Almost every number you have read about ad creative traces back to about a dozen studies, and several of the most repeated ones have no source at all. This series sorts the evidence from the folklore: what is actually measured, by whom, on what sample — and what gets quoted anyway.
Creative is the largest driver of advertising performance an advertiser still controls, and the research says so from two independent directions. It is also the driver most advertisers manage worst, because the two things that govern it — attention and repetition — behave in opposite directions, and almost nobody plots them together.
The short version
Three findings carry the rest of this series:
- Creative accounts for 49% of incremental sales in NCSolutions' model of roughly 450 campaigns, against 11% for targeting. The same ranking appears in a separate econometric analysis of about 28,000 campaigns.
- Longer exposure raises memory; repeated exposure lowers response. Both curves are published. They point opposite ways, and the tension between them is the whole practical problem of creative.
- Meta measured its own creative fatigue and found no wear-in period at all — performance only ever gets worse with repetition. That contradicts the model taught in most marketing courses.
And one finding about the field itself: several of the statistics that anchor the popular version of this topic do not survive being looked up. They are listed near the bottom, with what happened to each one.
Creative is the biggest lever you still control
The platforms took targeting back. Apple's App Tracking Transparency cut off the signal that made audience targeting precise, and Meta and Google answered by automating the buying itself through Advantage+ and Performance Max. The dials that used to sit in the media buyer's hands moved inside the algorithm. The ad did not.
NCSolutions models what actually produces incremental sales across roughly 450 consumer-goods campaigns, and has published the breakdown twice — once in 2017 and again in 2023. Creative held first place both times. What moved underneath it is the interesting part.
Creative did not move. Everything else did. Reach fell from 22% to 14% while brand factors rose from 15% to 21% — which is what commoditisation looks like in a chart. Buying eyeballs stopped being a differentiator, because everyone can now buy the same eyeballs through the same automated systems at roughly the same price.
Targeting sits at 11%. Creative sits at 49%. That is the ratio the rest of this series is about.
One caution about this number. You will also see "47%" and "56%" attributed to the same research. The 49% is the figure NCSolutions actually publishes for the five-keys model, in both editions. The others are a different cut or a restatement, and are handled in the source notes below.
The two curves nobody draws together
Here is the part that changed how I read all of this.
There is a published curve showing how ad memory responds to exposure duration. There is a separate published curve showing how ad response falls with exposure repetition. They come from different researchers, different decades and different methods, and they are almost never shown side by side — because they measure different things on different axes. But they describe the same creative, in the same feed, on the same afternoon, and they run in opposite directions.
Read left to right and the practical instruction falls out. Every extra second a person spends with your ad is worth something, and the first ten seconds are worth the most — the curve is steepest at the start. Every extra time you show the same person the same ad is worth less than the time before, with no floor and no recovery.
So the job is not "get attention". The job is to get more attention per impression, because impressions are the thing that decays.
What a better creative is actually worth
Strip out everything else and the mechanism is arithmetic. You buy impressions at a cost per thousand. The creative decides what fraction of those impressions become clicks. Hold the budget and the price of impressions still, change only the creative, and the traffic moves in proportion.
At a $15 cost per thousand impressions, $100 buys about 6,667 impressions:
Five times the click-through rate, five times the traffic, one fifth the cost per click — for the same money.
And the real gap is wider than that, because the auction is not indifferent to your click-through rate. Platforms price partly on predicted engagement, so a creative that earns attention tends to buy its impressions more cheaply as well as convert more of them. I have not found a published figure that puts a reliable multiplier on that second effect, so I am not going to invent one. Take the arithmetic as the floor, not the estimate.
Creative decays, and Meta published its own decay curve
The frequency statistic the industry still quotes is from 2018. It is a table from AdEspresso showing click-through rate falling and cost per click rising as frequency climbs, and it is genuinely useful. It is also eight years old, drawn from about 500 campaigns, and predates almost every material change to how these platforms deliver ads.
I went looking for a modern replacement — a public study from 2023 onward with a frequency-versus-performance table and a stated sample. There isn't one. Not from the platforms, not from the major measurement and analytics firms, not in the trade press. The industry is still citing 2018 because 2018 is what exists.
What does exist, and is barely quoted, is Meta's own analysis of creative fatigue.
Four things in that post deserve to be better known than they are.
The average impression is not the first. Across all Meta ad impressions, the mean number of prior exposures of that creative to that person is 4.2. More than 19% of impressions go to someone who has already seen the creative more than five times. Whatever you think you are buying, a large share of it is repetition.
There is no wear-in. This is the one that should change how the subject is taught. The classic model says repetition helps at first — the audience needs a few exposures to register the message — and only then does decay set in, producing a hump-shaped curve. Meta went looking for that effect in its own data and did not find it. Clicks and conversions become monotonically more expensive with repeated exposure. No warm-up, no peak, just decline from the first impression onward. Meta notes that brand objectives such as ad recall may behave differently, which is a real limit on the claim, but for direct-response advertising the textbook curve is not what the data shows.
By the fourth exposure, conversion likelihood is down about 45%. Not click-through — conversion.
And refreshing the creative is worth about 8%. This is the part nobody quotes, and it is the most useful part. Meta ran a two-cell split test across roughly 26,000 cases over two seven-day phases, and the effect was dose-dependent: the more fatigued the creative, the bigger the gain from adding a new one. In high-fatigue cases, acting on the fatigue guidance improved conversion rate by an average of 8%.
Hold those last two numbers next to each other, because the honest story is in the gap. Fatigue costs you around 45%. Fixing it gets you about 8% back. Creative refresh is a real, measured, causally-tested win — and it is nothing like a full recovery. The decay is not a mistake you can undo by rotating your ads. It is a tax on repetition, and the only way to pay less of it is to need less repetition.
The stats that don't check out
Researching this meant chasing a lot of numbers back to nothing. These are the ones repeated most often, and what actually happened when I looked for the source.
| The claim | What is actually there |
|---|---|
| "Human attention span is 8 seconds, shorter than a goldfish" | No study. The chain runs through a defunct statistics aggregator to a Microsoft Canada report that did not make the claim. Attention research finds sustained attention measured in tens of minutes. |
| "85% of Facebook video is watched without sound" | Publisher self-reports from 2016. Meta never published a sound-on split. Snap has stated that most of its content plays with sound on. |
| "User-generated content drives 4x higher click-through" | Every site citing it gives a different figure and none names a study. |
| "56% of Meta auction outcomes are attributed to creative" | Quoted everywhere, attributed to a Meta summit. No Meta-published source exists, in the live web or in archives. |
| "Creative diversification delivers +32% efficiency" | Traces to a Meta deck that survives only as a third-party re-upload. No Meta-hosted original exists anywhere, including archives. The related figures Meta does publish on its own site are +11% click-through and +7.6% conversion rate. |
| "79% say user-generated content highly impacts purchase decisions" | Real, but from 2019 fieldwork, and the publisher's site died after acquisition. The year mutates between citations. |
| "B2B buyers spend only 17% of the journey with suppliers" | Real and traceable — to a survey run in 2017. Almost always quoted as though it were current. |
None of these is a scandal. They are what happens when a number gets useful before it gets checked, and the sourcing degrades one citation at a time until the original is unreachable. The reason to list them is not to score points. It is that if you are making decisions about creative, you should know which of your beliefs rest on measurement and which rest on repetition.
There is a nice irony in the fatigue literature here. The essay practitioners cite for why creative decays — Andrew Chen's "The Law of Shitty Clickthroughs", which argues that every marketing tactic ends in a terrible click-through rate as audiences adapt — carries no publication date on the page, and opens with a banner-ad click-through figure that is itself disputed. The argument turned out to be right. Meta measured it fifteen-odd years later. The numbers in it should still be handled as the author's own.
The rest of this series
Seven parts, each one on a piece of the problem, each with its sources attached.
- Why creative is the new targeting — the effectiveness evidence, the signal loss that caused the shift, and what automated buying did to the media plan.
- Ad hooks and attention — what the attention research actually establishes, how much of it survives contact with a feed, and what changes in a hook between platforms.
- Ad creative formats — static against video, aspect ratio and length, and a catalogue of 43 static ad archetypes grouped by the job each one does.
- User-generated content and creator ads — the performance data, the branding trade-off nobody wants to make, and why the budget line is in the wrong place.
- B2B ad creative — the largest measured quality gap in advertising, and the cheapest arbitrage in it.
- AI ad creative — adoption against measured lift, what consumers say, and where the production pipeline actually breaks.
- Ad creative testing — volume, win rates, fatigue detection, and how to call a winner without fooling yourself.
How this was researched
Every figure on this page traces to a named source with a year and a method. That is the standard for the whole series, and it is worth stating what it meant in practice.
Sources were graded in three tiers. Tier one means the page was opened, the figure read first-hand and matched. Tier two means the primary source could not be reached from here but two independent paths agreed on the figure — those publish with the organisation, year and method named in the caption, never bare. Tier three means a single path through a weak carrier, such as a document re-upload or a content farm. Nothing in tier three appears on this page or anywhere in this series. Two rows in the working ledger sit there, and both were cut.
Where a source was blocked, the Internet Archive was used rather than a summary of it — that is how the Meta fatigue analysis above was recovered after its live page started returning an error. Where a number could not be verified at all, it went into the list above instead of into the argument.
Where the evidence contradicts itself, both sides are published. There are at least three genuine contradictions in this material, and resolving them by picking a favourite would be the least useful thing to do with them.
FAQ
How much of advertising performance does creative actually explain? NCSolutions' model of roughly 450 campaigns attributes 49% of incremental sales to creative, against 11% for targeting. A separate econometric analysis of about 28,000 campaigns produces the same ranking, which is why this series treats the finding as coming from two independent directions rather than one study.
Does an ad wear in before it wears out? Meta measured its own creative fatigue and found no wear-in period at all — performance only ever gets worse with repetition. That contradicts the model taught in most marketing courses.
Why do attention and repetition have to be read together? Because they point opposite ways. Longer exposure raises memory; repeated exposure lowers response. Both curves are published, and the tension between them is the whole practical problem of creative.
Are the widely quoted creative statistics reliable? Several of them do not survive being looked up. The ones that fail are listed near the bottom of this page, with what happened to each.
