Creative testing is where the effectiveness research either becomes an operating routine or stays a slide. Part seven, and the last, of the ad creative series.
Winners are rare, and that is the whole design constraint
Motion analysed 578,750 creatives across 6,015 accounts representing $1.29 billion of Meta spend. A winner is defined there as a creative reaching at least ten times the account's median spend and at least $500 — a definition built around whether the delivery system chose to back it.
About 5% of creatives become winners. Roughly half are switched off within 28 days, and the few winners take anywhere from about a quarter of all spend for the smallest advertisers to nearly two-thirds for the largest.
If one in twenty works, then the number of creatives you put into the market is not a stylistic preference. It is the mechanism.
Run the arithmetic on the bottom row. Under three creatives a week at a 3.8% win rate is roughly one winner every nine weeks. Motion's own framing puts four a week at about 0.2 winners per week and eighteen a week at about 0.9.
Note what else that chart says: the win rate more than doubles between the smallest and largest advertisers. Volume is not just buying more lottery tickets — advertisers who ship more also get better at it, because they see more outcomes.
The practical instruction for a small budget is not "spend like an enterprise". It is that a set of three to five genuinely different statics, as described in the formats guide, is the minimum unit that can teach you anything — and one ad running alone teaches you nothing at all, regardless of how it performs.
How to call a winner without fooling yourself
The rules do not change with account size. Only the volume does.
What I use is an impression allowance per ad rather than a calendar. Each creative gets somewhere between 500 and 1,500 impressions before a read is taken, and that threshold is what decides when the decision happens — not the reporting rhythm, not Monday morning. It works the same way on a $2,000 account, a $10,000 account and a $200,000 account, because impressions accumulate at whatever rate the budget allows and the decision waits for them.
Be precise about what that gate can decide. At 1,000 impressions and a 1% click-through rate you have about ten clicks. That is enough to screen a hook — to see whether the creative stops the scroll and earns a click at all — and it is nowhere near enough to call a conversion winner. Ten clicks will not distinguish a 3% landing-page conversion rate from a 6% one.
So it is two gates rather than one long test, and the mistake is running the first and announcing the second one's conclusion. What makes them different is not just the volume. They are not measuring the same object.
Gate one judges the creative. Five hundred to 1,500 impressions, read on click-through rate and thumbstop. Does this ad stop the scroll and earn a click. That is the entire question at this stage, and the obvious failures die here.
Gate two judges the conversion path. Around 2,500 impressions, which at a 5% click-through rate is about 125 clicks. What that volume buys is enough traffic to see whether the path behind the ad converts at all — I want to see at least one or two conversions out of it. If a hundred people have clicked and nothing has happened, the problem is very rarely the ad.
Notice what has changed between them. By gate two the creative has already passed; it did its job and delivered the click. What gate two actually tests is the landing page, the form, the offer and the follow-up. It is a funnel test wearing a creative test's clothes, and reading it as a verdict on the ad is how good creative gets killed for a broken form's mistake.
Be equally honest about what one or two conversions can support. That is a viability read — does this path convert at all — and it is not a ranking. Two conversions cannot tell you that ad A beats ad B. Treating them as though they can is the same error that produced most of the numbers in the section further down this page.
One number in that arithmetic does all the work, and it is the click-through rate. Five per cent is a LinkedIn figure — roughly the bar I hold creative to on those placements. Meta creative typically runs at 1 to 2%, where the same hundred clicks costs several times the impressions to buy.
| Click-through rate | Impressions to reach ~100 clicks | Where you see it |
|---|---|---|
| 5% | 2,000 | LinkedIn, creative clearing the bar |
| 2% | 5,000 | Meta, strong creative |
| 1% | 10,000 | Meta, average creative |
Same gate, same decision, three very different media bills. My own threshold sits at about 2,500 impressions because I am usually looking at placements where 5% is realistic, and that clears a hundred clicks with a little room. On a 1% account the identical decision needs five times the impressions, and the account that budgets for 2,500 and then calls the conversion path dead at twenty-five clicks has learned nothing except that it did not buy enough traffic.
The rule is written in impressions because impressions are what the platform accumulates while you wait. The quantity actually doing the work is clicks.
This is the operational form of a rule I keep coming back to: sample size sets the window, not the calendar. A weekly review cadence invites confident decisions on noise in any account that does not have the volume to fill a week.
Fatigue, and what the platform will do about it before you do
Fatigue is easy to spot and the first signal is the obvious one: click-through rate falls. The second signal is less obvious and more expensive — delivery systems are now good enough to simply stop serving a tired creative, and Meta will sometimes pause a campaign outright. If the platform is telling you, you have already paid for the lesson.
The measured behaviour behind that is Meta's own fatigue analysis, drawn on the hub page, and it has three parts worth restating here because they change what you should do.
- Meta's mean impression carries 4.2 prior exposures. More than 19% of impressions go to someone who has already seen the creative more than five times.
- There is no wear-in period. Meta looked for one in its own data and did not find it: clicks and conversions get monotonically more expensive from the first impression onward. The hump-shaped curve most of us were taught is not what direct-response data shows. Meta notes brand objectives may behave differently.
- Refreshing creative is worth about 8% in high-fatigue cases, from a two-cell split test across roughly 26,000 cases — dose-dependent, so the more fatigued the creative, the larger the gain.
That last pair is the honest framing of creative rotation. Fatigue costs around 45% of conversion likelihood by the fourth exposure. Refreshing wins about 8% back. Rotation is a real, causally-tested improvement and it is not a cure. The only structural fix is needing less repetition, which means a bigger creative pool or a bigger audience, not a faster rotation schedule on the same three ads.
The thinking behind this predates the measurement. Andrew Chen's essay "The Law of Shitty Clickthroughs" argued years earlier that every marketing tactic ends in a terrible click-through rate as the audience learns to recognise it — true of channels, and equally true of creative execution. Meta's data is that argument with numbers attached. Worth reading; worth noting that the page carries no publication date and its own headline statistic is disputed, so take the argument rather than the figures.
The thing that is not a creative problem
Asked what the most expensive creative mistake I have seen is, my honest answer is that it usually is not a creative mistake.
It is a broken funnel, or a spelling error on the ad. Both cost trust, and trust is what the conversion was made of. A creative can only lose what the funnel was going to convert — so before you spend a month optimising hooks, check that the form works, the page loads, and nothing on the ad is misspelled. That sounds too obvious to write down. It keeps being the answer.
Nobody has published a modern frequency table
One last note for anyone trying to do this properly. The frequency-versus-performance table the industry cites is AdEspresso's, from 2018, built on about 500 campaigns.
I went looking for a replacement — anything from 2023 onward with a frequency band against click-through rate, cost per click or cost per acquisition, and a stated sample. Not from the platforms, not from the major measurement and analytics firms, not in the trade or academic press. It does not exist publicly.
That absence is itself worth knowing. When you see a confident frequency cap recommendation, it is resting either on 2018 or on somebody's private data.
The series
- Why creative is the new targeting
- Ad hooks and attention
- Ad creative formats
- User-generated content and creator ads
- B2B ad creative
- AI ad creative
FAQ
How many creatives should a small advertiser ship each week?
The honest answer is more than most do, but the instruction is not "spend like an enterprise". Micro advertisers in Motion's data launch 2.80 a week at a 3.8% win rate, which is roughly one winner every nine weeks; enterprise advertisers launch 18.85 at an 8.2% win rate. Note that the win rate more than doubles across those tiers — volume is not just buying more lottery tickets, because advertisers who ship more also see more outcomes and get better at it. For a small budget, a set of three to five genuinely different statics is the minimum unit that can teach you anything, and one ad running alone teaches you nothing at all regardless of how it performs.
How many impressions before I can call a winner?
Two gates, not one long test, and they are not measuring the same object. Gate one judges the creative: 500 to 1,500 impressions, read on click-through rate and thumbstop, answering only whether the ad stops the scroll and earns a click. Gate two judges the conversion path at around 2,500 impressions — about 125 clicks at a 5% click-through rate — and what it actually tests is the landing page, the form, the offer and the follow-up, because by then the creative has already passed. The mistake is running the first gate and announcing the second one's conclusion, which is how good creative gets killed for a broken form's mistake. Both thresholds are my own operating rules, not published benchmarks.
Does rotating creative fix fatigue?
It helps and it is not a cure. Fatigue costs around 45% of conversion likelihood by the fourth exposure, and refreshing creative wins about 8% back in high-fatigue cases, from a two-cell split test across roughly 26,000 cases — dose-dependent, so the more fatigued the creative, the larger the gain. There is also no wear-in period: Meta looked for one in its own data and found clicks and conversions getting monotonically more expensive from the first impression onward, though it notes brand objectives may behave differently. The only structural fix is needing less repetition, which means a bigger creative pool or a bigger audience, not a faster rotation schedule on the same three ads.
Sources on this page
- Motion, "Creative Benchmarks 2026". 578,750 creatives, 6,015 accounts, $1.29 billion of Meta spend. Winner defined as at least ten times account median spend and at least $500.
- Analytics at Meta, "Creative Fatigue: How Advertisers Can Improve Performance by Managing Repeated Exposures", 2023. Logistic regression on matched impression and click data; 30-day lookback; dynamic product ads and brand ads excluded. Read from the Internet Archive capture of 20 May 2023 because the live page returns an error.
- AdEspresso, Facebook ad frequency analysis, 2018. About 500 campaigns. Dated, and still the industry's citation because nothing has replaced it.
- Andrew Chen, "The Law of Shitty Clickthroughs". Undated on the page; earliest Internet Archive capture 2014. The figures in it are the author's own and one is disputed — cite the argument, not the numbers.
- The 500-to-1,500-impression creative gate and the ~2,500-impression conversion-path gate are both my own operating rules across the accounts I run, not published benchmarks. The click-through rates in the table are the ranges I see on those platforms, used here to show how the same hundred-click threshold prices differently — they are not a benchmark either.
