You should test the largest ad set your budget can resolve, not a fixed number copied from a growth thread. For most Meta and TikTok accounts, that lands between 8 and 20 live variations at once. The exact number comes from dividing your test budget by the cost of one reliable read, then rounding down.
Most teams ask a production question first: how many variants can we make? The harder question is media capacity: how many variants can the campaign actually read before fatigue or budget runs out. Our 2026 fatigue benchmark finds that most creative is effectively dead within three weeks. CTR declines 15 to 20 percent in a creative's first two weeks, then the drop accelerates. Week three is a cliff at 45 to 70 percent below launch baseline. By week five a creative averages 38 percent below its peak.
Why 8 to 20 live variations is the range worth planning around
The benchmark is a synthesis of published platform and agency figures, not our own experiment. It puts 8 to 20 live variations as what a typical active campaign needs in rotation. Brands shipping 15 to 50 creative variants a month see 3 to 5 times longer campaign lifespan than quarterly refreshers. The operational conclusion is that throughput, not talent, is the bottleneck.
Meta internal research in the benchmark reports a CTR drop of 45 percent after a fourth exposure to the same creative. Weekly frequency of 2.5 on Meta prospecting is where decline begins. On TikTok, a typical in-feed hook rate is 20 to 30 percent, and fatigue tends to arrive above roughly 3 frequency.
That does not mean every account should test 20. It means if you are below 8, you will feel fatigue before you have enough data to replace tired creative. If you are above 20, you are usually testing more than a typical active campaign can rotate without a serious media budget.
How to calculate the largest resolvable set
A resolvable set is the number of ads that can each get enough impressions for a reliable read. The formula is:
Largest resolvable set = floor(test budget / (read impressions × CPM / 1000))
The read is the impression count you need before a CTR or hook rate difference becomes usable. On Meta feed or Reels I typically use 5,000 impressions per variation as a planning read. Use 3,000 when you only need a rough direction and can accept noise. Use 8,000 when you expect small CTR differences or need a buying decision with more confidence.
At those read sizes you are dealing with small but useful click counts. The benchmark gives Meta feed video CTR at 1.62 percent and Reels at 1.78 percent, so 5,000 impressions is roughly 81 to 89 clicks. TikTok is lower at 0.84 percent, or about 42 clicks per 5,000 impressions, which is why some teams use a larger read on TikTok.
At a EUR 12 CPM, 5,000 impressions cost:
5,000 × EUR 12 / 1000 = EUR 60
So a EUR 1,000 test budget buys 16 full reads:
floor(EUR 1,000 / EUR 60) = 16
That is the worked answer. At EUR 12 CPM and a 5,000 impression read, do not run 30 ads. Run 16 and actually learn something from each.
Why an unresolvable test costs the same as a resolvable one
The platform charges per impression, not per ad. If you spend EUR 1,000 at a EUR 12 CPM, you buy about 83,333 impressions. Split across 16 ads that is 5,208 impressions per ad, enough for a read. Split across 50 ads that is 1,667 impressions per ad, enough to see movement, not enough to separate signal from noise.
The media spend is the same in both cases. The difference is that the resolvable test produces 16 decisions. The unresolvable test produces 50 opinions and a spreadsheet nobody trusts. Cutting the set to fit the budget is not a loss. It is the same money spent on fewer, better reads.
Budget table at a EUR 12 CPM
| Read per variation | Cost per read at EUR 12 CPM | Budget for 8 ads | Budget for 16 ads | Budget for 20 ads |
|---|---|---|---|---|
| 3,000 impressions | EUR 36 | EUR 288 | EUR 576 | EUR 720 |
| 5,000 impressions | EUR 60 | EUR 480 | EUR 960 | EUR 1,200 |
| 8,000 impressions | EUR 96 | EUR 768 | EUR 1,536 | EUR 1,920 |
This table replaces the rule of thumb with arithmetic. If your weekly test budget is EUR 400 at a EUR 12 CPM, you cannot resolve 16 ads at a 5,000 impression read. You can resolve 8 at a 3,000 impression read and still keep some budget for follow-up tests. Most accounts should use the creative testing calculator before locking the set size.
When a smaller set is the right call
A EUR 300 budget at a EUR 12 CPM and a 5,000 impression read resolves five ads. Run five, read, cut the weak ones, and use the next budget window to test replacements. Forcing 10 into the same window does not make the test bigger. It makes the read worse.
The practical live range is still 8 to 20 for an active campaign with enough budget to rotate properly. Below 8, creative fatigue tends to arrive before replacements are ready. Above 20, most teams cannot manage the analysis or the rotation cleanly unless they have a dedicated media buyer.
Where Genyad fits into this
Production is usually not the binding constraint once the set is calculated. Genyad turns footage you already own into publishable ad variations. Each standard variation costs 1 credit on paid plans. Editing, re-exporting and uploading footage cost nothing. The hard part is not making 16 variants. It is giving 16 variants enough media budget to tell you something.
Set the set size first with the creative testing calculator, then use the ad variation calculator to check how many distinct variations your footage library can support before scripts start repeating. A variation is not a re-cut of the same timeline. The ad variation glossary entry covers what separates a real variation from a cosmetic one.
Genyad does not predict which ad will win. It has no predicted performance scores and no direct publishing to Meta or TikTok. If you need synthetic presenters, AI avatars, or product-feed-driven template rendering, other tools are a better fit. Arcads was EUR 100 per month for 10 AI actor videos as publicly listed in August 2026. Creatify was $39 per month as publicly listed in August 2026. Prices move. Genyad is the fit when you already have footage and need enough distinct variations to occupy a resolvable set.
Frequently asked questions
How do I calculate how many ads to test at a different CPM?
Use the same formula with your CPM. If your CPM is EUR 18 and your read is 5,000 impressions, cost per read is EUR 90. A EUR 1,000 budget resolves 11 ads. The creative testing calculator can do this without a spreadsheet.
Is 8 to 20 ads a hard rule?
No. It is an operating range from our 2026 fatigue benchmark, which is a synthesis of published platform and agency figures. A smaller budget may resolve five ads. A larger account can resolve more. The rule is that the budget must resolve the set.
What counts as a reliable read?
I use 5,000 impressions per variation as a planning read on Meta feed or Reels. The read should be set before launch, not after you see noisy results. If you need stronger confidence, use 8,000 impressions and accept a smaller set.
How much does Genyad cost to produce variations?
Genyad has a free plan with 5 video ad variations and one AI-generated video with no card. Starter is EUR 29 for 15 credits, Growth is EUR 99 for 65 credits, and Enterprise is custom. One standard variation costs 1 credit. AI-generated video costs 2 credits per second.
Does Genyad publish directly to Meta or TikTok?
No. Genyad does not have direct publishing to Meta or TikTok. You export the variations and load them into your ads manager. It also does not provide predicted performance scores or static banner formats.