The short answer
"Creative similarity score" is an industry name for how alike your ads look to Meta's delivery system. Meta does not publish such a score. On our board, ads that copy a sibling's opener, headline, format and angle reach the top tier 4.7% of the time. Ads that reuse an opener but change at least one of those reach it 17.3% of the time.
The short answer
"Creative similarity score" is not a number Meta shows you. It is a name the industry gave to a behavior advertisers started noticing after Meta moved ad retrieval to Andromeda: ads that look and sound alike seem to be treated as one option and share delivery, instead of each reaching new people.
Meta has not documented a similarity score, a threshold, or how grouping works. What it has said is that its ranking models use a "creative representation" of each ad as an input (GEM engineering post) and that Andromeda helps pick the right creative as advertisers upload more of them for diversification (Meta business update). The system can tell what your ads contain. It has little reason to treat two copies of one idea as two ideas.
So we skipped the theory and measured it on the ads that are winning.
What near-duplicates look like among 2,443 winners
Our October 9, 2026 snapshot has 2,443 live Meta ads with $50k+ in estimated spend. For every brand with more than one ad, we looked for ads whose first six words match another live ad from the same brand. Then we split those into two groups:
- Full copies: same opener, same headline, same format (video or image) and same narrative angle as a sibling.
- Reworked: same opener, but at least one of headline, format or angle is different.
| Group | Ads | Brands | Share reaching top tier | Median estimated spend |
|---|---|---|---|---|
| Full copies | 215 | 62 | 4.7% | $193K |
| Reworked (opener reused, something else changed) | 243 | 84 | 17.3% | $258K |
| All other ads from multi-ad brands | 1,589 | n/a | 9.8% | $225K |
Top tier is the top 10% of the board by winner score (live 21+ days), so about 10% is the baseline. Full copies sit at less than half of it. Reworked ads sit well above it.
A caveat on the reworked number. Some of those ads are winners that a brand chose to relaunch with a new headline or format, and a winner relaunched is more likely to stay a winner. That inflates the 17.3%. It doesn't explain the 4.7% on the other side, though. Copying an ad unchanged is the one move that clearly underperforms.
Headline reuse tells the same story
Headlines are the cheapest element to copy, so we checked them too. Among brands with more than one ad:
1,046
ads share their exact headline with a sibling ad
8.1%
of those reach the top tier
12.3%
of ads with a headline of their own do
One brand-wide headline is a reasonable choice for an offer. But when every ad carries it, the ads look more alike to the person scrolling, and very likely to the system as well.
What "different enough" looks like in practice
Earth Breeze runs three live ads that open with the same 21-word line. They are not copies. One is a product demo with an offer headline, one is a UGC selfie for unaware readers with a trust headline, one is a product-aware demo. The UGC version is the one in the top tier.
It took me 30 years of doing laundry to learn this, but I'll teach it to you in about one minute.
- Est. spend
- $290K
- Days live
- 110
- Format
- Video, 173s
Same opener as two sibling ads, but a different person on camera, a different headline and a different reader. This is the version in the top tier.
Your Health Journal does the same with a podiatrist script. Two cuts open with "four shoes I would never recommend." The top-tier version reframes it as a contrarian expert lesson with its own headline, "4 Shoes That Worsen Your Balance."
Here are four shoes I would never recommend to my patients struggling with balance problems.
- Est. spend
- $1.2M
- Days live
- 109
- Format
- Video, 213s
- Variants
- 3
The sibling ad keeps a generic product headline and a listicle voiceover. This one changes the frame and the headline, and it is the one in the top tier.
Cherish These shows a bigger jump. Same opening line, but one version runs as a first-person story with a root-cause angle and the other is a creator skit built on a wrong-question angle. Both are past $1.4M in estimated spend.
Yeah, I can cover Saturday too.
- Est. spend
- $1.5M
- Days live
- 111
- Format
- Video, 65s
- Variants
- 2
Its sibling uses the same first line inside a skit with a different angle. Same hook, two concepts, both funded.
The counterexample is Stepprs: 29 live winners, all video, one tagged angle, one awareness level, one hook pattern, and the same headline on 28 of them. Several scripts open with "This is Steppers, the viral insoles." The ads have run for 197 to 487 days, so they clearly sell. None of the 29 is in the top tier.
Which changes count
Here is how we would rank changes by how different they make an ad, based on the tags that separate top-tier ads on our board and on what a person scrolling would notice.
| Change | Likely seen as a new concept? | Use it for |
|---|---|---|
| New narrative angle (root cause to cautionary tale) | Yes | New reach |
| New awareness level (unaware story to product-aware offer) | Yes | New reach |
| New format or production style (founder video to static, UGC to demo) | Yes | New reach |
| New person on camera | Often | New reach, cheaper than a new script |
| New opener on the same body | Partly | Lifting a winner |
| New headline or primary text only | Weak | Small lift, offer tests |
| Color, crop, background, music swap | No | Variants of a winner |
The difference between hook pattern and hook mechanism matters here. Two openers can use different words but the same mechanism (a problem stated, then a fix). To a model reading meaning, that may be the same ad.
A similarity check before you launch
- List every live ad in the ad set with four fields: opener (first sentence), headline, format, angle.
- Flag any pair that matches on all four. Those are full copies. Keep the better performer and turn the other into a real rework or pause it.
- For each new ad, change at least two of the four. One change is a variant. Two is closer to a new concept.
- Check awareness spread. If every ad assumes the viewer already knows the problem, add one that doesn't. Our creative diversification playbook has a grid for this.
- Look at how competitors rework winners. Searching a brand's live ads for a repeated opening line shows you which elements it changes and which it keeps. On Ad Radar, search covers video transcripts, so a query for a hook returns every ad that uses it, across brands. More on grouping in Entity IDs on Meta.
Reusing an idea that works is good practice. Reusing it unchanged is what the data says to avoid.
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.