Meta Creative Similarity Score: What Sameness Looks Like in 15 Live Ad Libraries

Every article about Meta's Creative Similarity Score quotes a threshold. Six ads above 60% similarity. A 15 to 25% CPM swing. A score that climbs when you add a lookalike creative. None of those numbers come from Meta, because Meta has never published them, and none of the people repeating them can show you the score on a real account.

We cannot show it either. Brandsearch does not see inside Meta's auction, and a page that claimed otherwise would be inventing the same numbers with a different logo on them.

What we can show is the thing the score reads: how many structurally identical ads a brand is actually running right now. Brandsearch scans a brand's live Meta library, groups every ad by its opening line, and prints the count. That is a measurement, it renders on screen, and it is the closest honest proxy anyone has for the input Andromeda is scoring.

So this page measures it. Fifteen live Meta ad libraries, read on 12 September 2026. Between them they run 1,677 distinct opening lines across 11,588 ads, and 2,936 of those ads, better than one in four, sit behind whichever single line their brand repeats most.

The 15 ad libraries

Ordered by how much of the library sits behind one opening line, most concentrated first.

How to read the numbers

An opening line is not a creative. Brandsearch groups ads by the first line of the ad body, which is the part a scroller reads before deciding. Two ads sharing an opening line can use different footage. They are still, to a model scoring copy architecture, the same opening move made twice.

Catalog ads are excluded, and there are a lot of them. Meta writes the text on a catalog ad per product, so counting them would measure Meta's templating rather than the brand's. The 15 libraries hold 7,713 catalog ads on top of the 11,588 counted here. Every screenshot names its own catalog figure.

The percentage is arithmetic on two numbers in the same screenshot, the ad count on the top row divided by the ads scanned in the header. Nothing is modelled.

A concentrated library is not automatically a failing one. This is the finding that survived the data, and it is the opposite of the advice on most pages about this topic. Several of the most concentrated libraries here are carrying a worldwide top-ranked ad on the very line they repeat most. Rank badges in Brandsearch are global, computed across everything a brand runs in every market, while the euro figures are only the slice EU and UK disclosure rules force advertisers to publish. So a small euro number next to a #1 badge is a disclosure artefact, not weak performance.

1. Meller: 186 ads behind one opening line

See this brand in Brandsearch

The Barcelona sunglasses brand runs the most concentrated library in the set. Twenty-four opening lines cover 269 of the ads Brandsearch scanned, and 186 of those, 69% of everything counted, open with the same two words: "🌱 Sustainable collections". The second line, "✅ Improve Sleep", accounts for another 41.

The third row is the interesting one. "This is a test ad 🌱 Sustainable collections" has eight ads behind it, which is a live brand shipping its QA copy into a real library and leaving it there. It is a useful reminder of what these views are: not a curated case study, but whatever the advertiser actually has running.

2. DOONAILS: 67 ads behind one opening line

See this brand in Brandsearch

The smallest library here and the second most concentrated. Fourteen opening lines across 99 scanned ads, with "Bye Bye Press Ons 👋" carrying 67 of them. A further 284 of the brand's ads are catalog ads, which is the pattern you see when a store leans on Meta's product feed for breadth and keeps its written creative deliberately narrow.

Six of the fourteen lines are German and the other eight are English, so most of the variety in this library is a language split rather than a change of argument. This is a store that has decided what it sells and stopped restating it.

The catalog behind the narrow library

The Products view explains the shape. DOONAILS lists 196 products, and the bestsellers carry their own rank badges: Preparation Essentials at #1 for €22.90, the Starter Set at #2 for €49.95, then a run of Soft Gel Press Ons at €19.95. Fourteen opening lines is not a thin catalog being stretched. It is a wide catalog behind one message, with the product feed doing the variety and the copy staying still.

3. Elavate: 629 ads behind one opening line

See this brand in Brandsearch

The UK collagen brand runs 115 distinct opening lines, which by the standards of this list is a broad library. It is still 54% concentrated, because one line, "With Elavate's Multi-Collagen Superblend, your skin, hair and joints will look and feel better than ever!", sits behind 629 ads on its own.

That combination is worth pausing on. Elavate has not failed to write variations. It has written 114 other ones. It has simply put the majority of its volume behind the version that works, and kept the rest in rotation underneath. A count of distinct lines alone would have called this library diverse, and a count of concentration alone would have called it repetitive. It is both, and the two numbers only mean something together.

Pull the opening lines behind any brand's live ads

4. Essence Vault: 180 ads behind one opening line

See this brand in Brandsearch

Thirty-seven opening lines across 382 scanned ads, plus 1,156 catalog ads sitting behind them. The top line, "Looking for long lasting perfumes but tired of breaking the bank account?", carries 180 ads, and the second, "Love luxury fragrances but hate the £300+ price tags?", carries 55.

Those two lines are the same argument written twice. Both name the category, both name the price objection, both resolve it. A model scoring copy architecture rather than vocabulary has very little to separate them, which is exactly the trap that the phrase "we tested a new hook" usually hides.

5. prepmymeal: 636 ads behind one opening line

See this brand in Brandsearch

The largest single concentration in the set. The German prepared-meals brand runs 33 opening lines across 1,502 scanned ads, and "Einfacher geht gesunde Ernährung nicht!" is the opening line on 636 of them. The second line, "Leckere High Protein Gerichte - fertig in nur 7 Minuten 💪", adds another 288.

Only 55 of its ads are catalog ads, which makes this a written library rather than a feed-driven one. Someone chose to say the same sentence 636 times.

What that repeated line is actually attached to

Open one of the ads carrying that line and the case for the repetition appears. Brandsearch scores it Top 1% of 325 active ads, it ran 517 days, it exists in ×10 versions, and it disclosed €128k of EU and UK spend. The rank is worldwide, computed across everything prepmymeal runs anywhere; the euro figure is only the disclosed European slice, so the true spend behind it is larger than the card can show.

This is the row that breaks the standard advice. A brand told to diversify would kill this line. The data says the line is the brand's single best-ranked asset, and the ×10 versions are not an accident but the mechanism: one proven message, deliberately duplicated at the delivery level.

Check the rank on any ad before you copy it

6. Nutrition Geeks: 267 ads behind one opening line

See this brand in Brandsearch

Twenty-six opening lines across 902 ads. The most-used one is not a sentence at all: "⭐️⭐️⭐️⭐️⭐️", five star emoji, on 267 ads. "Want youthful skin?" follows with 234 and "Feeling bloated or sluggish after meals?" with 160.

A library topped by a row of emoji is a real result rather than a parsing quirk, and it says something specific about the category. The supplement shopper is being sold a rating before a claim, and the brand has found that the fastest way to say "reviewed well" is to not use words.

7. ARNE: 153 ads behind one opening line

See this brand in Brandsearch

The British menswear brand runs 148 opening lines, and the most-repeated is not a hook in any creative sense: "Shop the ARNE Collection Online or on the App now." on 153 ads. Below it, "Shop up to 50% off the Summer Sale at ARNE." on 25 and an A-TECH 18 pre-order line on 19.

That top line is a default, the sentence that goes on an ad when nobody wrote one. It is also, on volume, the single thing ARNE says most often to the market.

Where all that volume lands

The Landing pages view shows the other half of the structure. ARNE has 351 landing pages, and the top three each carry over a million euros of disclosed spend: /collections/sale at #1 with 397 ads and €1.4M, /collections/activewear at #2 with 322 ads, /collections/new-arrivals at #3 with 262 ads. The rank badges are worldwide; the euro figures are the European disclosure slice only.

A brand with one dominant opening line and 351 destinations is not running a narrow operation. The variety is in where the traffic goes, not in what the ad says on the way there.

See which pages the ads in your niche actually point at

8. Laura Geller: 214 ads behind one opening line

See this brand in Brandsearch

Fifty-eight opening lines across 889 scanned ads. "New to baked?" opens 214 of them, "Time to simplify your everyday makeup!" another 115, and a Baked Balance-n-Brighten Foundation line another 87.

Three words is an unusually short top line for this list, and it is doing a specific job: it addresses the one objection a baked-formula brand has, which is that nobody knows what baked means.

The same library, counted by full text instead of opening line

Switching from Hooks to Ad copy counts the whole body rather than the first line, and the number barely moves: 68 distinct texts across the same 889 ads, against 58 opening lines. The full text of "New to baked? We've got you covered!" is on 101 ads by itself.

That gap matters. If a library had many distinct bodies sharing a few openings, the repetition would be cosmetic. Here the bodies are nearly as repeated as the openings, which is the structural sameness the score is described as reading, rather than a surface effect.

Read the full ad copy behind a competitor's library

9. Reformed: 134 ads behind one opening line

See this brand in Brandsearch

Forty-seven opening lines across 631 ads. The top line, "What if your morning coffee did more than wake you up?", carries 134, and the second, a first-person matcha line, carries 119.

Those two are structurally different in the way that actually counts: one is a question aimed at a habit the reader already has, the other is a personal testimony. Same product, two different opening moves, and the volume split almost evenly between them. Of the concentrated half of this list, Reformed is the clearest case of a brand hedging across two distinct structures rather than one.

10. AYBL: 45 ads behind one opening line

See this brand in Brandsearch

Twenty-eight opening lines across 260 scanned ads, topped by "New in: Adapt collection!" on 45, then "New in: Enhance collection!" on 40 and "AYBL Summer Sale!" on 31.

The top two lines are the same sentence with the collection name swapped. Counted as distinct opening lines they look like variety; read as structure they are one template filled in twice, which is precisely the difference between counting creatives and counting concepts.

Everyone else selling the same thing

The Competitors view reframes the question from "is this brand repetitive" to "is this category repetitive". Brandsearch finds 200 stores selling products that look like AYBL's, offers to open the 460 ads running those products, and scores the closest matches: Gymshark at 94% match, TEVEO at 94% match with 738.9k monthly visits and 863 active ads.

If your opening line is a template and 200 stores sell your product, the template is probably theirs too. That is the version of the similarity problem worth worrying about, and it is not one you can see by looking at your own account.

Find every store selling what you sell

11. Wonderskin: 163 ads behind one opening line

See this brand in Brandsearch

A genuinely broad library: 149 opening lines across 1,158 ads, only 36 of which are catalog ads. The top line, "👁 Glide on at 9AM, still flawless at midnight.", carries 163, and a lip-stain line carries 134.

At 14% concentration this is the point on the list where no single message dominates. The top three lines are a wear-time promise, a product-origin story and a needle-free comparison, which are three different arguments rather than three phrasings of one.

12. MyVariations: 117 ads behind one opening line

See this brand in Brandsearch

The French mouthcare brand runs 48 opening lines across 1,009 ads, with a toothbrush-technology line on 117 and a straightforward "-15% sur une sélection de produits." discount line on 64.

The second row is the one to note. A discount line sitting that high is a library where a promotional mechanic, not a product argument, is the second most repeated thing the brand says.

13. BLACKROLL: 21 ads behind one opening line

See this brand in Brandsearch

The flattest distribution in the top half of this list. Fifty-one opening lines across 247 scanned ads, and the most-used one only reaches 21: a German line about an orthopaedic neck pillow. The second and third reach 19 and 16.

Nothing dominates. For a brand selling foam rollers, massage guns, compression boots and pillows, that is what you would expect, because each product line needs its own argument and none of them subsidises the others.

14. Fussy: 93 ads behind one opening line

See this brand in Brandsearch

Four hundred distinct opening lines across 1,416 ads, and the most-repeated reaches only 93. The top three are all variations on the same switch-to-Fussy claim, but they are three of four hundred rather than three of thirty.

A refillable deodorant is a single product with a single mechanic, so 400 lines is a deliberate choice rather than a consequence of a wide catalog. This is what the diversity advice actually looks like when a brand follows it.

15. On: 31 ads behind one opening line

See this brand in Brandsearch

The most diverse library measured: 499 opening lines across 1,117 ads, which is close to a distinct opening for every second ad. The most-repeated line, "Shop Last Season", reaches 31.

On is also the largest advertiser here by a distance. Thirty-five times more opening lines than DOONAILS, on a library only eleven times larger. Whatever the right answer is, the two ends of this list are not doing the same thing by accident.

What the 15 have in common

Read together, the list does not support the advice it was supposed to. Three things hold across it.

Concentration and diversity are independent, and you need both numbers. Elavate runs 115 opening lines and is 54% concentrated. AYBL runs 28 and is 17%. Neither figure tells you anything on its own, and most writing on this subject only ever quotes one of them.

The most repeated line is often the best performing asset. prepmymeal's 636-ad line is attached to an ad ranked in the top 1% of 325 active ads with €128k of disclosed spend across a 517-day run. Killing repetition for its own sake would have killed that first.

A lot of what looks like variety is one template filled in twice. AYBL's top two lines differ by a collection name. The Essence Vault's top two make the same price argument. Laura Geller's full ad texts are barely more varied than its opening lines, 68 against 58. If you count creatives you will conclude these libraries are diverse. If you count concepts you will not.

The practical read: measure both numbers on the brands you actually compete with, before deciding whether your own library is too samey. The answer is category-specific, and the spread on this page is 69% to 3%.

The harder version of the problem is the one AYBL's Competitors view points at: when 200 stores sell your product, the sameness that matters is between you and them rather than inside your own account. We took that apart separately in why your competitor's ads all look the same, and the question of which signals still read clearly after Andromeda is covered in three competitor research signals that still work.

Measure your own library against the 15

What the Creative Similarity Score actually is, and what it is not

Meta's Andromeda ranking work made creative the lever it pays to move, and creative diversity is something Meta itself talks about in that context. What Meta has not published is a score, a threshold, or a formula.

So treat every specific number you read about this, including the widely repeated six-ads-above-60% rule and the CPM ranges attached to it, as unsourced until someone shows you where it came from. This page does not repeat them, and an earlier version of this page did, which is why it was rewritten.

What is safe to say is narrower and more useful. Structural sameness is about layout, pacing and copy architecture rather than palette or font. Two ads with different footage, the same opening line and the same body text are the same opening move made twice. And the count of how many times a brand makes the same opening move is a real, observable number that you can pull for any brand on Meta, including your own competitors, in about a minute.

The same caution applies to the other Meta surface everyone is guessing about right now. Nobody outside Meta can see how Meta AI Shopping picks the brands it recommends either, so the useful move there is the same one: measure the signals you can actually observe rather than the ranking you cannot.

Method

Fifteen Shopify DTC brands, pulled on 12 September 2026 from the Brandsearch brand index of 13.7M stores (9M Shopify), filtered to brands running at least 150 active Meta ads with at least 400,000 monthly visits, then sorted by disclosed EU spend and hand-checked.

Every "Sells" cell was read from the brand's own store description on its Brandsearch brand card, never from memory. That check is in the process because an earlier article described a collagen brand as a skateboard company.

Opening-line counts, ads scanned and catalog-ad counts are read off the Hooks view for each brand, which is the same view each screenshot shows. Percentages are the top row's ad count divided by the ads scanned in that view's own header. Regional storefronts were dropped rather than counted twice, because they show a different ad set and their fallback ads return unusable spend figures.

Three things were dropped rather than softened. Per-line rank and run-length render only when a Hooks row is expanded, so no rank or run-length is quoted from a Hooks row anywhere on this page. The Scripts view returned an empty state on the brands tried, so nothing here rests on ad transcripts. And ad-level rank, version count and spend are quoted only for the single prepmymeal ad whose card is shown, because that is the only ad card photographed.

Ad counts move daily. The figures here are a reading, not a constant.

The API and MCP behind this list

Every brand-level figure on this page came out of the public API before it was checked against the app. The filter that produced the candidate set:

curl -s -H "X-API-Key: $BRANDSEARCH_API_KEY" \
  "https://api.brandsearch.co/v1/brands?platform=shopify&brand_type=dtc%20brand\
&meta_active_min=150&monthly_visits_min=400000\
&sort_by=eu_total_spend&sort_order=desc&page_size=100\
&fields=id,name,description,niche,country_code,monthly_visits,last_meta_active_count,eu_total_spend"

And the ad-level pull behind the prepmymeal card, which is where the version count and the disclosed spend come from:

curl -s -H "X-API-Key: $BRANDSEARCH_API_KEY" \
  "https://api.brandsearch.co/v1/brands/prepmymeal.com/ads?platform=meta\
&sort_by=eu_total_spend&sort_order=desc&page_size=25\
&fields=id,status,start_date,end_date,total_active_time,duplicate_count,reach_rank,eu_total_spend,eu_total_reach,creative"

duplicate_count is the field the app renders as "Versions", and reach_rank is the field behind the rank badge. The same data is available over MCP, so Claude, Cursor or ChatGPT can run this measurement on your own competitor set without you writing the loop.

Connect Brandsearch to your AI tool

FAQ

What is the Meta Creative Similarity Score?
It is the industry name for how Meta's Andromeda ranking treats structurally similar ads from the same advertiser, grouping them rather than letting each compete separately. Meta has not published a score or a threshold, so any specific number you see quoted for it is someone's estimate.

Can I see my own similarity score?
No. There is no such field in Ads Manager and no API that returns one. What you can measure is the input: how many of your ads share an opening line, a body text and a structure.

How many creatives should I actually run?
The 15 libraries here range from 14 distinct opening lines to 499, and both ends include brands with top-ranked ads. The useful benchmark is the brands you compete with, not a universal number.

Is repeating one hook across many ads bad?
Not on this evidence. The single most repeated line in the set, on 636 ads, belongs to an ad ranked in the top 1% of its brand's 325 active ads with a 517-day run. Repetition of a proven message is different from failure to test.

Why are the spend figures in euros when the rank is global?
Rank badges are computed worldwide across everything a brand runs. Spend and reach are only the slice EU and UK disclosure rules force advertisers to publish. A modest euro figure next to a high rank means most of the budget is undisclosed, not that the ad underperformed.

Where does the catalog-ad number come from?
Meta fills in the text on catalog ads per product, so Brandsearch separates them out and reports them beside the counted ads. The 15 brands here carry 7,713 catalog ads on top of the 11,588 ads counted.