In 1972, Herbert Krugman sat down at General Electric and wrote a paper arguing that three exposures to an advertisement might be enough. The logic was elegant in its simplicity: the first exposure asks "what is it?" The second asks "what of it?" The third produces a decision. Any additional exposures, Krugman suggested, were repeats of the third effect. This was the origin of the "three exposures" rule that would govern media planning for the next fifty years.
It was never validated. It was adopted.
Michael Naples formalized it in 1979 with an ANA book titled Effective Frequency, which became the field's foundational text despite containing, by later academic reckoning, almost no rigorous empirical support for its central claim. Jones (1995) would later study 142 brands and find that one exposure may be sufficient. Gibson (1996) found that a single exposure could produce major attitude changes. Yahoo!'s field experiments found linear returns continuing all the way to fifty exposures, with each additional impression worth approximately 3.7 cents in measurable sales lift. Yet the three-exposure rule persisted, not because the evidence supported it, but because it was useful — a planning heuristic that gave buyers permission to spend at levels that kept the industry's infrastructure running.
The frequency question — how many times must someone see an ad before it works — is one of advertising's oldest and least settled arguments. And the Elsen, Pieters, and Wedel paper published in February 2025 in the Journal of Marketing Analytics does not resolve that argument so much as it dissolves it from underneath.
The paper is titled "Effects of advertising exposure duration and frequency: a theory and initial test," and its central contribution is a concept called Identification Threshold Theory. The authors — Millie Elsen of Tilburg University and the University of Maryland, Rik Pieters of Tilburg University, and Michel Wedel of the University of Maryland — argue that the industry has been asking the wrong question. The question is not how many times someone is exposed. The question is whether each exposure lasted long enough for the brain to identify what it was looking at.
An identification threshold is the minimum time required for a viewer to recognize the advertised product. Below that threshold, repetition does not compound. A viewer who sees an ad for 50 milliseconds, twenty times, may have accumulated one full second of exposure — but if each individual exposure fell short of the threshold, the learning and evaluation effects that advertisers rely on did not occur. The total time matters less than whether each discrete exposure crossed the minimum.
This directly challenges the Total Time Hypothesis, which has underpinned media planning's frequency models since the 1970s. Under the Total Time Hypothesis, learning is a function of cumulative exposure — ten exposures of one second each should produce the same outcome as one exposure of ten seconds. Elsen, Pieters, and Wedel's experiments found this is not true. Duration per exposure matters independently of total duration.
The practical implication is uncomfortable: if the average digital display ad is visible for less than the typical identification threshold for that creative, then the frequency metric that media planners use as their primary optimization target is measuring something that may not be happening.
The paper's second finding complicates any universal rule about exposure duration. The authors found that identification thresholds vary significantly depending on the type of advertisement.
Typical ads — those conforming to category conventions, using expected formats, displaying familiar product categories in familiar ways — have lower identification thresholds. Viewers can recognize them faster. Atypical ads, which use unexpected formats, unfamiliar category presentations, or creative structures that require more cognitive processing to decode, have higher thresholds. They need more time per exposure to register. Earlier work by Elsen, Pieters, and Wedel (2016) identified three ad type patterns: "upfront" ads, which register positively at both brief and long exposures; "mystery" ads, which register negatively at brief exposures but positively at longer ones; and "false front" ads, which attract at brief exposure but alienate at longer ones.
The implications for media planning are significant. A brand running typical, formulaic creative in a high-frequency digital campaign may be fine with shorter individual exposures — the category-conformant format lowers the bar for recognition. A brand running unusual, distinctive, or complex creative — the kind that agencies win awards for — may be systematically underperforming because each exposure is too brief for the threshold to be crossed. The creative that breaks convention requires more time to process, but digital environments are engineered to deliver exactly the opposite: ultra-brief, scroll-past impressions optimized for completion rates rather than cognitive engagement.
The authors tested these relationships across two controlled experiments using 48 advertisements, manipulating both exposure frequency and exposure duration while varying ad typicality. The dataset is publicly available via Mendeley Data (DOI: 10.17632/dry79fnv2z.1). The findings on the typicality-by-duration interaction were robust across both experiments.
The Elsen, Pieters, and Wedel paper contributes to a broader literature on the time course of ad recognition, and the numbers are clarifying.
Research by Pieters and Wedel (2012, Marketing Science) established that consumers can distinguish an advertisement from editorial content at durations under 100 milliseconds — essentially at a glance. But this is gist discrimination, not product identification. Recognizing that an ad exists is different from recognizing what is being advertised.
A landmark study by the Mobile Marketing Association in partnership with Neurons Inc. and the Advertising Research Foundation, using EEG and eye-tracking, found that the human brain requires approximately 400 milliseconds to cognitively engage with mobile advertising. At that threshold, 67 percent of mobile ads were cognitively recognized. On desktop, where viewing conditions differ, two to three seconds were required before two-thirds of ads were cognitively recognized. Emotional responses — the affective processing that brands often rely on to drive preference — can form in as little as 700 milliseconds.
These are not marginal differences. They describe a gap between what the industry measures and what the brain requires.
Additional eye-tracking research from InSkin Media and Research Now found that display ads need to be visible for approximately fourteen seconds to accumulate one second of gaze time — the actual focused attention that drives encoding. To accumulate two seconds of gaze time, an ad needed to be present for roughly thirty-three seconds on average. Yet the IAB's standard viewability requirement — fifty percent of the ad visible for at least one second — is treated as a meaningful quality threshold by most programmatic systems.
One second of impression time is not equivalent to one second of cognitive engagement. The brain does not process what it does not have time to see.
The Elsen, Pieters, and Wedel paper enters a landscape where the frequency orthodoxy is already under pressure from multiple directions.
Sky Media's analysis of 135 studies found the optimal frequency window to be between eight and fourteen exposures — a significant upward revision from Krugman's three. Magna and Roku's research — published on Roku's advertising research portal — analyzing 53 campaigns across 14 verticals and 36,832 survey responses found general audiences reached maximum ad recall at ten to fifteen exposures, while in-market consumers — those already considering a purchase — peaked at just four exposures. The implication: frequency needs are heterogeneous, not universal, and the current industry practice of applying uniform frequency targets across heterogeneous audiences is a category error.
More fundamentally, the Yahoo! field experiments — a controlled experiment by Yahoo! Research with a major American retailer across over a million shared customers, published in Quantitative Marketing and Economics (2014) — found that returns from exposure continued linearly without visible diminishing effects up to fifty exposures in a two-week period, with each additional exposure worth approximately 3.7 cents in measurable sales lift. This suggests the entire concept of an "optimal" frequency may be a planning fiction imposed on a continuous response function.
What none of this prior work had adequately addressed was the duration question. Frequency planning assumes that exposures are comparable — that the seventh impression is meaningfully equivalent to the first. If identification thresholds mean that brief impressions are not equivalent to longer ones, then a campaign delivering twenty ultra-brief exposures may be producing fewer effective impressions than a campaign delivering eight exposures of sufficient duration. The metric being optimized may be measuring the wrong variable.
Media planning has operated for decades on a shared assumption: that impressions are fungible, that frequency is the primary driver of advertising effectiveness, and that the industry's preferred frequency targets — whether three, eight, or ten — represent something real about how human memory and recognition work.
The Elsen, Pieters, and Wedel paper suggests this assumption deserves the scrutiny that the original three-exposure rule never received. If identification thresholds vary by creative type, and if sub-threshold exposures fail to produce the learning and evaluation effects that justify the campaign spend, then the frequency models that govern billions of dollars in annual advertising investment may be optimizing for a number that has no consistent relationship to actual effectiveness.
The counterargument is that viewability standards have improved, that longer on-screen time has become more common as publishers optimized for viewable impressions, and that digital advertising has moved toward higher-quality inventory. This is partially true. Industry measurement platforms — including Integral Ad Science, DoubleVerify, and Moat — have reported meaningful increases in viewable impression rates over the past decade, though exact figures vary by platform, format, and vendor methodology.
But viewability — an ad being technically visible on screen — is not the same as cognitive engagement. An ad can be viewable for 1.2 seconds and technically pass the IAB standard while remaining below the 400-millisecond threshold at which the brain begins to cognitively process what it is seeing. The industry has been measuring a proxy for the thing it actually cares about, and optimizing that proxy, and calling it effectiveness.
The identification threshold is not a metaphor. It is a minimum. And minimums, by definition, cannot be wished away by better measurement.
The Elsen, Pieters, and Wedel paper provides a theoretical framework and initial empirical results. What it does not provide is a simple lookup table that media planners can use to set duration minimums for specific campaigns. The identification threshold for a specific ad in a specific context depends on ad typicality, viewer expertise, platform, creative complexity, and other variables that are not easy to specify in advance.
This is where the practical application of the research runs into the organizational reality of media planning. The systems that buy and optimize digital advertising operate on frequency and cost per outcome metrics. Duration per individual exposure is not a standard optimization target in most DSPs or campaign managers. Even if a planner wanted to ensure each exposure exceeded the identification threshold, the infrastructure to measure and act on that information is not broadly available.
The research points toward a shift in what is being optimized — from exposure count to exposure quality — and that shift requires changes in how creative is designed, how media is bought, and how effectiveness is measured. That infrastructure does not yet exist at scale.
What the Elsen, Pieters, and Wedel paper offers is a diagnosis: the industry has been treating the symptom (frequency) rather than the condition (threshold crossing). The symptom is easier to measure. The condition is what actually determines whether advertising works.
The paper "Effects of advertising exposure duration and frequency: a theory and initial test" by Millie Elsen, Rik Pieters, and Michel Wedel appeared in the Journal of Marketing Analytics, Vol. 13, pp. 386–404, February 2025. DOI: 10.1057/s41270-024-00373-4.