The Meta creative fatigue benchmark

Why we ran this

Ask when to refresh a Meta creative and you will be told a number. Frequency 2.5 is the early warning. Frequency 3.5 means refresh now. Every agency deck repeats some version of it, and almost none of them show the data underneath.

We had the data sitting in our reporting warehouse, so we looked. 372 ads across 15 DTC brands, 10,508 ad-days of daily delivery, 48,484,188 impressions and $686,316 in spend, every creative tracked from the day it first delivered.

The result reframed the question for us. The fatigue debate assumes a creative lives long enough to wear out. For the overwhelming majority, that assumption is simply false, and the practical consequence is that most of the effort operators put into refresh timing is aimed at the wrong problem.

Finding 1: most creatives die before they can fatigue

Start with the whole population, every one of the 372 ads, no filters, no minimum spend. This is the part of the analysis with no survivorship problem at all, because nothing has been excluded.

Four out of five creatives never reach 100,000 impressions. Put that next to the frequency thresholds everyone quotes: to hit frequency 2.5 against even a modest 200,000-person audience, a creative needs half a million impressions. 94% of these creatives never got there. The fatigue conversation is about a stage of life most creatives never reach.

The lifespan distribution says the same thing from the other direction.

The median creative runs 18 days and delivers 10,127 impressions. A quarter of them are done inside 5 days having delivered 338 impressions, which is a rounding error. Only the top decile gets past 285,517.

Finding 2: a tiny minority carries the spend

Because most creatives die young, budget concentrates hard into the few that survive.

The top 10% of creatives absorb 69% of spend. The top 20% absorb 85%. The bottom half accounts for about 1%.

This is the economic shape of the whole exercise, and it has a direct operational reading. The value of a creative programme is almost entirely in how quickly it finds the few that scale. A process optimised for squeezing more life out of an average creative is optimising a rounding error. A process optimised for testing volume and cutting losers fast is working on the part that matters.

Finding 3: what fatigue looks like for the creatives that survive

Now restrict to the creatives that actually reached scale: the 57 ads across 13 brands with at least 150,000 lifetime impressions. Each creative is measured against its own performance over its first 100,000 impressions, and every chart below shows the percentage change from that starting point.

So 0% means "no different from how this creative began", -5% means it got 5% worse, and +15% means it got 15% worse if the metric is a cost. Because everything is relative to each creative's own starting point, a brand running at a $200 CPA sits on exactly the same scale as one running at $12, and neither can dominate the average.

Each line is the percentage change against that same creative's own performance over its first 100,000 impressions, so 0% means "no different from how this creative started" and every brand can be pooled regardless of its absolute costs. Buckets beyond 1M-2M are suppressed: they fall below the 15-creative and 5-brand minimum this study publishes at.

Click-through falls early, then stops falling

CTR drops 6.3% between the first 100,000 impressions and the 100-250k bucket. Then it essentially stops: -6.2% in the 250-500k bucket and -4.7% in the 500k-1M bucket, both measured against where each creative started. That is flat within the noise of a sample this size.

That shape matters. The mental model most operators carry is a steady slide, which implies that the longer you run a creative the worse it gets, forever. What this shows is a step down early, then a plateau. The creative takes an initial hit as it moves past the most responsive slice of the audience, and then settles.

Cost per acquisition is the number that actually moves

CPA rises 15.2% by the 250-500k bucket and stays there. It is a larger, more durable move than CTR, and it is the one that hits the P&L directly.

The practical implication is that CTR is a poor refresh trigger. It moves early, moves a little, and then goes quiet while CPA is still elevated. Watching click-through for a fatigue signal means reacting to the smallest, earliest, least economically meaningful part of the pattern.

CPM does not creep up

This one contradicts the received wisdom directly. The commonly repeated sequence is that frequency climbs, click-through slides, CPM creeps up, and conversion rate falls, in that order. In this dataset the CPM step never happens. It stays within 4.3 percentage points of where the creative began, across the entire published range.

Return on ad spend, and where we stop believing it

ROAS declines 11.9% by the 250-500k bucket, and then appears to recover sharply. We do not believe the recovery and neither should you. A creative only reaches the highest impression buckets if somebody kept funding it, and people keep funding creatives that are working. The late-life climb is the shape of survivorship, not the shape of performance. It is shown because hiding it would be worse, but no claim rests on it.

Finding 4: you cannot call it in the first three days

The most common creative-testing heuristic is to judge early click-through and cut the laggards. We tested it directly. Taking every creative with at least 3,000 impressions in its first three days (87 of them), we split by early CTR and looked at what each group went on to do.

Every figure in this table is per individual creative, not per group. The bottom third's $2,829 is what the median single creative in that group spent across its entire life, not the group's combined total.

Early click-through separates cleanly, from 1.08% to 3.83%, a spread of more than three times. And it predicts almost nothing. The bottom third went on to deliver a median 344,637 lifetime impressions, more than the top third's 124,909. The relationship is not weak, it is not even in the expected direction.

Is 29 creatives per group a lot? No. It is a small sample, and it is why this finding is stated as a negative rather than a relationship. Showing that a heuristic fails needs far less statistical power than quantifying one that works: the groups separate cleanly on the input (a threefold spread in early CTR) and then do not separate at all on the outcome, and they do not even order in the expected direction. That pattern is hard to produce from a real predictive relationship. What 29 per group cannot do is tell you the size of any effect, which is why no number is attached to the claim.

We are not claiming low early CTR causes success. With 29 creatives per group the sensible reading is that early click-through carries no usable signal about eventual scale. There are plausible mechanisms, most obviously that a high-CTR creative can be cheap attention that converts badly, while a lower-CTR creative that converts well earns budget. But the honest headline is the negative one: the heuristic does not work, so a three-day CTR read should not be deciding which creatives get killed.

What is actually happening when a creative "fatigues"

The word fatigue implies one process. The data suggests at least three, running on different clocks, and separating them explains why the curves look the way they do.

Audience exhaustion. The most responsive slice of an audience sees the creative first and acts first. Once they are used up, the same creative is being shown to progressively less responsive people. This is the mechanism behind the early CTR step: it is not that the creative got worse, it is that the remaining audience is different. It explains why the drop is front-loaded and then plateaus, because after the initial skim the marginal audience stops changing much.

Creative wear-out. The same person sees the same ad repeatedly and tunes it out. This is the mechanism the frequency thresholds are actually about, and it is the one we could not measure directly without reach data. If it dominated, we would expect continuous decay rather than a plateau, and we do not see continuous decay.

Auction and market conditions. Competitor bidding, seasonality and audience supply move CPM independently of the creative. This is almost certainly why CPM looks flat against cumulative impressions: it is being driven by something that has nothing to do with how long your creative has been running, so it does not correlate with exposure.

Read that way, the flat CPM result stops being surprising. CPM was never a creative metric. The industry folded it into the fatigue narrative because CPM and CTR often deteriorate in the same week, and correlation in time got mistaken for a shared cause.

Where the 2.5 and 3.5 numbers came from

Worth being fair to the received wisdom. The frequency thresholds are not invented. They come from an era of much smaller audiences, narrower interest targeting and manual placement, where an advertiser could genuinely exhaust a saturated audience and watch frequency climb week over week. In that world, frequency was a reasonable proxy for how thoroughly you had worked an audience.

Two things changed. Broad targeting and automated placement made the addressable pool for most DTC advertisers far larger, so frequency rises much more slowly relative to spend. And the optimisation layer now reallocates delivery away from saturated pockets automatically, which partially absorbs the effect the threshold was designed to catch.

The threshold survived because it is easy to remember, easy to check and produces a decision. Our argument is not that it is stupid, it is that it is a proxy for a proxy, and the underlying quantity it stands in for is better measured directly.

What we would actually do with this

Five things follow, in order of how much money they are worth.

Stop optimising refresh cadence and start optimising hit rate. With 80% of creatives never reaching 100,000 impressions and the top decile carrying 69% of spend, the return on finding one more scalable creative dwarfs the return on extending the life of an average one.

Trigger refreshes on CPA, not CTR and not frequency. CPA is where the durable 15.2% move shows up. CTR moves first but barely, and then goes quiet.

Stop reading CPM as a fatigue signal. It did not move with exposure here. When CPM rises, look at auction pressure, audience size and seasonality.

Give creatives more than three days before judging them. The early-CTR read that most testing frameworks depend on had no predictive value in this data.

Expect the plateau. After the initial step down, click-through was broadly stable through the published range. A creative that took an early hit and settled is not necessarily dying, and killing it on the basis of that first drop may be destroying one of the few that would have scaled.

How we would structure creative testing given this

If the top decile carries 69% of spend and early signals do not identify it, the process has to be built around that rather than around picking winners cleverly.

Volume over precision at the top of the funnel. When your hit rate on scalable creatives is roughly one in ten, the number of genuinely distinct concepts entering testing is the main lever you control. Distinct means different angle, different hook, different proof, not five crops of the same asset.

Judge on a spend threshold, not a time window. "Three days" means completely different exposure for a creative at $50 a day and one at $2,000. Give every creative the same opportunity measured in delivery, then judge it.

Let the plateau run. Since click-through steps down early then flattens, a creative that took its initial hit and stabilised at an acceptable CPA is not dying. The instinct to refresh on that first drop is the instinct to kill creatives at exactly the point they have finished their most expensive phase.

Refresh on economics. Set the trigger on CPA against target, which is where the durable 15.2% move showed up, rather than on a frequency threshold or a CTR wobble.

Keep the winners running longer than feels comfortable. With this much spend concentrated in so few creatives, prematurely retiring a proven one is expensive, and the data does not show a cliff that justifies it inside the published range.

Frequently asked questions

What is creative fatigue in Meta ads?

Creative fatigue is the decline in performance as a creative accumulates delivery. In this study of 372 creatives, it shows up as a 6.3% drop in click-through within the first 250,000 impressions and a 15.2% rise in cost per acquisition by the 250-500k mark. It does not show up as rising CPM. The more important finding is that 80% of creatives never reach 100,000 impressions, so most never fatigue at all.

How often should I refresh Meta ad creative?

Refresh on cost per acquisition against target rather than on a fixed schedule or a frequency threshold. CPA is where the durable move appears in this data. A creative that took an early click-through drop and then stabilised at an acceptable CPA is performing normally, not fatiguing.

Is frequency 2.5 or 3.5 the right threshold to refresh at?

We could not measure frequency directly, because it requires reach data this dataset does not carry. What we can say is that 94% of creatives never reach 500,000 impressions, which is roughly what it takes to hit frequency 2.5 against even a modest audience. For most creatives the threshold is never reached, so it cannot be the operative decision rule.

Does CPM go up as a creative gets older?

Not in this data. CPM stayed within 4.3 index points of where each creative started across the entire published range. Rising CPM is better explained by auction competition, audience supply and seasonality than by creative age.

Can I tell in the first few days whether a creative will win?

Not from click-through. Splitting 87 creatives by their first three days of CTR, the top third went on to deliver a median 124,909 lifetime impressions against 344,637 for the bottom third. Early click-through carried no usable signal about eventual scale.

What percentage of Meta creatives actually scale?

About one in ten carries meaningful budget. The top 10% of creatives absorbed 69% of spend and the top 20% absorbed 85%, across 372 creatives and $686,316 of spend.

How long does a Meta creative last?

The median creative in this study was live 18 days and delivered 10,127 impressions in total. The 90th percentile reached 75 days and 285,517 impressions.

Do video and static creatives fatigue differently?

We could not answer this to our own standard. The video cohort peaked at 14 qualifying creatives per bucket against the 15-creative minimum we publish at, so the comparison was cut rather than published thin.

What we would measure next

Three open questions this dataset could not close.

Reach, so frequency can be tested directly. The single most valuable addition would be reach at ad level, which would let us test the 2.5 and 3.5 thresholds against outcomes instead of arguing about them.

A longer window. 2026-04-08 to 2026-08-10 cannot separate seasonality from exposure. A multi-year pull would.

Creative attributes. Whether hook style, angle or proof type predicts scale is the question operators most want answered, and it needs creative-level tagging across a far larger sample than we currently hold.

Method

Daily ad-level delivery for every creative across 15 managed DTC accounts, pulled from the Meta Marketing API into our reporting warehouse. The unit of analysis is the individual creative, not the brand or the campaign, so one large advertiser cannot dominate a curve.

Fatigue is measured against cumulative delivered impressions rather than days live, because two creatives of the same age have had completely different exposure if one ran at $50 a day and the other at $5,000. Each creative is indexed to its own performance in its first 100,000 impressions, which lets brands with very different economics be pooled and means no figure here reveals any individual advertiser's costs or returns.

Rates are computed on pooled sums inside each bucket, never as an average of daily ratios, which would let a single low-volume day swing a result. Whole-population findings on mortality and spend concentration use every creative with no minimum. Decay-curve findings use creatives with at least 150,000 lifetime impressions.

No individual advertiser is identified anywhere in this report. Every published figure aggregates at least 15 creatives across at least 5 distinct brands; any cell below either threshold is suppressed and the suppression is disclosed rather than hidden. All performance figures are indices against a creative's own baseline, never absolute costs.