In 1977, a group of college students at Villanova and Temple were handed a list of sixty trivia statements and asked to rate how true each one was, on a scale of one to seven. The statements were chosen so that nobody in the room could plausibly know the answers — the first air force base, the year basketball became an Olympic discipline. Two weeks later they were handed another sixty. Two weeks after that, a third. Twenty statements appeared on all three lists. The other forty were new every time. Nobody told the students which was which, because there was nothing to cover up: this was a test of how belief moves, and the movement was the whole point. On statements that showed up once and vanished, confidence stayed flat across the six weeks. On the twenty that came back, the average rating climbed — 4.2, then 4.6, then 4.7.
Nothing argued with them. Nothing was added to the statements. The only thing that changed was that they had read the sentence before, and that alone was enough to make the sentence a little more true.
Lynn Hasher, David Goldstein and Thomas Toppino had named the thing they found, and the name they gave it is the reason this study is still cited: the illusory truth effect. Their explanation was careful and slightly unflattering. When you judge whether something is true, you use two signals. One of them is whether it fits with what you already know — slow, effortful, and correct more often than not. The other is whether it feels familiar. Familiarity is fast and free, and in a normal life it is a decent proxy, because true information does tend to circulate more than invented information. It is a good-enough shortcut right up until someone repeats something false at you, and then it quietly starts paying out in the wrong currency.
🧠 The Answer Key Doesn’t Help
The obvious objection is that the 1977 statements were things nobody could know. Sure — repetition fooled people on unanswerable trivia. But on a claim I can check, the machinery should stall out. If I know the Atlantic Ocean is not the largest ocean on Earth, one more reading of the claim should bounce off.
It doesn’t. In 2015, Lisa Fazio and her colleagues tested exactly this, using statements that participants could verify from their own knowledge — the kind where the correct answer is already sitting in your head. Repetition still pulled their truth ratings upward, including on the false ones they could have corrected on the spot. The paper’s title is the finding: knowledge does not protect against illusory truth. It has been cited more than six hundred times, mostly by people trying to argue with it.
Warnings don’t help either, and the evidence for that is old enough to be almost embarrassing. In 1979, participants were told directly that statements which appeared repeatedly were no more likely to be true than new ones. They rated the repeated statements as truer anyway. The information was delivered and understood. It lost to a feeling that arrived earlier and felt simpler.
Nor is it a trait you can screen for. In seven studies with a combined sample of 2,196 people, Jonas De keersmaecker and colleagues tested whether cognitive ability, the need for cognitive closure, or a preference for analytical over intuitive thinking made any difference. All seven studies found a healthy effect, and none of the three traits moderated it. Being smarter changes the arguments you can construct later. It does not change what familiarity whispers first.
The most recent wrinkle is the one that should worry anyone who writes about consequences. Earlier work had found that conspiracy statements were somewhat more resistant to repetition than ordinary trivia, which invited a comforting reading: maybe extreme claims have some internal guardrail. A 2026 registered replication by Shauna Bowes and Lisa Fazio aligned the two sets of statements on baseline plausibility and reran the comparison. The effect of repetition was the same size. Conspiracy claims aren’t special. They had simply been starting from a lower base.
🤔 Familiarity, Not Truth
The mechanism most researchers reach for is processing fluency. A repeated statement is read more quickly and with less friction than a new one, and that frictionlessness gets misread as a signal from the world rather than a property of your reading. Hassan and Barber put the position plainly in a 2021 paper: fluency and truth are frequently correlated in the real world, so people learn to use processing fluency as a marker for truthfulness. It is a learned heuristic, earned honestly, and it has a hole in it exactly the size of a repeat.
The sharpest demonstration came from the fake-news era. Gordon Pennycook, Tyrone Cannon and David Rand took real fabricated headlines and showed them the way they had actually spread — as Facebook posts — then asked people how accurate the headlines were. A single prior exposure raised accuracy ratings, both within the same session and a week later. It held when the headlines contradicted the reader’s own politics, which is where the tribalism explanation usually goes. It held even when the stories were labeled as contested by fact checkers. Their conclusion is worth keeping: tagging a story as disputed is not an effective solution, because a warning is a piece of information and the effect runs beneath information.
There is a boundary, and it’s narrower than you’d hope. Statements that are entirely impossible — the 2018 experiment’s example was “The earth is a perfect square” — were immune. More recent preregistered work softened even that: repeat a plausible and an implausible claim side by side, and belief rises for both, with the biggest gains on the ambiguous ones. So the real guardrail is absolute impossibility, and most of the claims a person actually encounters live comfortably on the near side of it.
🔗 Why the Third Ad Works
Advertising has been using this since before it had a name for it. The conventional explanation for why repetition works is memory — you see the brand enough times that it’s still there when you’re standing in the aisle. The illusory truth effect says the money is being spent somewhere else. Repetition buys default credibility. People will not remember that you told them four times. They will remember that you sound like a brand that has been around, which is a much cheaper thing to achieve and a much harder one to argue with.
The same mechanism is doing quiet work in any product that hands a person a description of themselves. A personalized report is read once, then returned to, then quoted to a friend, then encountered again in an export or a notification. Each pass raises the odds that the sentences inside it stop reading like interpretations and start reading like facts about the reader. Pair that with a description general enough to be claimed and a hit rate loose enough to be rewarding, and you have the full arc — vague enough to be accepted, repeated enough to be believed — with no dishonesty anywhere in the loop. Which is precisely what makes it worth knowing about.
For anyone studying anything, the effect has an uglier face. Rereading material manufactures fluency, and fluency is exactly the feeling people use to decide they have learned something. The chapter you’ve been through four times feels solid in a way a fresh chapter does not, and the feeling is not tracking retrieval. If you want a test the effect can’t fool, close the source and reconstruct it from nothing. Recall that fails is information. Familiarity that succeeds is not.
And in an information environment where the same sentences get generated, paraphrased, aggregated and reposted at a volume no previous decade could match, this is the load-bearing effect. The provenance labels are doing far less work than we assume, and the repetition is doing far more.
🎲 The Second Time Is the Expensive One
Almost every study on this used three repetitions or fewer, which left an obvious question: what does the tenth reading do? Hassan and Barber ran it out — up to nine exposures in one experiment, up to twenty-seven in another. Perceived truthfulness kept rising, but logarithmically. The single largest jump came from the second exposure. Every reading after that added less, and the curve flattened into a shape that tells you where to spend your attention: the first repetition is where a claim acquires most of its unearned weight, and after that you are mostly paying for ceiling.
Two footnotes for the shelf. Vellani, Zheng, Erçelik and Sharot showed in 2023 that a single repetition also makes people more willing to share a false statement, so the bias compounds through a network instead of sitting still in your head. And in 2014, Eryn Newman and colleagues found that the same sentence is rated truer when it’s attributed to a person whose name is easy to pronounce. No repetition at all — just a slightly smoother path through the sentence, and the truthiness arrives anyway.
Which leaves the honest test of your own beliefs, and it is a duller one than it sounds. Have I heard this before, and was the second time ever anything other than the first time again. Most of the sentences you are surest about are just the ones you have met the most often, which is a fact about your history with them and not about the sentences themselves.
The 1977 study is Hasher, Goldstein & Toppino’s “Frequency and the conference of referential validity.” The follow-ups cited here are Bacon (1979), Fazio, Brashier, Payne & Marsh (2015), Pennycook, Cannon & Rand (2018), De keersmaecker, Dunning, Pennycook, Rand & Sanchez (2019), Fazio, Rand & Pennycook (2019), Hassan & Barber (2021), Vellani, Zheng, Erçelik & Sharot (2023), and Bowes & Fazio (2026).