Relative, Not Absolute

Every decision about advertising frequency, email cadence, or how often to appear in front of a prospect rests on an assumption about repetition. This article covers what the evidence says, including the part that undermines the usual reading of it.

Our Grades For These Claims

Applying the scheme from the first article in this series.

Grade A that repetition increases liking under some conditions. Two meta-analyses in Psychological Bulletin, twenty-eight years apart, and a literature of over 200 experiments in its first two decades alone.

Grade B for the effect size of r = 0.26, which reaches us through a citing source rather than the meta-analysis itself.

Grade B that subliminal exposure produces larger effects, on the meta-analysis plus a direct two-experiment test.

Grade C for the salience reinterpretation, which is the most consequential claim here and which reaches us through a manuscript we did not obtain in full.

Our position: the effect is one of the better established in this series, and what produces it is genuinely open, with a live proposal that it is about standing out rather than about repetition.

A Note On Method

Everything here is verified to August 2026.

We obtained the 1989 meta-analysis's abstract verbatim from two independent sources, and a passage of its own text from a hosted copy[4][5]. We did not obtain the meta-analysis.

The headline effect size does not come from the meta-analysis, but from a paper citing it[1], and we flag that at every use.

We did not obtain the 2017 re-examination and have only its citation and a bibliographic service's summary of its conclusion[6].

The salience challenge reaches us through a manuscript page[3]. We did not obtain that paper either.

We did not obtain Zajonc's 1968 monograph.

All arithmetic is ours, and the curve parameters are entirely invented.

This article discusses research on preference. It is not marketing, advertising or media buying advice.

The Effect

The claim, and the scale of the literature behind it.

A source describes the original: "Zajonc (1968) empirically demonstrated mere exposure, whereby repeated exposures of a particular stimulus, as long as there is nothing inherently negative about the stimulus, will result in a more positive attitude toward that particular stimulus."[1]

From the 1989 paper's own text: in the twenty years after the monograph, "over 130 articles reporting more than 200 separate experiments assessing the exposure-affect relationship appeared in the psychological literature."[5]

Two observations, ours.

The qualifier in the first quotation is doing real work. As long as there is nothing inherently negative about the stimulus. Repetition does not rescue something disliked; on this framing it amplifies a neutral or mild starting position.

And 200 experiments in two decades is a large literature by the standards of this series, which has more often dealt with famous findings resting on a single study.

How Large It Is

The number, and our conversions of it.

A citing paper reports "a robust effect size of r = 0.26" from the 1989 meta-analysis of 208 studies[1]. We did not obtain the meta-analysis and cannot confirm this against it.

Our own conversions, standard formulas: that is roughly d = 0.54, about 6.8 percent of variance explained, and a 79 percent overlap between the distributions.

Placed against what this series has reported: brainstorming penalty 1.39, mere exposure 0.54, the disputed nudge headline 0.43, feedback interventions 0.41, framing 0.31, growth mindset 0.08, choice overload 0.02.

Two observations, ours.

This is among the larger effects covered in forty-eight articles, and we would rather say that plainly than treat every finding as fragile by default.

And the two framings of the same number are both true. Six point eight percent of variance sounds negligible. Sixty-five percent probability of superiority sounds substantial. Neither is misleading, which is the twenty-eighth article's problem arriving again.

Everything Moderates It

The meta-analysis's actual conclusion, which is less quotable than the effect size and more useful.

From the abstract: "Stimulus type, stimulus complexity, presentation sequence, exposure duration, stimulus recognition, age of subject, delay between exposure and ratings, and maximum number of stimulus presentations all influence the magnitude of the exposure effect."[4]

Three observations, ours.

That is eight moderators in one sentence, and it is the abstract's central finding rather than an aside.

A headline figure averaged across a literature in which eight variables move the effect describes no particular situation. This series has made that observation about growth mindset, expectancy effects and loss aversion, and it applies here to a finding we are grading well.

And note that maximum number of stimulus presentations is on the list, which is the variable an advertiser actually controls.

Better When Nobody Notices

The finding that makes this article uncomfortable to write.

A description of the meta-analysis states: "A meta-analysis of research on Zajonc's (1968) mere exposure effect indicated that stimuli perceived without awareness produce substantially larger exposure effects than do stimuli that are consciously perceived."[2]

Two observations, ours.

Substantially larger, not merely present. The effect is stronger when the person cannot report having seen the thing.

Which inverts the usual assumption about advertising. Recall is treated as the measure of whether an impression landed, and on this finding recall may be the condition under which it works least well.

And It Was Tested Directly

Because a meta-analytic moderator is weaker evidence than an experiment designed to test it.

A study reports: "this finding has not been tested directly in the laboratory. Two experiments were conducted comparing the magnitude of the exposure effect produced by 5-ms (i.e., subliminal) stimuli and stimuli presented for longer durations (i.e., 500 ms). In both experiments, 5-ms stimuli produced significantly larger mere exposure effects than did 500-ms stimuli."[2]

It adds that the results held "for polygon, Welsh figure, and photograph stimuli"[2].

We did not obtain this paper and report no sample sizes or effect sizes from it.

Two observations, ours.

The authors noticed that a moderator found across studies is confounded with everything else that differs between those studies, and ran the direct comparison. That is the right instinct and this series has repeatedly wished for it elsewhere.

And it held across three stimulus types, which addresses the first moderator on the 1989 list.

Why That Is Uncomfortable

Naming the implication rather than leaving it. Ours.

Three points.

An influence that works better when unnoticed cannot be resisted by the person it acts on, because resisting requires noticing.

It also cannot be consented to in any meaningful way. The twenty-eighth article's test for honest framing was whether a person shown both descriptions would feel they had been told the same thing. That test cannot be run on something they did not perceive.

And we note this is a property of the mechanism, not a description of anyone's conduct. A firm placing ordinary visible advertising is not doing anything covert. The finding is about which exposures move preference most, not about what advertisers are doing.

The Inverted U

The limit, which is the part with direct budget consequences.

A source describes the meta-analysis's treatment: it "identifies satiation/boredom as a limiting factor to the effect, resulting in an inverted-U shaped relation between exposures and liking: the first exposures result in a positive increase on the liking of a stimulus, as exposures increase, liking attains a peak point where further exposures will only decrease it."[7]

Another notes limiting conditions "such as boredom", citing Bornstein, Kale and Cornell (1990)[1]. The underlying account is attributed to a two-factor model proposed by Berlyne[7].

Two observations, ours.

An inverted U means there is a wrong answer on both sides. Too little exposure and the effect has not accumulated; too much and it has reversed.

We did not obtain any source stating where the peak is, in exposures or in time, and none of our sources gives one.

We Modelled The Shape

To show what the described dynamics imply. The parameters are entirely ours and invented; no source gives a saturation rate, a tedium rate, or a peak. This demonstrates that a peak exists under the described dynamics, not where it is.

We modelled two opposing forces: a familiarity gain that saturates, and tedium that accumulates linearly.

At 3 exposures, net liking 0.30. At 6, 0.45. At 10, 0.51, the peak. At 15, 0.47. At 20, 0.36. At 30, 0.09. At 40, negative.

Three observations.

The curve is asymmetric. It climbs quickly and declines slowly, which means the cost of slight overexposure is modest and the cost of gross overexposure is that you end up below where you started.

The region around the peak is flat, so precision is not required. This is the same shape as the thirty-seventh article's advice-weighting curve, and the practical lesson is the same: the decision that matters is order of magnitude, not fine tuning.

And because the parameters are invented, the numbers above should be read as a shape and nothing else. A firm that took ten exposures from this article as a target would be taking a number we made up.

The Challenge

The claim that would reframe the whole literature.

A manuscript proposes that "repeated exposure to stimuli influences evaluations by increasing salience, the relative quality of standing out from other competing stimuli, which partially explains effects previously attributed to absolute 'mere' exposure."[6]

The mechanism it describes: "When exposure is manipulated among stimuli within the same presentation, this likely manipulates the relative salience of stimuli in addition to absolute exposure. In contrast, when only one stimulus is presented, that stimulus is likely salient whether it is presented many times or few times, because there are no other stimuli competing for salience."[3]

We did not obtain this paper and report its proposal as a proposal.

Two observations, ours.

The distinction is between how often you saw it and how much it stood out relative to what else was there. Most experiments confound the two.

And note the hedge: partially explains. The authors are not claiming the effect is entirely salience.

No Significant Effect Alone

The empirical claim attached to that proposal, and it is the most consequential sentence in this article.

"According to a meta-analysis, there was a moderate effect size of mere exposure on increased liking in heterogeneous presentation experiments, but no significant effect in homogeneous" ones, with our source truncating[3].

Three observations, ours.

In plain terms: where several different things were shown and some appeared more often, the effect appeared. Where one thing was shown alone, it did not reach significance.

We report this at one remove, from a manuscript's characterisation of a meta-analysis we did not obtain, and grade it C accordingly. We also cannot tell from our source which meta-analysis is meant, though the 2017 re-examination is the likely candidate.

And it is a structural claim, not a psychological one. If the effect requires competitors, then what repetition does is change relative standing, and an experiment with a single stimulus removes the thing that was doing the work.

What That Would Mean

Following it through, if it holds. Ours.

Three consequences.

Frequency would be a relative variable, not an absolute one. Appearing ten times matters because competitors appeared twice, not because ten is a number with properties.

Which means the same spend does different work in different markets. In a crowded category, frequency buys relative salience. In a category where you are the only advertiser, on this account, it may buy considerably less.

And it would explain the eight moderators in the 1989 abstract better than an absolute-exposure account does. If what matters is standing out, then everything about the presentation context should move the effect, which is precisely what that meta-analysis found.

The Pattern Again

Placing this alongside the series. Ours.

Three points.

This is the fifth time in this series that a well-known effect has turned out to have a live structural reinterpretation: after the Dunning-Kruger chart reproducible from noise, the judges' decline reproducible from time management, the hot hand debunking resting on a selection artefact, and the confirmation bias task using the one configuration where its strategy cannot work.

The difference here is instructive. In the earlier cases the reinterpretation weakened the finding. Here it may relocate it. A salience account does not say repetition fails to increase liking; it says the reason involves competitors.

And the practical advice barely changes either way, which is worth noticing. Whether frequency works absolutely or relatively, a firm still buys frequency; it just thinks about the competitive context differently.

What This Means For Marketing Spend

The application. Ours, untested, and offered with the grades attached.

Four points.

There is a wrong answer on both sides. An inverted U means underexposure wastes the effect and overexposure reverses it, and no source we found states where the turn happens.

Judge frequency against competitors, not against a rule of thumb. If the salience account is right, the relevant question is share of impressions in the category rather than a count.

Do not use recall as the sole measure of whether an impression worked. On the meta-analysis and a direct test, unrecalled exposures produced larger effects, which means a recall metric systematically misses the exposures doing the most work.

And repetition will not rescue a bad offer. The effect is described as operating where there is nothing inherently negative about the stimulus, which is a boundary condition and not a footnote.

The Line We Would Draw

Ours, in the manner of the twenty-eighth article.

Three positions.

Ordinary visible advertising is not covert influence, and nothing here suggests otherwise. That the effect is larger below awareness is a fact about the mechanism, not an accusation about practice.

Deliberately engineering exposure to be unnoticed is a different act, and the test we would apply is the one from the twenty-eighth article: would the person, shown what was done, feel they had been told the same thing? That test cannot be passed by something designed not to be perceived.

And the honest middle is the ordinary case: a firm that shows up consistently, visibly, in places its audience expects to see it, is using the mechanism in the form that survives being explained.

What To Do

Treat frequency as having a peak. The described dynamics imply an inverted U, so more is not monotonically better, and no source states where the turn is.

Do not fine-tune it. On our own invented curve the region around the peak is flat, so order of magnitude is the decision and precision is not.

Measure share of impressions, not just count. If the salience proposal holds, frequency does its work relative to competitors.

Stop treating recall as the test. Unrecalled exposures produced larger effects on the meta-analysis and on a direct experimental comparison.

Fix the offer first. The effect is stated to hold where nothing is inherently negative about the thing being repeated.

Distrust the headline number. The same abstract lists eight variables that move the effect, so an average across that literature describes no particular campaign.

Note this is one of the stronger findings here. At roughly d = 0.54 on a citing source's figure, it sits above most of what this series has covered, and we would rather say so than hedge uniformly.

The Limits Of This Analysis

Several caveats matter. This article discusses research on preference and is not marketing, advertising or media buying advice. Everything is verified to August 2026. We did not obtain the 1989 meta-analysis, and have its abstract verbatim from two independent sources plus one passage of its text. The headline effect size of r = 0.26 comes from a paper citing that meta-analysis, not from the meta-analysis itself, and we could not confirm it against the source. We did not obtain the 2017 re-examination, the subliminal comparison study, the manuscript proposing the salience account, or Zajonc's 1968 monograph, and report titles, citations and quoted passages only. The claim that there was no significant effect in homogeneous presentation experiments is graded C: it reaches us from a manuscript's characterisation of a meta-analysis we did not obtain, our source truncates mid-sentence, and we cannot confirm which meta-analysis is meant. All arithmetic is ours; the effect size conversions use standard formulas applied to a figure we could not verify, and the inverted-U curve parameters are entirely invented, with no source stating a peak, a saturation rate or a tedium rate, so the numbers demonstrate a shape and nothing more. The marketing applications and the section on where the line falls are our own reasoning, untested.

Related Reading

No. 49 Real, And It May Not Pay A gift raised charitable donation frequency by 75 percent in a field experiment of 10,000 letters. A wage gift raised output 25 percent and the effect... Behavioural Finance Three Different Things Called One Name The test for whether a presentation choice is framing or something worse. Behavioural Finance Ninety-Nine The other commercial effect with real evidence and heavy moderation.

Frequently Asked Questions

Does repetition really increase liking?

On the evidence we found, yes, under conditions. A citing source puts the 1989 meta-analysis of 208 studies at r = 0.26, which converts to about d = 0.54 and is among the larger effects covered in this series. The boundary condition is that nothing be inherently negative about the thing repeated.

Is more always better?

No. A source describes an inverted-U relationship: liking rises to a peak and further exposures reduce it, with satiation and boredom as the limiting factors. We found no source stating where the peak sits, and the curve we modelled uses parameters we invented entirely.

Why does it work better subliminally?

We cannot tell you why, only that it does on the evidence we saw. The meta-analysis reported substantially larger effects for stimuli perceived without awareness, and a later study tested that directly, comparing 5-millisecond and 500-millisecond presentations, and found the shorter produced larger effects across three stimulus types.

What is the challenge to it?

A proposal that repeated exposure works by increasing salience, meaning relative standing among competing stimuli, rather than by absolute repetition. The attached empirical claim is that a meta-analysis found a moderate effect where multiple stimuli competed and no significant effect where one was shown alone.

How confident should I be in that challenge?

We grade it C. It reaches us through a manuscript we did not obtain, characterising a meta-analysis we also did not obtain, in a sentence that truncates before it finishes. It is the most consequential claim in this article and the least well sourced, which is worth stating plainly.

What should I actually change?

Judge frequency against what competitors are doing rather than against a rule of thumb, stop using recall as the sole test of whether an impression worked, and accept that there is a wrong answer on both sides of the frequency question without pretending anyone knows exactly where it turns.

References

1. Repository page for the 1989 meta-analysis carrying scholarly citing text, on Zajonc (1968) having empirically demonstrated mere exposure, whereby repeated exposures of a stimulus, as long as there is nothing inherently negative about it, result in a more positive attitude toward it; on both traditional literature reviews and meta-analyses of the hundreds of studies documenting the association between repetition and liking, citing Bornstein (1989) and Montoya, Horton, Vevea, Citkowicz and Lauber (2017), concluding that the mere exposure effect is robust; on there being limiting conditions such as boredom, citing Bornstein, Kale and Cornell (1990); and on Bornstein's 1989 meta-analysis of 208 studies having established that repeated exposure to a stimulus increases preference for it, with a robust effect size of r = 0.26, with the effect operating even under subliminal conditions and following an inverted-U curve first described by Berlyne (1971). Note: a repository page reproducing text from papers citing the meta-analysis. The effect size of r = 0.26 comes from this citing text and not from the meta-analysis itself, which we did not obtain. researchgate.net
2. Repository page for the 2017 re-examination carrying citing text, on a meta-analysis of research on Zajonc's mere exposure effect having indicated that stimuli perceived without awareness produce substantially larger exposure effects than do stimuli that are consciously perceived, citing Bornstein (1989a); on that finding not having been tested directly in the laboratory; on two experiments having been conducted comparing the magnitude of the exposure effect produced by 5-millisecond subliminal stimuli and stimuli presented for 500 milliseconds; on 5-millisecond stimuli having produced significantly larger mere exposure effects in both experiments; and on those results having been obtained for polygon, Welsh figure and photograph stimuli. Note: a repository page reproducing an abstract of the direct comparison study, which we did not obtain. No sample sizes or effect sizes are reported here because we do not have them. researchgate.net
3. Manuscript page proposing a salience theory of mere exposure, on exposure manipulated among stimuli within the same presentation likely manipulating the relative salience of stimuli in addition to absolute exposure; on a single presented stimulus likely being salient whether presented many or few times because no other stimuli compete for salience; and on a meta-analysis having found a moderate effect size of mere exposure on increased liking in heterogeneous presentation experiments but no significant effect in homogeneous ones, our source truncating. Note: a manuscript page, not a published paper we obtained. The homogeneous-presentation claim is the most consequential in this article and the least well sourced; we cannot confirm which meta-analysis is meant and grade it C accordingly. psycnet.apa.org
4. Academic database record for Bornstein, Robert F., Exposure and affect: Overview and meta-analysis of research, 1968-1987, Psychological Bulletin, 106(2), 265–289, September 1989, DOI 10.1037/0033-2909.106.2.265, reproducing the abstract, on a review and meta-analysis of methodological and subject variables influencing the exposure-affect relationship having been performed on studies published in the 20 years following Zajonc's 1968 monograph; and on stimulus type, stimulus complexity, presentation sequence, exposure duration, stimulus recognition, age of subject, delay between exposure and ratings, and maximum number of stimulus presentations all influencing the magnitude of the exposure effect. Note: an academic database record with full text access restricted. We obtained the abstract and not the meta-analysis. proquest.com
5. Study resource copy of the 1989 paper's opening pages, reproducing the abstract identically and adding that modifications of the two-factor model of exposure effects are described; recording that Zajonc found a negative relationship between exposure frequency and mean galvanic skin response change, suggesting repeated unreinforced exposure results in habituation and decreased autonomic arousal; and stating that in the 20 years following the 1968 monograph, over 130 articles reporting more than 200 separate experiments assessing the exposure-affect relationship appeared in the psychological literature, across topics including advertising effects, children's reactions to novel stimuli, subliminal influences on behavior and social perceptions. Note: a study resource copy of the paper's opening pages. Our source for the scale of the literature and for the two-factor model reference; we did not obtain the meta-analysis itself. studocu.com
6. Bibliographic service record for Montoya, R. M., Horton, R. S., Vevea, J. L., Citkowicz, M., & Lauber, E. A. (2017), A re-examination of the mere exposure effect: The influence of repeated exposure on recognition, familiarity, and liking, Psychological Bulletin, 143(5), 459–498, DOI 10.1037/bul0000085, summarising its suggestion that repeated exposure to stimuli influences evaluations by increasing salience, the relative quality of standing out from other competing stimuli, which partially explains effects previously attributed to absolute mere exposure. Note: a bibliographic service's summary. We did not obtain the 2017 re-examination, its data or its effect estimates. semanticscholar.org
7. Academic preprint literature section, on Bornstein's meta-analysis of over 20 years of studies elaborating how factors such as stimulus type and complexity, presentation sequence, and individual traits including age and personality modulate the magnitude of the effect; on Bornstein identifying satiation and boredom as a limiting factor, resulting in an inverted-U shaped relation between exposures and liking in which the first exposures increase liking, liking attains a peak, and further exposures decrease it; and on Bornstein calling attention to the two-factor model for stimuli preference proposed by Berlyne. Note: an academic preprint's literature section, not a primary source. Our description of the inverted U comes from here and from reference 1; no source we obtained states where the peak lies. arxiv.org

This article discusses research on preference and is not marketing, advertising or media buying advice. Neither meta-analysis discussed was obtained. The headline effect size comes from a paper citing the 1989 meta-analysis rather than from the meta-analysis itself. All arithmetic is the authors' own, and the inverted-U curve parameters are entirely invented, demonstrating a shape rather than any quantity; no source obtained states where the peak lies.