Reach, frequency and gross rating points have exact published formulas. The three-exposure rule, on the other hand, inverts what its author actually argued, rests on no original experiment, and appears in no measurement standard anywhere.

Media planning is an unusual field in that its vocabulary is rigorously standardized and its rules of thumb are almost entirely undocumented. Separating the two is most of the value here.

This page gives the standardized definitions, traces the famous rule to its source, and says plainly what a small B2B company can and cannot compute.

What is genuinely standardized

The measurement council publishes formulas. They are worth knowing precisely, because they constrain what any of the derived numbers can mean.

Reach. Unique users, unduplicated homes or audience who have been exposed to ads, meaning who have generated a viewable impression, at least once during a time period. The formula: unique audience with a viewable impression, divided by the measured population or target, multiplied by 100.

Frequency. The number of times a user, home or audience generated a viewable impression and contributed to reach within a session or time period, expressed as an average. The formula: total viewable impressions divided by unique audience with a viewable impression.

Gross rating points. The sum of all ratings for an advertisement or campaign, reported as a gross number. Reach multiplied by frequency equals gross rating points.

The viewable impression these all rest on. For display, currently stated as at least 50% of the ad viewed for one continuous second or more.

Note what that makes frequency. An average over people who were reached at least once. It says nothing about the distribution, so a campaign with an average frequency of 4 might have shown the ad four times to everyone, or twice to most people and twenty times to a few.

Which is the practical warning. Average frequency is the least informative number on a media plan, and it is usually the one people argue about.

The standards prescribe no frequency level

This is the finding that reframes most media planning conversations.

What was checked. Every occurrence of recommendation language in the audience measurement standard.

What those occurrences are about. Audit procedure. Annual certification, disclosure to advertisers, the time zone used in reporting.

What they are never about. An effective exposure level, a recommended frequency, or any threshold at which advertising is said to work.

What the standards bodies actually do. Define how to count reach, frequency, impressions and gross rating points, and define what makes an impression countable. That is the entire remit.

So where does “you need X exposures” come from. Not from any measurement standard. It comes from a 1972 paper, and from what the industry did to it.

The published measurement formulas for reach and frequency and the absence of any prescribed effective levelTable presenting the formulas published by the industry measurement council for reach, frequency and gross rating points, alongside the observation that those same standards prescribe no recommended level of frequency. Reach represents unique users, unduplicated homes or audience who have been exposed to advertisements, meaning who have generated a viewable impression, at least once during a time period, and is calculated as the sum of unique audience with a viewable impression divided by the measured population, universe or target, multiplied by one hundred. Frequency represents the number of times a user, home or audience generated a viewable impression and contributed to reach within a session or time period, expressed as an average, and is calculated as the sum of viewable impressions divided by the sum of unique audience with a viewable impression. Gross rating points are defined as the sum of all ratings for a specified advertisement or advertising campaign reported as a gross number, with reach multiplied by frequency equalling gross rating points. The viewable impression underlying all three is currently stated for display advertising as at least fifty percent of the advertisement being viewed for one continuous second or more. A significant consequence of the frequency formula is that it produces an average across people reached at least once and therefore conveys nothing about the distribution, so a campaign reporting an average frequency of four might have shown the advertisement four times to everyone, or twice to most people and twenty times to a small number, making average frequency the least informative figure on a media plan despite being the one most frequently debated. Searching the audience measurement standard for recommendation language returns only matters of audit procedure, namely annual certification, disclosure to advertisers and the time zone used in reporting, and never an effective exposure level or any threshold at which advertising is said to work.Defined precisely, and deliberately incompleteStandardized, with published formulasReachunique audience with a viewable impression ÷ measured population × 100”exposed to ads… at least once during a time period”Frequencytotal viewable impressions ÷ unique audience with a viewable impressionAn average. It tells you nothing about the distribution.Gross rating pointsreach × frequencyAnd the viewable impression underneath: 50% of the ad, 1 continuous second, for display.Not standardized anywhere: how much frequency is enoughWhat “recommend” refers to in the standardAnnual certification. Disclosure to advertisers.The reporting time zone. Audit procedure only.What it never refers toAn effective exposure level. A recommendedfrequency. Any threshold at which ads work.
Formulas for counting, published precisely. Nothing at all about how much frequency is enough. Source : MRC Digital Audience-Based Measurement Standards (2017)

The three-exposure rule says the opposite of what its author argued

The paper exists, it is retrievable, and reading it is the fastest way to lose confidence in the rule built on it.

The source. A December 1972 paper in an advertising research journal, four pages long, titled around the idea that three exposures may be enough. Note the direction in the title itself.

What the author sets out to argue, verbatim. That he would like to argue against single exposure potency and also against any large number of repeated exposures, and that he stops at three because there is no such thing as a fourth exposure psychologically: fours, fives and so on are repeats of the third exposure effect.

So the claim is a ceiling, not a floor. The argument is that additional exposures beyond the second or third add little, not that three are required before anything happens. The industry inverted it.

The three stages he actually describes. The first exposure is by definition unique and produces a “What is it?” response. The second produces a more evaluative and personal “What of it?” response. By the third, the viewer knows he has been through both, and the third becomes the true reminder.

Whether he ran an experiment. No. He states that the special importance of two or three exposures is attested to by a variety of converging research findings based on different research methods, and cites three prior studies: his own 1968 eye-movement work on print, a 1969 study of responses to television spots, and a 1970 conference paper combining purchase diary and media data.

What came later. A 1995 book using single-source panel data argued that effective frequency is in fact one, and that continuity of presence beats concentrated repetition. A planning school built on the same idea, arguing that advertising effects decay quickly so weekly reach matters more than accumulated frequency.

What to take from all of it. There is a genuine debate about diminishing returns from repetition. There has never been a standardized minimum, and the paper everyone cites for one argues against large repetition rather than for a threshold.

The argument of the nineteen seventy-two exposure paper compared with the industry rule derived from itComparison of what the nineteen seventy-two paper on advertising exposure actually argues with the rule the advertising industry subsequently derived from it. The paper, published in December nineteen seventy-two in an advertising research journal and four pages in length, states that its author wishes to argue against single exposure potency and also against any large number of repeated exposures, and that he stops at three because there is no such thing as a fourth exposure psychologically, since fours, fives and subsequent exposures are repeats of the third exposure effect. The argument is therefore a ceiling describing diminishing returns rather than a floor describing a minimum requirement. The three stages the paper describes are that the first exposure is by definition unique and produces a what is it type of cognate response, the second produces a more evaluative and personal what of it response, and by the third the viewer knows he has been through both of these and the third becomes the true reminder. The paper reports no original experiment, stating instead that the special importance of two or three exposures compared to a much larger number is attested to by a variety of converging research findings based on different research methods, and citing three prior studies: the author’s own nineteen sixty-eight eye movement work on print advertising, a nineteen sixty-nine study of responses to television spots, and a nineteen seventy conference paper combining purchase diary data with media data. Subsequent work challenged the derived rule, including a nineteen ninety-five book using single-source panel data which argued that effective frequency is in fact one and that continuity of presence outperforms concentrated repetition, and a planning school arguing that advertising effects decay quickly so that weekly reach matters more than accumulated frequency.The paper argues the opposite of the ruleWhat the 1972 paper says”argue against single exposure potencyand also against any large number ofrepeated exposures""there is no such thing as a fourthexposure psychologically”A ceiling of diminishing returns.What the industry made of it”You need at least three exposuresbefore an ad works.”Used to justify buying more frequencyagainst a narrower audience, which isthe opposite prescription.A floor you must reach.And the paper ran no experiment of its own”attested to by a variety of converging research findings based on different research methods”,citing three prior studies. It is a synthesis essay, not original work.What came afterA 1995 single-source panel analysis argued effective frequency is in fact one, and that continuity beats repetition.
The paper argues against large numbers of exposures. The rule built on it says you need at least three. Source : Krugman, Why Three Exposures May Be Enough, Journal of Advertising Research (1972)

The numbers you have been given did not come from the places running your ads.

The largest social platform. Documents a reach and frequency buying type letting advertisers control how many times people see their ads, and describes it as valuable for understanding who you reach and how often. It describes the control. It names no target.

The search platform. On frequency capping, its guidance is that you can let the system optimise how often your ads show, and it marks that option as recommended. Again a mechanism, not a number.

The professional network. Defines average frequency as the average number of impressions shown to each member account that received at least one impression. A definition, and nothing else.

What that leaves. The specific figures in circulation, five impressions a week, three to five for display, six to ten a month on the professional network, appear in none of the documentation I could open. They come from agency blogs.

Why the platforms decline. Reasonably, because the answer depends on the message, the category, the creative and the buying cycle. A universal number would be wrong for almost everyone.

What that means for your plan. A frequency target in your media plan is a decision you made, not a standard you followed. Write it down as such, with the reason, so someone can revisit it.

What advertising platforms publish about frequency compared with the frequency figures circulating in the industryTable comparing what each major advertising platform actually publishes about advertising frequency with the specific frequency figures that circulate widely in the industry. The largest social platform documents a reach and frequency buying type which it describes as allowing advertisers to control how many times people see their advertisements, and as a valuable tool for understanding who is reached and how many times the audience sees the messaging; it describes the control mechanism and names no target level. The search platform, on the subject of frequency capping, states that the advertiser can let the system optimise how often advertisements show, and marks that automatic option as recommended, again describing a mechanism rather than a number. The professional network defines average frequency as the average number of impressions shown to each member account that received at least one impression, which is a metric definition and nothing further. None of the three publishes a recommended frequency level. The specific figures widely quoted in the industry, including five impressions per week, three to five for display advertising and six to ten per month on the professional network, appear in none of the platform documentation that could be opened and originate instead from agency publications. The platforms plausibly decline to publish a number because the appropriate frequency depends on the message, the category, the creative and the buying cycle, so any universal figure would be wrong for most advertisers. The practical consequence for an advertiser is that a frequency target appearing in a media plan represents a decision the advertiser made rather than a standard the advertiser followed, and should be recorded as such together with its reasoning so that it can later be revisited.Nobody running your ads publishes a numberPlatformWhat it publishesTarget given?Largest socialA buying type to “control how many times people see”NoSearch”let Google Ads optimize how often your ads show”NoProfessional networkA definition of average frequency, and nothing elseNoThe numbers in circulation”5 a week”, “3 to 5 for display”, “6 to 10 a month”.In none of the documentation. Agency blogs.Why they decline, reasonablyIt depends on message, category, creative andbuying cycle. A universal number is wrong for most.So a frequency target in your plan is a decision you made, not a standard you followedWrite it down with its reason, so somebody can revisit it later.
Three mechanisms, one metric definition, and no target anywhere. The weekly figures come from agency blogs. Source : Platform help documentation, read directly (2026)

Excess share of voice, with its own confidence interval

The one quantified planning rule with a published sample also publishes how much it explains, and that number is rarely quoted.

The report. A 2017 industry effectiveness study, drawing on a databank of award submissions.

The sample, verbatim. 497 for-profit cases submitted since 1998, and 118 submitted in 2014 and 2016, with 58% of the 2014 and 2016 cases supported by full econometric models.

The coefficient. Roughly 0.6 points of market share growth per 10 points of excess share of voice, in both the 1998 to 2006 and 2008 to 2016 periods.

The number nobody quotes. The R-squared: 6% in the earlier period, 12% in the later one. Excess share of voice explains between a sixteenth and an eighth of the variation in market share growth between campaigns. Everything else is creative, category, budget and luck.

The funding, disclosed on the report’s own front matter. Made possible through the support of a search company and a television marketing body. The report also argues for mass reach against narrow digital targeting, which is a conclusion those two sponsors benefit from. The statistics are not thereby wrong; the sponsorship belongs in the citation.

Three coefficients circulate and they are not the same study. 0.5, 0.6 and 0.7 are attributed variously to a 2013 report, this 2017 one, and a commissioned B2B cut. Only the 0.6 is verified here. Do not average them.

The excess share of voice coefficient and the proportion of variance it explainsTable presenting the excess share of voice finding from a two thousand and seventeen industry effectiveness report together with the explanatory power figure that its citations routinely omit. The report analysed four hundred and ninety-seven for-profit cases submitted since nineteen ninety-eight and one hundred and eighteen submitted in two thousand and fourteen and two thousand and sixteen, drawn from a databank of award submissions, with fifty-eight percent of the two thousand and fourteen and two thousand and sixteen cases supported by full econometric models. The published coefficient is approximately zero point six points of market share growth per ten points of excess share of voice, and this figure is the same in both the nineteen ninety-eight to two thousand and six period and the two thousand and eight to two thousand and sixteen period. The figure that is rarely quoted alongside it is the R-squared, which is six percent in the earlier period and twelve percent in the later one, meaning that excess share of voice explains between one sixteenth and one eighth of the variation in market share growth observed between campaigns, with the remainder attributable to creative quality, category, total budget and chance. The report’s own front matter discloses that the publication was made possible through the support of a search company and a television marketing body, and the report argues for mass reach against narrow digital targeting, a conclusion from which both sponsors benefit, which does not invalidate the statistics but belongs in any citation. Three different coefficients circulate in the industry, namely zero point five, zero point six and zero point seven, attributed variously to a two thousand and thirteen report, this two thousand and seventeen report, and a commissioned business-to-business analysis; only the zero point six figure is verified here and the three should not be averaged or conflated.The coefficient, and how much it explainsPeriodShare points per 10 pts ESOVVariation explained1998 to 20060.66%2008 to 20160.612%The sample, verbatim497 for-profit cases since 1998, plus 118 from2014 and 2016. 58% of those had econometrics.What 12% meansExcess share of voice explains about an eighthof the variation. Creative and category do more.Disclosed on the report’s own front matter”made possible through the generous support of” a search company and a television marketing body.Three coefficients circulate: 0.5, 0.6 and 0.7. They are three different studies. Do not average them.
0.6 points of share per 10 points of excess share of voice. And an R-squared of 6% to 12%. Source : Binet and Field, Media in Focus, IPA (2017)

What a small B2B company can actually compute

The honest answer to the planning question, and it is not the one the textbooks give.

What the classic models need. Reach curves and audience duplication models are published mathematics, freely available. Their inputs are not: exposure frequency distribution by media property, and audience duplication between properties, which come from proprietary panels.

What that costs. A panel subscription, at a price built for national advertisers. A company spending five figures a month on advertising is not the customer for that data, and cannot buy the inputs the models require.

What is actually available to you. The platforms’ own reach estimators, inside their campaign tools. These are usable and they are the right instrument at this scale.

What they are not. Published methods. They are proprietary, unaudited, not comparable between platforms, and not documented in any paper you can check.

So the answer to “how do I build a reach-based media plan”. At small scale, you do not build one from published method. You use the platform estimators, you write down what they told you, and you treat the output as a planning assumption rather than a forecast.

And what to plan on instead. Money, and outcomes. What you will spend, over what period, on which audiences, judged on cost per meeting held. That is computable from data you own.

What to put in the plan

Six items, all of which you can defend to a finance director.

The audience, defined as a list or a set of criteria you can actually target. Not a persona document. A targetable definition.

The money, by period and by platform. With a stated minimum viable amount per platform, below which you are buying noise.

The message, and how many versions of it. One promise per audience, and the number of executions you can actually produce and maintain.

The measurement, chosen in advance. Which number decides whether this worked, and where it comes from. Decide before, because deciding after is how plans get marked as successes.

The assumptions, written as assumptions. Including any frequency target, marked as a choice you made and not a standard you followed.

The stopping rule. What result, by what date, causes you to stop or change. Plans without one run until the budget does.

What a small advertiser can and cannot compute when building a media planDiagram distinguishing what a small business-to-business advertiser can compute when building a media plan from what requires data it cannot obtain. The classic reach models, including reach curves and audience duplication models, are published mathematics that is freely available, but their required inputs are not: they need exposure frequency distribution by media property and audience duplication between properties, both of which originate in proprietary audience measurement panels sold at prices constructed for national advertisers, so a company spending a five-figure monthly sum on advertising cannot purchase the inputs those models require. What is actually available at that scale is the reach estimators built into the advertising platforms’ own campaign tools, which are usable and are the appropriate instrument at this size, but which are proprietary, unaudited, not comparable between platforms and not documented in any publicly checkable methodology. The honest conclusion is therefore that a small advertiser does not build a reach-based media plan from published method: it uses the platform estimators, records what they reported, and treats the output as a planning assumption rather than a forecast. What can be planned reliably instead is money and outcomes, specifically what will be spent over what period on which audiences, judged on cost per meeting held, all of which is computable from data the advertiser already owns. The plan itself should contain six defensible items: a targetable audience definition rather than a persona document; the money by period and platform with a stated minimum viable amount per platform; the message and the number of executions that can actually be produced and maintained; the measurement chosen in advance including where the number comes from; the assumptions written explicitly as assumptions including any frequency target; and a stopping rule specifying what result by what date causes the plan to stop or change.What you can compute at your scalePublished models you cannot useReach curves and duplication models arepublic mathematics.Their inputs are panel data priced fornational advertisers.Tools you can use, with a caveatThe platforms’ own reach estimators.Usable, and the right instrument here.Proprietary, unaudited, not comparablebetween platforms.What you can plan on reliablyMoney and outcomes: what you spend, over what period, on which audiences, judged on costper meeting held. All computable from data you already own.And the item most plans are missingA stopping rule: what result, by what date, causes you to stop or change.Plans without one run until the budget does.
The classic reach models need panel inputs. The platform estimators are usable and unauditable. Plan on money and outcomes. Source : MRC standards and platform planning tools (2026)

Where to go next

You want the buying mechanics underneath. Media buying explained.

You are deciding how many creatives to run. How many creatives to test.

You are weighing brand against performance spend. Brand vs performance in B2B.

You want to know why most B2B ads score badly. Why B2B ads fail.

You are splitting budget between platforms. How to split budget between Google and Meta.

You want the cost of an impression. CPM, cost per thousand.

In short

  • Reach, frequency and gross rating points have published formulas. Reach times frequency equals GRP, and the viewable impression underneath is 50% of the ad for one continuous second.
  • Average frequency hides its distribution, which makes it the least informative number people argue about most.
  • The measurement standards prescribe no frequency level. Every recommendation in them is about audit procedure.
  • The three-exposure rule inverts its source. The 1972 paper argues against large numbers of exposures and says there is no fourth exposure psychologically.
  • That paper ran no experiment, describing itself as resting on converging findings from three prior studies.
  • No platform publishes a recommended frequency. They document the capping tool, the automatic option and the metric definition.
  • Excess share of voice buys about 0.6 share points per 10 points, with an R-squared of 6% to 12%, in a report sponsored by a search company and a television body.
  • No published method lets a small company plan reach, because the models need panel inputs you cannot buy.

Plan money and outcomes, and write your assumptions down as assumptions. Book a diagnostic, or see how we approach B2B paid acquisition.