The Art of the Revisit: Why There Is No Magic Ad Frequency

Every advertiser eventually asks the same question. How many times should people see my ad? Three? Five? Eight?

This page won't give you a number. It will show you why there isn't one, in seven short lessons. Each lesson is a small simulation you can play with. Nothing here needs a background in math or media planning. You only need to remember things, and occasionally forget them.

Lesson one: Forgetting

In the 1880s, a German psychologist named Hermann Ebbinghaus spent months memorizing lists of nonsense syllables. WID. ZOF. KEB. He picked nonsense on purpose, so that nothing he already knew could help him. Then he measured how much was left after twenty minutes, an hour, a day, a month.

The result is called the forgetting curve. Memory drops fast at first, then more slowly. By the next day, he had to relearn about two thirds of what he had learned.

Ads mean more than nonsense syllables, but the curve has the same shape. Meet Sam. Sam just saw your ad.

Try the memorability slider. A great ad lasts longer than a forgettable one, but the curve keeps its shape. Even the best ad fades.

Forgetting is a given. What you control is when the next ad arrives.

Lesson two: The second time

So you show the ad again. The obvious question is when.

Drag the second TV along the timeline. The dashed grey line is what would happen without it. Watch how long Sam's memory lasts after the second ad, measured as a half-life: the time it takes for half of the memory to fade.

Too soon, and the second ad does almost nothing. Sam still remembers. There is nothing left to strengthen. You paid for an ad that changed nothing.

Too late, and Sam has mostly forgotten. The second ad still helps, because forgotten isn't gone. Ebbinghaus noticed it himself: relearning a list he could no longer recall always took less time than learning it the first time. But the ad helps less than it could have, and Sam spent days not remembering you in between.

In between, there's a sweet spot, when the memory has faded a lot, but not completely. Psychologists call this desirable difficulty: recalling something that is slipping away strengthens it much more than seeing something you already know well. The same effect shows up with ads (Appleton-Knapp, Bjork & Wickens, 2005).

Lesson three: Growing gaps

Now you have five ads. Your job: keep the ad in Sam's memory for as long as possible after the last one.

Before you touch anything, take a guess.

Same five ads, same budget. Spaced the right way, Sam remembers for weeks. Crammed into one evening, Sam remembers for about a day.

Notice that one ad every six days beats one a day, even though Sam has forgotten almost everything by the time each new ad lands. Forgotten isn't gone, so each late ad still builds on something. The daily ads come too close together: after the first couple, each one lands while Sam still remembers, and builds less and less.

The gaps should grow because every well-timed ad slows down the next round of forgetting. So the next sweet spot arrives later than the last one. Lab studies still argue about whether growing gaps beat equal gaps when the equal gap is well chosen (Karpicke & Roediger, 2007). They agree on the part that matters here: the best gap depends on how long the memory has to last.

In a study of more than 1,350 people, Nicholas Cepeda and his colleagues measured it (Cepeda et al., 2008). If the test is a week away, short gaps between study sessions win. If it's a year away, much longer gaps win. That is the whole point of lesson seven.

Lesson four: A whole class

Sam is one person. A campaign talks to a crowd. Here are 400 households.

In real life, you can't show your ad to anyone at any time. On streaming TV, you can only show it while a household is watching. Some households stream every night, others once a week. In this class, one household in five is a heavy viewer, and together they produce more than half of all the ad slots. That skew is everyday life in connected TV, and default delivery hands them more than half of your budget.

Now let's run a campaign the way most platforms do by default: split the budget evenly across the days, and buy the ad slots that come up. Before you press play, a guess.

Look at the histogram first. Almost a third of the class never sees the ad, while a few heavy viewers see it ten to fifteen times, often several times in the same evening. A daily cap trims the worst of it, but it doesn't decide when anyone should see the ad.

Then look at the blue line. Memory goes up after each ad and leaks away between them. Most of what you paid for, the class has already forgotten.

Now try Reach first. Almost everyone gets the ad. But the second ad waits until everyone has had a first one, about nine days later, and by then most households have forgotten. Average memory ends up no better than with even pacing. Reach is the right first step, and only a first step.

Lesson five: Total awareness

When a campaign looks wasteful, the usual reflex is to aim it better. That reflex has a reason. What an awareness campaign actually achieves is hard to see while it runs, and when you can't see the ends, you tighten the means. Narrow the audience down to the households that look most likely to buy. Start early, to warm them up. Cap the frequency. It feels like precision: fewer people, more impact on each.

Here is what it does. Take a 45-day campaign where only the last ten days count, like a launch or an election. Plan A is the reflex: a quarter of the class, from the first day. Plan B targets the whole class and starts a week before the exam. Same budget, same classic delivery. The only things that change are who you target and when you start.

Look at the share of each audience that remembers. Plan A wins easily, and that's the number most reports show you, because it's measured inside the target. It's also the wrong number to manage. What you need is the number of people who remember when it counts. Call it total awareness: how many people you reach, times how strongly they remember. On the chart, it's the area of each rectangle.

Plan B gets about half the memory per household, on four times the audience: twice as many people remembering during the exam. Plan A loses for two reasons you already know. The households it reaches see the ad fourteen times each, and seven ads in ten land too soon to add anything (lesson two). And it starts on day one, so much of what it builds has faded by the time the exam comes (lesson one).

Now use the four buttons above the chart to change Plan A one step at a time. Start it a week before the exam, and almost nothing happens: squeezed into fewer days, the same hundred households just get more repeats that land too soon. Widen it to the whole class instead, and awareness jumps from 56 to 101. Do both, and you have Plan B. Breadth is what makes a late start pay off, and the late start is what keeps memory fresh.

The fair objection is that a narrow audience is more likely to buy. The highlighted test under the chart puts a number on it: how much more likely Plan A's households would need to be just to break even. That's the number to bring to a targeting debate, and it's often more than targeting can honestly promise.

And memory isn't the only thing working against depth. In field experiments, the first ad a person sees tends to do the most, and each extra one does less, sometimes much less (Lewis, 2015). Push repetition far enough and it starts to annoy rather than persuade (Pechmann & Stewart, 1988). Everything points the same way: one ad each to two people usually beats two ads to one.

“Half the memory on an audience three times bigger is one and a half times the awareness.”

Lesson six: Delivery that remembers

What if the delivery knew what you now know? It would follow four rules.

Rules two and three pull against each other: a first ad for someone new, or a reminder for someone whose memory is fading?

Here are the first three rules up close: ten households, one week, the same three ads every evening. Each dot is an ad slot, a moment when a household is watching and an ad could run. The blue band behind each row is that household's memory of the ad. Switch between the two bidders, and tap any dot to see why it was bought or skipped.

Classic delivery spends most of the week on the two heavy viewers, because they produce most of the slots, and it often buys two in the same evening. Delivery that remembers skips those repeats, catches the light viewers on the rare evenings they watch, and comes back to each household once its memory has faded. Same 21 ads. All ten households reached instead of eight, no ad wasted, and at any given moment, 3.5 households remember the ad on average instead of 2.4.

Now scale it up to the whole class, and add the fourth rule. Pick a goal. On the left, classic even pacing. On the right, delivery that follows the four rules. Same budget, same households, same TV habits, both targeting the whole class and both scored on total awareness. The yellow band is the exam: the days that count. Then compare with Last-minute burst. It spends in the same final stretch, so most of the gap that remains comes from choosing who sees each ad.

Switch between the goals and watch the bars under the chart. For a launch or an election, delivery that remembers waits, pushes hard just before the window opens, then tapers off with spaced reminders, so a high share of households remembers for the whole window. For a flash sale it waits, then spends only in the week before the sale. For an always-on brand it spreads them out, and the share remembering keeps rising as each household's memory gets sturdier.

For the launch and the flash sale, it also reaches more households, not fewer: the money it stops spending on fifth ads to heavy viewers goes to someone new. Now drag the budget all the way down on the always-on goal. It does the opposite, and gives fewer households enough ads to stick rather than everyone one ad they will forget.

In 1959, Hubert Zielske mailed the same 13 ads to two groups of women: one group every week, the other every four weeks. The weekly group's recall climbed fast, then collapsed once the mailings stopped. The four-weekly group never peaked as high, but ended the year remembering far more (Zielske, 1959). Which schedule was better depends on when you needed them to remember.

Lesson seven: No magic number

Back to the question everyone asks: what's the right frequency? The famous rule of three comes from Herbert Krugman, and he was describing three stages of attention, not a media plan (Krugman, 1972).

Below, the same simulation looks for the best delivery plan for three different goals. It tries different start dates, ramps and priorities, and keeps the plan that leaves the most people remembering during the window. Change the budget and the audience size, and watch what the "right frequency" does.

There's no optimal frequency. There's an optimal exposure plan, and it follows from three things: the budget you have, the size of your audience, and when the exam is.

Frequency is an output, not an input. So is spacing. Push the budget up and the best plans reach more of the audience before they add reminders. Grow the audience with the same budget and they look for the breadth that maximizes total awareness: sometimes everyone, sometimes fewer people often enough to stick, never a small group hammered too soon. And the closer the exam, the shorter the gaps: compare the always-on brand with the flash sale, exactly like Cepeda's students.

This lines up with what the research on ad frequency keeps finding. In 30 field experiments at Yahoo!, the value of extra impressions varied widely: a few campaigns wore out fast, while about a third showed almost no drop-off (Lewis, 2015). Memory for an ad keeps building over several exposures in controlled settings (Schmidt & Eisend, 2015). And people are more likely to buy when the same ads are spread apart rather than bunched together (Sahni, 2015). A fixed frequency cap can't hold all of that at once. A plan built around memory can.

Final exam: Your campaign

Everything is unlocked. Set your own campaign length and exam window, your budget, how memorable your ad is, and which classic strategy to compare against. Try to find a setup where classic delivery wins. It's possible, and finding out when is a good way to check you understood the rest.

When can classic delivery win? When the exam starts within a day or two and only lasts a few days, spending everything as fast as possible wins, by a small margin. There is no time for spacing to pay off, so the best move is to reach as many people as you can, right now. Everywhere else, timing and choosing who sees each ad make the difference.

The same logic, in a live bidder

"What frequency should I use?" comes up in almost every awareness conversation I have. The useful answer takes ten minutes of play rather than one number, so I built the ten minutes.

Brand lift studies are still the best way to check what a campaign did. But they usually arrive after it's over, and they can't tell a bidder whether a given household should see the ad tonight. So at Vibe, Awareness campaigns now come with a Memory Score: an estimate of how much of your audience is likely to still remember the ad, built only from how many times and how recently it ran. In a campaign's Insights tab, you see the households you reached next to the households still likely to remember it.

Since our latest release, Awareness campaigns also run on Laurence, a bidder built on the logic you just played with. It lowers the value of a household that was just reached and restores it as time passes, so the money stops going to repeats that change nothing and goes where there is more memory to create. More total awareness, for the same budget.

Awareness isn't the opposite of performance. It's just rarely measured like it.

Methodology

This page runs a teaching model, small enough to fit in your browser and simple enough to reason about. It is not Vibe's production bidder, and its numbers are illustrations, not forecasts. Real memory varies with the person, the creative, the category and everything else competing for attention.

Memory after an ad decays exponentially. The first ad gives a half-life H₁ (1 day for a typical ad, set by the memorability slider). Each later ad resets memory to full and multiplies the half-life by a boost that depends on how much was left when it landed:

f is 0 when m is 1, peaks at 1 near m = 0.3, and is about 0.24 when m is 0. So an ad that lands while memory is still full adds nothing, an ad that lands around the sweet spot more than doubles the half-life, and an ad that lands after everything is forgotten still stretches it by about a third. That last part is Ebbinghaus's savings: a memory you can no longer recall still makes relearning easier. Robert and Elizabeth Bjork call it storage strength, which keeps building even after retrieval strength has faded (Bjork & Bjork, 1992). It's also why Zielske's four-weekly group kept building recall.

The class has 400 households: 20% heavy viewers (watching most days, five ad slots a night), 45% regular viewers and 35% light viewers. Ads can only run when a household is watching. "Remembering" on a given day is the average memory across all 400. Classic strategies buy whatever slots come up.

In lesson five, each plan targets the first households of a fixed random order and spends only inside that audience, so a narrow audience is a subset of a broader one. In lesson six's ledger, ad slots come up in a fixed random order each evening; classic delivery buys the first ones, while delivery that remembers ranks the households watching tonight and keeps any budget it doesn't need for the next evening. In the larger simulations, delivery that remembers ranks the households watching today by the value of showing them the ad now and never spends after the exam ends, and the page searches a small set of spending schedules (start date, ramp, priority between new and returning households) for the one that maximizes average memory during the exam window.

Sources and further reading

Credits

The format owes everything to Nicky Case, whose How To Remember Anything Forever-ish teaches spaced repetition by making you do it, and to Bret Victor's idea of explorable explanations. Words and model by Arnaud Blanchard.