What counts as a look-alike
Two files that look the same at a glance in a feed: the same picture with a new headline, the same ad translated, often the same ad cut to a square and a vertical frame, or a video whose frames match an image you also uploaded. The check compares how each file is laid out, where it is light and dark and what colour each part is, and groups the files that score close.
- Look alike
- Files that would read as one ad to someone scrolling past.
- Stand apart
- Files with no close match in the batch.
- Different looks
- The number of groups, a lone file counting as one. It is the number to compare against how many ads you planned.
How videos are compared
A video is read at three points, near the start, in the middle and near the end, and two videos are compared frame for frame at those points. Two cuts that show the same shots at those points land in one group, whatever their opening line says. A video and an image are compared on whichever of the three frames matches the image best, so a video built around one of your statics usually joins that static's group.
Sound, voice-over and captions that change over time don't count. Two videos with the same pictures and a different voice are one look here, and in a muted feed they mostly are.
What it can't see
Meaning. The check doesn't know what is in a picture, so the same product shot on a table and on a windowsill count as two looks, and two ads with the same message and different photos are not grouped. A batch can pass here and still say one thing twenty times. Read the copy and the concept of each ad as well.
Meta's own creative diversity label
Meta reports a version of this for ads already in an ad set. Its help page says creative diversity “is calculated by comparing the images and video thumbnails in your ad set to each other to determine similarity”, and the score is “given a label of High, Medium, or Low by applying thresholds.” Meta calls the metric “estimated and in development” and doesn't publish its method or thresholds, so the groups here can't predict its label. What this check adds is timing and detail: it runs before the ads are uploaded, and it shows which files are alike rather than one label for the set.
Why it matters, in Meta's words from its December 2025 post on diversification: “When multiple ads look or feel alike, these are seen as variations of the same creative, meaning learnings and delivery optimizations are shared at the creative level”. To the system, ten look-alikes are one creative in ten versions.
Why the count matters on Meta
Meta's best-practices page on creative diversification recommends “maintaining at least 20 diversified ads in Advantage+ shopping campaigns”. Its Andromeda page defines diversification as “a wide range of ad creatives with different themes, messages, and visuals to cater to diverse audience segments.” A batch of near-copies meets the count and misses the definition. Meta's one creative-volume number goes through both pages.
What makes two ads different
Meta's definition names three things that can change: the theme, the message and the visual. This check only sees the visual, so use it for that third part and judge the other two yourself. Visual differences that register here include:
- a different format: a static, a video, a carousel's first card;
- a different subject in the frame: the product alone, a person using it, a hand holding it, a screenshot;
- a different setting or background, and a different main colour;
- a different layout: text on top, text in a side panel, no text at all.
Differences that don't register: a new headline on the same picture, a translated ad, a new button colour, a price change. Those are worth testing as variants of one ad, but they are one look, and counting them as separate ads overstates how varied a batch is.
Brand colours
Sharing a palette doesn't make two ads look-alikes. Ads in the same brand colours with different layouts and different subjects stay apart, because the check weighs where things sit in the frame as much as their colour. A batch that is all one colour can still be varied; a batch that is all one picture can't.
How to use the result
Look at the biggest group first. If it holds the same picture with different headlines, keep the strongest headline and spend the other slots on a new picture, a new format or a new angle. Sizes of one ad belong together, so a group made of one ad in several shapes is fine. Then compare the number of different looks with the number of ads you meant to test.
A routine for a test batch
- Export the batch exactly as you will upload it, one file per ad, sizes included.
- Drop it all in at once and note the number of different looks.
- For each group, write down in one line what idea the ads in it share. If you can't tell them apart in a line, they are one ad.
- Swap the extras for ads with a new subject, setting or format, and drop the batch in again.
- Stop when the number of looks is close to the number of ideas you meant to test.
The files never leave your browser, so an unreleased campaign can be checked here as safely as a live one.