How to Use Competitor Ad Data to Fix Your Creative Similarity Score

Your CPMs are rising because Meta thinks your ads are the same ad. Here's the fix — and it starts with your competitors' creative library, not yours.

What the Creative Similarity Score Actually Measures

Meta's Andromeda update changed how ads compete in the auction. Your creatives now get an Entity ID — a fingerprint based on visual structure, layout, and copy patterns.

When two or more of your ads share a high similarity fingerprint, Meta groups them. They compete against each other instead of against other advertisers.

The result: fragmented delivery and higher CPMs. You're bidding against yourself.

The threshold most advertisers report trouble at is 6 ads above 60% similarity. Past that point, your Advantage+ Shopping Campaigns start cannibalizing reach. Your cost per result climbs while delivery spreads thin across creatives that Meta treats as duplicates.

This isn't about image quality or production value. A $50K shoot and a $200 UGC clip can both get flagged if the visual structure — framing, text overlay placement, color palette, pacing — is too similar.

The score is structural, not aesthetic.

The Andromeda system evaluates three layers:

Visual layout. Where are the elements on screen? Product left or center? Text overlay or clean image? Split-screen or full-bleed?

Temporal structure. For video — what's the pacing? Hook in 0–2 seconds, demo in 2–8, CTA at 12? Two videos with identical pacing score as similar even if the footage is completely different.

Copy architecture. Headline length, body text density, CTA type. Two ads with "Shop Now" buttons, 3-line headlines, and 2-paragraph body text look identical to the model — regardless of what the words say.

Why Small Tweaks Don't Lower Your Score

The first thing most advertisers try: swap the background color, change the headline font, crop the image differently.

It doesn't work.

Meta's similarity model reads deeper than surface elements. It looks at layout composition, text-to-image ratio, color distribution, and copy cadence. A green background instead of white doesn't move the needle when the framing, CTA placement, and visual hierarchy are identical.

This is the trap of incremental variation. You end up with 12 creatives that look "different" to you but register as the same ad to the algorithm.

I've seen accounts with 12 "different" creatives that Meta grouped into 2 entities. Same product photo centered. Same headline-above-image layout. Same 15-second video pacing. The only differences were color and copy — which the model largely ignores when scoring structural similarity.

The performance impact on Advantage+ campaigns is brutal. ASC relies on Meta's ML to allocate budget across your creatives. When the system groups 8 of your 12 creatives as one entity, your effective creative count drops to 5. You're paying for 12 slots but getting the distribution of 5.

Real diversity means changing the angle — not the styling. A testimonial UGC ad, a product demo, a comparison chart, a founder story, and a meme-format hook are five structurally distinct creatives. Five color variations of the same product-on-white-background are one creative in Meta's eyes.

Incremental tweaks vs structural diversity — Meta sees the top row as one ad

The Real Fix: Map What's Already Working in Your Niche

The advice you'll find everywhere is "diversify your creatives."

That's useless without a method. Diversify how? Into what formats? Based on what?

Guessing at diversity means testing random ideas. Some work, most don't. You burn $500–$2,000 per angle finding out which ones have legs — when brands in your niche already figured that out months ago.

The problem with "be more creative" is that it isn't a brief. Your creative team needs direction. "Make something different" produces a mood board session, not a structurally distinct ad.

The better approach: look at what's running profitably in your category and identify the distinct creative clusters. Not individual ads. Clusters — groups of ads that share a visual strategy, hook type, or structural format.

A fitness supplement niche might have 6–8 active clusters right now:

UGC testimonial — creator talking to camera, before/after text overlay
Ingredient breakdown — close-up product shots with callout labels
Comparison format — side-by-side "Brand X vs us" split screen
Problem-agitate — opens with a pain point, 3-second hook
Lifestyle context — product used in a gym, kitchen, morning routine
Data-driven — screenshots of lab results, clinical stats, percentages

Each cluster is structurally different. Different framing, different pacing, different visual weight. Run one creative from each cluster and your similarity score stays low — with every angle validated by the market.

You don't need to invent new formats. You need to see which ones exist.

Stop reading about winners. Find them yourself.

Search 6.5M+ brands, their ads, revenue, and products — all in one place.

Try Brandsearch free

How to Find Your Niche's Creative Clusters

Here's the workflow. It takes about 20 minutes and gives you a creative brief your team can execute against.

Step 1: Pull winning creatives in your niche.

Open Discovery, filter to your category, set `Phase: Winning` with `Running Days: 25+`. Sort by reach. You want ads that survived — not tests that ran for 3 days and died.

Filter to video first. Video creatives have more structural variety than statics, and Meta's similarity model weighs motion and pacing heavily. You can also hit the "Video ad winners" preset — one click applies video format, 25+ running days, and 100+ active ads.

Save 20–30 winning ads across 5–8 different brands to a Swipe File folder. Don't cherry-pick from one brand — spread it across competitors so you see the full range of what works in your vertical.

Discovery page filtered to winning video ads sorted by reach

Step 2: Group them by visual strategy.

This is where most people stop. They save a swipe file and call it research.

Go through your 20–30 saved ads and sort them into structural groups. You're looking for shared patterns in framing, hook type, visual layout, and CTA placement — not shared products or brands.

You'll land on 6–8 distinct clusters. Label each one: "UGC talking head," "split screen comparison," "product demo close-up," "text-heavy scroll stopper," "lifestyle montage."

Brandsearch Creative Tests does this automatically for individual brands. Open a competitor's Brand Analysis page, click the Creative Tests tab. You'll see their ads grouped by visual and copy strategy — ranked by run-length and reach. The clusters that survived longest are the ones the market validated.

Do this for 3–4 competitors and you have a map of every creative format that works in your niche. You're not looking at individual ads anymore — you're looking at the structural vocabulary of your entire category.

The key: don't copy the ad. Extract the structure. Hook type, format, visual weight, CTA placement. Those patterns transfer to your own product. Specific creatives don't.

Step 3: Build your creative brief from the cluster map.

Take your 6–8 clusters. For each one, write a one-line brief:

Format type (UGC, studio, product demo, comparison)
Hook structure (question, stat, pain point, outcome)
Visual weight (text-heavy, image-heavy, motion-first)
Reference ad (link to the best-performing example)

Hand this to your creative team. They're not copying ads — they're building one creative per cluster. Each one is structurally different by design, because the clusters themselves are structurally different.

Your similarity score stays low because you built the brief around structural diversity, not random variation.

Step 4: Score each cluster for your brand.

Not every angle works for every product. A $200 kitchen gadget won't benefit from meme-format hooks — but it will from product demos and founder authority. A $29 impulse-buy supplement leans hard into UGC and before/after.

Cross-reference the clusters against your existing ad library. If you're already running 3 UGC testimonials and zero product demos, that's your gap. The cluster you're missing is the one that lowers your similarity score the most.

Pick the 4–6 clusters most relevant to your product. Discard the rest. That's your creative brief — backed by market data instead of a brainstorm.

The Numbers That Matter After Launch

After you launch creatives from 5–6 different clusters, check two things in the first 7 days.

CPM trend. If your CPMs drop 15–25% compared to your previous batch, the algorithm is treating your creatives as distinct entities. You stopped bidding against yourself. In a typical $12–$14 CPM niche, accounts with high similarity scores often see CPMs spike to $18–$20. After switching to cluster-based creative, that same niche drops back to the $12–$14 range within 72 hours.

Delivery distribution. In Ads Manager, check the breakdown by creative. Healthy distribution looks roughly even — each creative getting 10–20% of spend. If one creative eats 70% while others get nothing, you still have a clustering problem. The goal is balanced distribution where ASC can test each angle with enough budget to exit learning phase — usually $50–$100 per creative per day for most e-commerce accounts.

Learning phase exit speed. With structurally diverse creatives, ASC typically exits learning in 3–5 days instead of cycling indefinitely. When the algorithm can distinguish between your ads, it finds the winner faster and allocates spend accordingly.

Brands that consistently scale on Meta run 6–10 structurally distinct creatives at any given time. Not 20 variations of the same concept. Not 3 safe bets. A deliberate spread across validated formats.

The specific clusters that tend to perform best across e-commerce verticals: UGC testimonial, product demo, and problem-agitation hook. Those three alone typically cover 70%+ of winning spend. Founder authority and lifestyle angles fill the remaining slots and keep the similarity spread healthy.

One more thing. The similarity score isn't static. Every time you add a creative that's structurally similar to existing ones, the score climbs. Every time you add a genuinely different angle, it drops. Monitor it the same way you monitor frequency — if it creeps up, your next creative should be from a new cluster, not a variation of your best performer.

Track this over time by watching competitors in Brandsearch Brand Analysis. Check the Overview tab for traffic trends and ad scaling. A brand with steady traffic growth and a diverse Creative Tests tab is running this playbook.

The 15-Minute Weekly Creative Audit

Run this every Monday before your team starts new creative.

Open your Brandsearch Swipe File folder from last week. Check which saved ads are still running. If they're still live, the cluster is still valid.

Open Brandsearch Discovery. Check your niche for new winning creatives in the last 7 days. New cluster emerging? Add it to your brief.

Check your Ads Manager. Which clusters are getting the best delivery distribution? Double down on winning structures. Kill the ones Meta isn't distributing.

Update your cluster map. Remove dead formats, add new ones, adjust reference ads.

15 minutes. Your creative pipeline stays structurally diverse without guessing, and your similarity score stays below the penalty threshold.

The brands that keep CPMs low aren't more creative than you. They're more systematic. They map the angles that work in their niche, produce one ad per angle, and replace underperformers with new clusters — not new color variations.

The Bottom Line

Meta's Creative Similarity Score isn't a mystery. It's a structural analysis of your ad library. When your creatives share the same layout, framing, and visual patterns, the algorithm groups them — and your CPMs pay the price.

The fix isn't random variation. It's deliberate structural diversity based on what's already working in your niche.

The method:

Pull 20–30 winning competitor ads — Brandsearch Discovery, Meta video, 25+ running days
Group them into 6–8 structural clusters — Brandsearch Creative Tests tab on 3–4 brands
Build one creative per cluster — each structurally different by design
Audit weekly — check similarity score, add new clusters, kill dead formats

Stop guessing at diversity. Map it from the market and build your brief around proof.

Your CPMs spiked because Meta grouped your ads together. Give it something genuinely different to work with.