Meta Creative Diversity Score: Low, Medium, High, and what to do

According to the Common Thread summary of 2026 Meta Ads changes, Meta introduced a Creative Diversity Score on 2026-08-26, shown as Low, Medium or High. The practical reading is simple: ads in one account that look and say the same thing are likely to score lower than ads that differ. The summary does not give a formula, and this post does not claim one.

Everything about Meta below comes from that one third-party page, read on 2026-10-02. Check Meta's own Ads Manager help before you rely on any threshold. Sume facts come from Sume's video generation docs and Image API docs.

What is the Creative Diversity Score?

It is a three-level rating, Low, Medium or High, that the summary dates to 2026-08-26. The same page lists several other August and September changes, so it helps to see them side by side.

Only the rows marked as the summary's words are Meta facts here. The last column is what Sume can do about it, from the docs.

What counts as a different creative?

Meta's definition is not in the summary, so treat any answer as a hypothesis to test in your own account. A reasonable working rule is that a variant differs when a viewer would notice it in the first second: another opening line, another subject, another setting, another format.

A caption restyle on the same footage changes how the ad looks but not what it shows. A new first frame and a new hook change both.

How do I make more different creatives with Sume?

Vary one thing at a time on purpose, and keep a record of which axis changed. Sume's Image API takes a prompt plus up to 16 reference images on the ChatGPT Image 2.5 models, so one product shot can sit in several different scenes.

For video, POST /v1/videos accepts frame_images for a first frame and input_references for style guidance, so the same product can open each clip differently. Send an Idempotency-Key so a retry does not buy a second clip.

• Change the hook line, not only the font.
• Change the setting or subject in the first frame.
• Mix still ads and short video ads in one ad set.
• Keep a sheet with one row per ad and one column per axis.

What should I not conclude?

Do not conclude that more ads always raise the score, or that the score moves delivery. The summary reports the levels only. Make a batch, read the score in Ads Manager, and let your own numbers decide the next batch.

If you use Meta's own generative tools, their labels and defaults are separate questions; see the related posts on Advantage+ and AI labels.