Twenty variants of one idea count as one ad

Twenty two ads sit in the folder, ready for Monday. One concept sits under all of them. The same creator, the same kitchen, the same complaint at second three. Nineteen of them change only the opening line of text. Two swap the music track. One is the same video with a fresh thumbnail. The team calls it a heavy test week. The client deck will say twenty two tests. The media buyer knows better and says nothing, because volume is what gets praised on the Monday call.

It is one test. That was always true, and most of us have quietly known it for years. What changed in August is that Meta now prints it inside the account, in a box your client can read without you in the room.

Meta started scoring variety in August

On 26 August 2026 Meta shipped a Creative Diversity rating in Ads Manager. Common Thread Collective wrote it up on 9 September 2026. Every active ad set now carries a rating of Low, Medium or High. The score reads five things:

• Content style, meaning the look and production level of the asset
• Messaging theme, meaning the argument the ad actually makes
• Hook type, meaning how the first two seconds work
• Format, meaning video, static, carousel and so on
• Spokesperson, meaning who is on camera

The stated mechanic matters more than the label. Per Common Thread Collective, near identical ads get collapsed and counted as one creative by delivery. So the number of ads you upload and the number the auction sees are two different numbers. They have probably been different for a while. Meta calls the score estimated and in development. It has published no methodology document, which is a tell in itself.

Twenty two rows in the table. One creative in the auction. Your test week tested one thing.

Foxwell and Sanchez are both right about their own accounts

Andrew Foxwell argued on 31 August 2026 that the bottleneck moved. Execution is no longer the hard part. Ideation is. He also argues that AI generated user style video fatigues faster than authentic creative. Fatigue here means the point where enough of the same people have seen an ad that response drops off. Good human work, in his account of it, can hold the top of an account for months rather than days.

Savannah Sanchez runs the other side of the argument, and runs it well. Motion published her setup in 2026. More than 200 ads a week, across more than 50 clients. She has 40 creators and 10 editors on a five day cadence. That is a real machine. Note what is not published alongside it. No win rate. No conversion benchmark. The number that travels is the output number.

Both descriptions are honest. They describe different businesses. Sanchez has 40 creators and 10 editors, so her volume is varied by construction. Forty people do not converge on one idea. A two person team shipping 200 ads a week is shipping about six ideas in twenty outfits each. Under the new rating those two setups score nothing alike, at an identical ad count. The ad count was never the thing being measured.

The objection I would make against my own case

Here is the strongest version against me. The score is labelled estimated and in development. Meta published no methodology. Advertisers have been handed unaudited quality labels before and were right to ignore them. Optimising toward a number nobody outside Menlo Park can inspect is a good way to make an account worse. And if Sanchez ships 200 ads a week and keeps 50 clients, why should a Low rating change anyone's process?

Because the rating and the mechanic are separate things. The rating is a report. The collapsing is the behavior being reported. You can distrust the label completely, refuse to chase it, and still be paying for twenty ads that deliver as one. The score is new. The collapse is not. Ignore the box if you want. You cannot ignore what the box describes.

Price settles the rest. Meta's Q2 2026 filing, dated 29 July 2026, reports ad impressions up 14 percent year over year. Average price per ad rose 12 percent. At flat spend you are buying roughly 12 percent less auction weight than a year ago. Auction weight here just means how much delivery your money actually wins. Redundant ads were affordable when inventory was cheap. Inventory is not cheap now, and the redundancy no longer even registers as separate entries.

Five ideas beat twenty versions

What counts as a genuinely different idea is not mysterious, and the axes give you the map. Motion, in its 2025 Ad Creative and Creative Strategy Trends report, found 42 percent of top spending ads use lo-fi production. Lo-fi means phone shot and unpolished, the opposite of a studio edit. The same report found only 14 percent of social ads use humour. Among ads spending over 1 million dollars, that rises to 25 percent. Humour is a whole messaging axis most accounts have never once tested.

So here is Monday. Open one ad set that is live right now. Put every active ad in a sheet and tag each one on Meta's five axes. Content style, messaging theme, hook type, format, spokesperson. Then count the distinct combinations instead of counting ads. Most accounts land on three or four. That number is what you brief against this week, not the ad count. Then turn off the near copies yourself. Delivery has already decided they are one ad, and you are paying for the difference.