Mere exposure effect

The textbook version of this principle is a straight line: familiarity breeds liking. The meta-analytic version is a hill. Montoya, Horton, Vevea, Citkowicz and Lauber modelled 268 curve estimates from 81 articles and found a positive slope with a negative quadratic term — an inverted U, not a monotonic rise. It appeared for visual but not auditory stimuli, and across exposure durations from under 10 seconds to over a minute. No existing model of the effect accounts for the pattern, so the curve is the finding and the explanation is open.

The downturn has been measured directly. Bornstein, Kale and Cornell showed stimuli 0, 1, 2, 5, 10, 25 or 50 times. For the more complex stimuli, affect ratings rose through about five exposures and then fell; by 50 exposures they were rated no differently from stimuli never shown at all. Their first experiment adds the limit: participants high in boredom proneness showed no exposure effect whatsoever, while low-boredom participants rose monotonically. Overexposure is not a separate moral category from exposure — it is the same curve past its peak, in someone who has run out of patience.

The effect is also less reliable than its fame suggests. Chow, Rhodes, Rule, Buchsbaum and Hasher ran four online experiments on neutral faces, repeating images 3, 5 or 7 times; repetition raised liking in none of them, with substantial Bayesian evidence against it every time, BF01 > 16.41 — and the same data reproduced the known preference for younger faces, so the procedure could detect an effect. de Zilva, Mitchell and Newell found the preference for familiar faces disappeared once the nonsense-word cue each face had been paired with was rearranged or replaced at test. What is liked is a stimulus in its learned context, so a familiar component moved to a new screen does not arrive with its liking intact.

Almost all of this runs on novel, meaningless material: polygons, ideographs, nonwords, strangers' faces. The closest work to a product is Hekkert, Thurgood and Whitfield, who found rated attractiveness of everyday object shapes rose with how common the objects were, and then needed fictitious but plausible products to get exposure frequency under control at all. There attractiveness rose linearly with exposure frequency, and the slope was steeper when the product was presented as something nobody had paid attention to designing. The window closes at the other end too: Pugnaghi, Memmert and Kreitz used an inattentional-blindness paradigm (N = 240) and found no exposure effect under preconscious processing, against the folklore that unnoticed exposures are where the effect is strongest. The moderator is not being scrutinised, not not being seen.

Mrkva and Van Boven reframe what is being counted. They showed stimuli 9, 3, 1 or 0 times and found repetition raised liking, salience, evaluative extremity and emotional intensity — but in their fourth experiment it was relative exposure that did the work: nine showings raised liking only when the stimuli around it had been shown less often, and absolute exposure had no significant effect. Salience mediated the results better than fluency did. Fluency is not dead as a mechanism — Reber, Winkielman and Schwarz raised prettiness and liking ratings with a matching prime, higher figure-ground contrast and longer presentation, without changing exposure frequency — and it survives for liking where it failed for truth judgments (see processing fluency). For a build, the reading is that a visual language everybody sees equally and constantly has no relative advantage to trade on.

Schmidt and Eisend's meta-analysis of advertising repetition puts the experimental maximum for brand attitude at roughly ten exposures while recall keeps rising and does not level off before the eighth; low involvement and spaced exposures strengthen the attitude effect, massed and embedded exposures strengthen recall, and both decay over time. Courbet, Fourquet-Courbet, Kazan and Intartaglia had 398 participants browse a health site carrying four three-second pop-ups for a fictitious mineral water and never mentioned the brand again; 7 days and 3 months later the exposed groups rated attitude and purchase intention higher than controls and responded faster, the image format keeping its advantage at three months. Four barely attended glances, still readable three months on, are both the evidence that this works at product scale and the reason a frequency cap needs an ethical justification and not only a performance one.

On the usual product recommendation — keep the visual language consistent across touchpoints — there is no controlled evidence at all: nothing in this literature manipulates internal consistency. What has been tested is close to the opposite construct, conventionality against the category. Tuch, Presslaber, Stöcklin, Opwis and Bargas-Avila had participants rate real website screenshots and found prototypicality — looking like a typical site of its kind — carried a larger effect on aesthetic ratings than visual complexity, ηp² = .812 against .581, detectable at 50 ms of exposure and still present at 17 ms, with low-complexity, high-prototypicality pages rated most appealing. That is a case for looking like what users already know, not for matching yourself. The opposite folk claim fares no better: Horsham and colleagues' registered replication with 801 participants found that learning more traits about a person did not reduce liking, r = −0.027, even though people expected it to.