The short answer
Meta allows up to 50 ads per ad set, but the number that matters is distinct concepts: 3 to 6 is a workable range for most budgets. Winners rarely carry many versions: 71% of the 2,443 ads on our board run as a single version, and the median brand has just 2 ad versions live among its winners.
The short answer
The hard limit is 50 ads per ad set. That applies to Advantage+ sales campaigns too, which brought ad sets back after the original Advantage+ shopping format's single structure with up to 150 ads (summary of the change).
The useful answer is different. Count concepts, not ads. For most budgets, 3 to 6 distinct concepts per ad set gives Meta's delivery system real choices without starving each ad of spend. Variants of a proven winner can sit next to them, but they don't count as extra concepts.
Practitioner guidance lands in a similar place. One agency recommends a minimum of six meaningfully different ads per ad set. Treat that as directional: Meta has not published a recommended number.
Ad count vs concept count
Since Andromeda, near-identical ads appear to be grouped and treated as one option (the industry calls this an Entity ID; it is not an official Meta term). That changes the question.
| Ad set | Ads | Distinct concepts | What delivery has to work with |
|---|---|---|---|
| 12 cuts of one founder video | 12 | 1 | One idea, split 12 ways |
| 4 hooks on one body + 2 statics of the same offer | 6 | 2 | Two ideas |
| Founder story, product demo, UGC objection video, offer static | 4 | 4 | Four ideas for four kinds of buyer |
The third ad set has the fewest ads and the most options.
What variant counts on winners tell us
In the Meta Ad Library, one ad can carry several versions (different images, cuts or text combinations). We record that as variants. Here is how the 2,443 winners in our October 9, 2026 snapshot break down:
| Versions per ad | Ads | Share | Median estimated spend | Share reaching top tier |
|---|---|---|---|---|
| 1 | 1,743 | 71% | $177K | 4.5% |
| 2 to 3 | 524 | 21% | $332K | 13.0% |
| 4 to 9 | 148 | 6% | $528K | 43.9% |
| 10 or more | 28 | 1% | $1.12M | 82.1% |
Read the last column with care. Variants launched is one of the signals in Ad Radar's winner score (alongside days live and engagement growth), so the top-tier share rises with variants partly by construction. The spend column is independent of that and shows the same pattern: ads that get more versions carry far more estimated spend.
The order of events matters. Brands don't launch ten versions and hope one wins. They launch one, see it work, then add versions around it. Variants are what a winner gets, not how you find one.
Two more numbers:
71%
of winning ads run as a single version
43%
of image winners have 2+ versions, vs 25% of video
2
median ad versions live per brand, across all its winners
14
at the 90th percentile of brands
Images get more versions than video because they are cheap to vary: a new background, a new headline, a new crop. And most brands keep fewer live versions of their winners than people assume. The median brand on the board has two.
Video and image ads scale differently
The split by format is worth a closer look, because it changes how you count an ad set.
| Format | 1 version | 2 to 3 versions | 4 or more versions |
|---|---|---|---|
| Video: ads | 1,471 | 426 | 71 |
| Video: median estimated spend | $187K | $356K | $1.09M |
| Image: ads | 272 | 98 | 105 |
| Image: median estimated spend | $131K | $209K | $291K |
22% of image winners carry four or more versions. Only 3.6% of video winners do. A static concept tends to show up as a family of versions (new crops, new headlines, new backgrounds), while a video concept usually runs as one or two cuts and gets very large when it does get more.
For your ad set count, that means a static with five crops is still one concept, and it is cheap to keep that family fresh. A video concept is expensive to vary, so each new cut should earn its slot by changing something that matters, like the opener or the person on screen.
Three ads at both ends
Online Certification for Grief Coaches
- Est. spend
- $4.7M
- Days live
- 84
- Format
- Image
- Variants
- 15
Fifteen versions of one image ad in 84 days. This is what scaling a static winner looks like: one idea, many executions.
It's time to retrieve all those files you thought you lost.
- Est. spend
- $2.5M
- Days live
- 1442
- Format
- Video, 57s
One version, almost four years live, in the top tier. A single strong concept needs no crowd around it.
Hey, before you skip this video, let me tell you why my product sucks.
- Est. spend
- $2.5M
- Days live
- 2260
- Format
- Video, 147s
Over six years live as a single version. A founder video that admits a flaw is hard to tire of and hard to imitate.
How to set your number
Step 1: Work out how many conversions you can afford
Meta's help center says an ad set generally needs about 50 optimization events since its last significant edit to stabilize. Divide weekly budget by your target cost per purchase:
- $300 a day at a $50 CPA = about 42 purchases a week
- $1,000 a day at a $50 CPA = about 140 a week
At 42 a week, the ad set is near the edge of exiting learning, and each extra concept splits that further. Three or four concepts is plenty. At 140 a week you can run six to eight concepts and still give each one a fair read.
Step 2: Fill the slots with different ideas
Pick concepts that differ in angle, reader awareness or format. Our creative diversification playbook has a map of where winners cluster.
Step 3: Add variants only to winners
When an ad wins, give it two or three new hooks or cuts in the same ad set. That is where variants belong. See ad variants as a scaling signal.
Step 4: Prune monthly
Pause ads with no meaningful spend after two to three weeks unless they are new. They are not hurting much, but they clutter your read of what works. More on cadence in how many creatives to test.
When you study a competitor, the variant count on each ad tells you which ones it is scaling. Ad Radar's winner score already counts variants launched, so following a brand and filtering for top-tier ads shows you the concepts it is putting money behind.
Figures marked as estimated spend come from Ad Radar's model of engagement on public Meta Ad Library ads. They are estimates, labeled as such, and are best used to rank ads against each other.