me." None of these needs an argument. They need airtime. The same mechanism explains why the person on the receiving end can know, in a quiet moment, that the statements are false, and still feel their weight. Fazio's participants knew about the Pacific.
The public version is the slogan, the talking point, the claim that circulates for a week and is corrected once. Corrections have to fight repetition with a single exposure. They rarely win.
The defence the research supports is modest but real. Familiarity is not evidence, and the feeling of truth that comes with the fifth hearing carries no more information than the feeling that came with the first. Noticing that the only thing a claim has going for it is that you have heard it before is the whole skill.