Creative diversity in Meta ads: what Meta's rating means and how to raise it

Creative diversity in Meta ads is how different the ads in an ad set look from each other. Meta now rates it in Ads Manager as Low, Medium or High per ad set, by comparing the images and video thumbnails for similarity. The rating is labelled estimated and in development, and Meta has not published its thresholds.

Last checked 10 October 2026 against Meta's Business Help Center definition and press coverage. We update this page when Meta publishes more.

What Meta's creative diversity rating is

Creative diversity is a column in Meta Ads Manager that gives each ad set a label: High, Medium or Low. Meta's Business Help Center definition explains the calculation in two sentences. It compares the images and video thumbnails in the ad set with each other, turns their similarity into a score, and applies thresholds to that score to get the label. The stated purpose is to show when an ad set needs more varied images and videos, to help prevent creative fatigue and improve performance.

The metric carries two of Meta's metric labels. Estimated means the figure is modelled. In development means Meta can change how it works. Practitioner write-ups date its arrival in Ads Manager to around 26 August 2026, and it has not reached every account at the same time.

What is confirmed and what is only reported

Most of what is written about the rating goes beyond Meta's two sentences. This is where each claim stands.

When a blog quotes a similarity percentage as the cut-off for Low, ask where it came from. None of the figures we found cite Meta, and two of the most repeated ones (60% and 65%) contradict each other.

How to find the Creative diversity column

1. Open Ads Manager and go to the Ad sets tab. The label is per ad set, so the Campaigns tab will not show it.
2. Click Columns, then Customize columns.
3. Search for “Creative diversity”, tick it and click Apply.
4. Save the layout as a column preset so the rating sits next to Amount spent, Cost per result and Frequency every time you open the account.

If the search finds nothing, the metric has most likely not reached your account yet. The rollout has been uneven.

Why an ad set with many ads can still read Low

Search Engine Land reported client accounts that ran static images, video, carousels, UGC, partnership ads and boosted organic posts, and still saw Low on each one. The exact thresholds are unknown, but Meta's definition gives one plain reason: it names images and video thumbnails, and nothing else. It does not mention copy, audio or the script.

Read literally, that means a video is compared by its thumbnail. Ten UGC videos with ten different scripts, each opening on a person talking to camera in a similar kitchen, give ten similar thumbnails. A static, a carousel card and a video can also look alike when all three show the product on the same white background. A different format does not guarantee a different picture.

Ad count does not appear in Meta's description either. Adding a ninth and tenth ad that look like the first eight should not move the label.

Creative diversity, Andromeda and the similarity score

Andromeda is the ad retrieval system Meta described in a December 2024 engineering post. Retrieval is the first step of delivery: it narrows tens of millions of candidate ads to a few thousand before the ranking models pick what a person sees. The common reading among buyers is that near-identical ads give that step little to choose between.

Meta has said little beyond that. In a September 2026 statement to Marketing Brew, a Meta spokesperson called diversifying creative a longstanding best practice, and did not explain how its models represent individual creatives. Agency executives in the same story said Meta told them that ads sharing the same look or opening hook are treated as the same ad. That is reported. Meta has not documented it.

You will also see a “creative similarity score” discussed. Meta told the same publication that the creative metrics it is testing in its reporting API have gone to a small number of agencies since June 2026 and are not widely available. Treat any similarity threshold you read as a guess.

How to raise a Low rating

• Change what the first frame shows. Meta compares images and thumbnails, so a new person, a new setting or the product in use moves the comparison more than a new headline. Set a custom thumbnail on videos that open the same way.
• Build new concepts before new variants. A different reason to buy (price, a problem solved, what other customers say) usually forces a different picture. A new background colour or call-to-action button does not.
• Vary the people. Three creators filmed in three homes look different. One creator in three shirts does not.
• Mix composition along with format. A text-led graphic, a lifestyle photo and a demo video differ. A carousel whose first card repeats your main static adds little.
• Pause near-copies. If the label works the way the definition reads, five ads with the same thumbnail weigh on the comparison. Removing the weakest of them can help as much as adding new ones.

How to check whether Low costs you anything

Meta has not said that a Low label reduces delivery. Before you rebuild your production plan around it, test it against your own numbers. Once a week, export the Ad sets view with Creative diversity, Amount spent, Cost per result and Frequency (Reports, then Export table data) and keep every file.

After a month, compare cost per result for ad sets that read Low with those that read Medium or High, within the same objective and result type. If the Low ad sets are no more expensive and show no signs of creative fatigue, the label can wait for the next production round.

Common mistakes

• Treating the label as a penalty. Meta describes it as a signal to add varied images and videos.
• Adding more ads to fix it. Meta's description says nothing about ad count.
• Counting formats as concepts. The same script cut as a Reel, a Story and a square feed video is one idea, often with three near-identical thumbnails.
• Copying a vendor's similarity threshold into a creative brief as if Meta had published it.
• Comparing labels across objectives. A retargeting ad set with two ads and a prospecting ad set with fifteen answer different questions.

Where the money went: the spend-weighted view

Meta rates what you uploaded. This shows where the budget went. An ad set can hold six different ads while delivery puts most of its spend behind one of them, and the people who saw your ads then mostly saw one idea.

To measure it, group each ad into a bucket by format (video, image, carousel) and by content type (UGC, testimonial, demo, pack shot, lifestyle, talking head, before and after). A bucket is a combination such as “video × testimonial”. Take each bucket's share of spend, square it, add the squares and invert the total:

Effective number of creative types = 1 ÷ (s₁² + s₂² + … + sₙ²), where s is each bucket's share of spend

Economists use the same calculation, the inverse Herfindahl index, to measure how concentrated a market is. The result reads as a count. A score of 1 means one bucket takes all the spend. A score of 4 means spend behaves as if it were spread evenly over four. Squaring the shares makes a dominant bucket pull the number down hard, while small buckets barely lift it.

By hand: export the Ads tab for 30 days, tag format and content type (a naming convention such as “VID_UGC_hook2” makes this a filter), sum spend by bucket and apply the formula. Do it per campaign too, since an account can score well while each campaign runs one type.

Worked example

Example with illustrative numbers: four accounts, each with a month of spend.

Account A runs three types of creative, but with 70% of spend on one it behaves like an account with fewer than two. Account D has the most ads of the four and the lowest score. B and C both run four types, and the even split gives C the higher number.

Read the two measures together. A Low label with a score near 1 means the uploads look alike and the money follows them. A Medium or High label with a score near 1 means you uploaded variety but delivery backs one type, so check whether that type's cost per result earns the concentration. A low score can wait on a small account, where one product at 5.000 kr a month needs only two types to compare.

How Adsverse.io measures it

Adsverse.io calls this score the Creative Diversity Index (CDI) and calculates it exactly as above. It uses the ads optimised for the account's conversion result (purchases, leads, registrations, installs), so reach and traffic spend does not dilute the number.

Format and content type are tagged automatically from the ad name and the creative, and you can correct any content-type tag by hand. When too few ads carry a content type to form comparable buckets, the score falls back to format alone and says so.

Next to the score you see the largest bucket and its share of spend, plus spend, ad count and cost per result for each bucket. A bucket is only ranked against the others when it has at least three ads and 5% of spend, so one lucky ad does not crown a winner.