In 1949, a team led by Carl Hovland sat American soldiers down in front of Frank Capra’s Why We Fight films — straight wartime propaganda, made to stiffen the resolve of men about to be sent overseas. Then the researchers did something patient. They measured the soldiers’ opinions five days after the screening, and again nine weeks later. If you had to guess which measurement showed the films working harder on the men who watched them, you would probably say the first one, while the thing was still fresh. It was the second. The gap in opinion between soldiers who had seen the films and soldiers who had not was larger at nine weeks than at five days. The message had not worn off. It had gone to work late, quietly, after everyone had stopped thinking about it.
That delayed increase is what Hovland’s group ended up calling a sleeper effect, and it is one of the strangest findings in the study of persuasion, because it inverts an otherwise boringly reliable pattern. Eagly and Chaiken’s 1993 review of longitudinal persuasion research found that attitudes usually drift back toward where they started. You get moved, then you drift home. That is the default.
The sleeper effect happens when a message arrives carrying what researchers call a discounting cue — a signal that tells you to take it with a grain of salt. A low-credibility source. A disclaimer. A label saying the content was paid for. A byline from the opposing campaign. The cue does its job immediately: it suppresses the attitude change that the message would otherwise have produced on its own. Then, over weeks, the suppression stops holding and the message keeps paying out.
🧠 The Ad You Already Decided Not to Believe
The clean example is negative political advertising. You are an undecided voter, you watch an attack ad, and at the end you catch the small print: paid for by the other side. Something in you files it away correctly — of course they would say that — and you are not persuaded. According to the sleeper effect, that is not the end of the story. Weeks later, at the moment that actually matters, the content may be sitting there doing the work the ad was paid to do, with the objection detached from it.
Ruth Ann Weaver Lariscy and Spencer Tinkham looked at exactly this in a 1999 Journal of Advertising paper on negative political advertising, and the underlying mechanism has been replicated and argued over for the better part of seventy years. Consumer researchers picked it up early, too — Darlene Hannah and Brian Sternthal published on detecting and explaining it in the Journal of Consumer Research in 1984, and David Mazursky and Yaacov Schul followed in 1988 with a study on how the way an ad is encoded affects whether you end up failing to discount it. This is not a curiosity from an old journal. It is a standing feature of how advertising can work on you.
🤔 Fifty Years of People Trying to Kill It
The honest part of this story is that the sleeper effect spent decades being nearly impossible to reproduce.
Hovland’s team first assumed simple forgetting. The soldiers forgot that the material was propaganda, so their resistance faded. Then they checked. The recall data said the subjects still remembered the non-credible source at the delayed follow-up (Hovland and Weiss 1951; Weiss 1953). Forgetting was wrong as an explanation, so the hypothesis changed shape: the memory of the source survives, but its association with the message loosens. The cue is still in there. It is just harder to retrieve in connection with the claim.
Then came the attempts to bury it. In 1974, Paulette Gillig and Anthony Greenwald published a paper under the title “Is it time to lay the sleeper effect to rest?” Thomas Cook and Brian Flay countered that earlier failures simply had not met the four conditions a real test requires: a genuinely persuasive message, a discounting cue strong enough to suppress immediate change, enough time between the two measurements, and a message still potent at the delayed one. In 1978, Charles Gruder, Cook, Karen Hennigan, Flay and Cynthia Alessis ran the test properly and got the version they called the absolute sleeper effect — an actual rise in persuasion from immediate to delayed measurement, not merely slower decay. A decade later, Anthony Pratkanis, Greenwald, Michael Leippe and Michael Baumgardner published the paper whose title has followed the phenomenon ever since: “The sleeper effect is dead: Long live the sleeper effect.”
The current answer comes from G. Tarcan Kumkale and Dolores Albarracín’s 2004 meta-analysis in Psychological Bulletin. They pooled the judgment and memory data and found the increase was real under specific conditions: when the message arguments and the discounting cue both had strong initial impact; when recipients had more ability or motivation to think about the message; and, most usefully, when the discounting cue came after the message rather than before it.
Sit with that middle condition for a second. The effect is stronger in people who thought harder about what they were reading. The lazy version of the story — that the sleeper effect works on people who were not paying attention — is backwards.
🔗 The Order of the Label Is the Whole Game
For anyone who makes things meant to persuade, including ads, the practical rule falls straight out of that third condition. A disclaimer placed before the content gets absorbed as part of how you read the content. The same disclaimer placed at the end arrives after the claim has already been encoded, at which point it is a separate object that can be separated from the message. The content stays. The label drifts off on its own.
That is also the shape of a much more ordinary modern problem. A screenshot outlives the account it came from. A striking statistic gets shared with the handle cropped out and keeps circulating years after the paper it misread has been corrected. Every one of those is the sleeper effect running without a researcher attached: the claim is what you keep, the provenance is what you lose, and the loss is not random — it is the part your memory was never going to hold onto for long. It is also an argument for checking things while you still remember who said them, since that window closes on its own schedule.
There is a quieter version in narrative work, too. A character who says something wrong, or self-serving, or a little cruel, gets discounted by the reader in the moment, because the reader can see who is talking. The line arrives again weeks later, unattached, and by then it has your ear.
🎲 Where You Put the Disclaimer
Two things you can actually do with this.
The first is a test you can run on your own material. If you write or make anything with a disclosure, a caveat, or a source note, move it. Put it before the content once, then after, and notice which version feels like the caveat did something. The research says the trailing version is the leaky one.
The second is a small inventory. Name three sentences you have repeated in the last month whose origin you cannot point to — a statistic, a piece of advice, a line someone once said to you. Check whether any of them came from a source you would have dismissed at the time. The messages you declined to believe are not necessarily the ones that stopped working on you. Some of them were simply waiting for you to forget who sent them.