Creative Diversity: Low. Why 30 New Ads Won’t Fix It

Monday, 9:40 a.m.

Sam, our growth lead from Issue #2, opens Ads Manager. A new column called Creative diversity says Low next to the ad set that pays the bills. Sam asks the agency for thirty new ads by Friday.

The agency takes the brief literally: new backdrops, new colours, a static, a carousel, an unboxing. Thirty ads that look nothing alike. Two weeks later, the old ad still takes the budget.

Every one of the thirty spoke to the same buyer, at the same moment, with the same reason to buy. They looked different. To Meta’s new process, they were similar.

Thirty new looks. One buyer, one reason.

Meta rebuilt three layers in October

The one part you can see

The Creative diversity column started appearing in late August and reached most accounts in the last week of October. It labels each ad set high, medium or low:

“Creative diversity is calculated by comparing the images and video thumbnails in your ad set to each other to determine similarity.”

Meta Ads Manager help text. Meta calls the metric estimated and in development.

The column only compares pictures. The three layers behind it read far more: your words, your buyer and your product. If the column rates an ad set Low, treat that as a smoke alarm, not a verdict.

What Galvor saw in the data

Across the brands we work with, we keep seeing the same pattern. Here is how it looked in the last six months of ads Galvor analysed: every new video ad launched from May to September 2026, and how much of the budget it got in its first month.

The bottom row is the one to remember. A new message over a familiar shot got 16% less: the script changed, but the ad still looked like one already running. A proven message in a new look did best, with 47% more. A new reason needs a new first shot.

How to think about diversity: Galvor’s five layers

“Creative is the new targeting.” “Your job is to give Meta enough creative diversity to match the right message to the right person at the right time.”

Lines media buyers have repeated over two years of Andromeda.

If the system reads ads, shoppers and products, a new ad has to be new in more than its look. Galvor scores every ad on five layers. Two are what the camera sees. Three are who the ad is for and what it says.

The shift for marketers and leaders

Five moves Galvor recommends.

4 Split every angle into motivation, situation and objection

A broad angle like “better sleep” is a family of ads, not one ad. Each motivation, situation and objection gives Meta a separate reason to match a different person.

5 Organise ad sets by buyer, not by angle

One shopper can care about saving time, proof and price at once, and Meta can now show them several angles in turn. One ad set per buyer, with each angle in the format that proves it best, gives it room to do that.

Next issue: Beyond Narratives and Categories: How Galvor Uncovers the Psychology Driving Your Brand

Narratives and categories tell you what an ad is about. They don’t tell you why it works. Next time, we show how Galvor decodes the buyer psychology behind a brand’s best ads, and builds hooks, scripts and campaigns on it.

The three systems, explained

Andromeda. Meta’s retrieval system. For every person and every ad slot, it narrows tens of millions of ads to a shortlist of a few thousand worth ranking. An ad that misses the shortlist never reaches the auction. Meta’s post, 28 Oct 2026

GEM. Meta’s Generative Ads Model, which Meta calls the central brain of its ad ranking. It learns from what people do across Facebook and Instagram, in ads and in posts, and trains the models that rank ads. It doesn’t pick ads itself. Meta’s post, 10 Nov 2025

GenRecs. Meta’s generative recommendations work. It uses language models to describe every ad and product with a shared set of building blocks, which Meta calls semantic tokens, so a brand-new ad can borrow what Meta has learned about similar ones. Meta’s post, 19 Oct 2026