Creative diversity isn't more ads. What Meta's Andromeda actually rewards before Q4.

Meta's Andromeda ranks on creative, not audiences. Creative diversity means different hooks and angles, not more variants. A Q4 plan for small ad teams.

‍

Meta's Andromeda ranks ads on creative rather than audiences. The advice that followed — run 15 ads, run 20, run 30 — is the least useful part of the story, because creative diversity has very little to do with how many ads sit in the ad set.
‍

If you are being told to triple your output and you have one designer and a fixed Q4 calendar, this is the piece worth reading before you book anything.

‍

What Andromeda actually changed

Andromeda replaced audience-based targeting with creative-based targeting across most objectives and placements. Instead of you selecting interests, demographics and lookalike seeds, the system reads the creative to work out who should see it.

‍
The consequence is easy to state and hard to internalise: you are no longer only competing in the auction. You are competing to be retrieved for it. An ad that reads as a near-duplicate of another ad in your account is not a second chance at the auction — it is the same chance, split.

‍
This is why the skill that produced returns in 2022 was audience configuration, and the skill that produces them now is creative production. It is a genuinely different job, and most teams have not been resourced for the swap.

‍
Why the numbers everyone quotes disagree with each other

Six to eight creatives per ad set. Ten to fifteen. Fifteen to twenty. Twenty to thirty. All four of those ranges are in circulation right now, and they come from agency and tool-vendor blogs rather than from Meta. They disagree because they are describing different accounts at different spend levels, and because a round number makes for a better headline than "it depends on how different your ads actually are".

‍
Chasing the number is how you end up with twenty colourway variants and a low diversity rating. Meta has introduced a creative diversity indicator in Ads Manager rated Low, Medium or High — and practitioners report accounts that look genuinely varied, mixing statics, video, user-generated content and partnership ads, still coming back Low. The thresholds and the signals behind that rating are not public. Treat it as a directional prompt, not a target to game.

‍

‍

What counts as different

‍Akvile DeFazio's framing in Search Engine Land is the most usable list going: hook, creator or spokesperson, visual style, messaging angle, offer structure, customer pain point, emotional appeal, and format. Vary those and you have diversity. Vary anything else and you have exports.
‍

The working test is simpler than the list. If you removed the logo from two ads and showed them to a stranger, would they say those are two different ads, or the same ad twice?

‍
Same product, same model, blue background versus green background: one ad. Same product, with the founder explaining why they made it in one and a customer explaining why they kept it in the other: two ads. Aspect ratio, length and caption crop are not concepts. They are file formats.

‍

The small-team version

Almost all the published advice assumes creative capacity that most brands do not have. Here is the version for one designer, one videographer day, and a Q4 calendar that is already half-full.

‍
Shoot for hooks, not for ads

A single creator session should come away with five openings, not one video. The first three seconds carry more weight than anything else in the ad and cost almost nothing to vary — same set, same talent, same lighting, five different ways in. Each one addresses a different pain point or objection. That is five genuinely distinct ads out of one booking.

‍
Mine what you already have

Reviews, support tickets and returns reasons are a free, honest list of distinct customer pain points, written in the customer's own language. Each one is a messaging angle you do not have to invent, argue for, or brief. Most brands are sitting on a year of this and using none of it.

‍
Recreate the winner in ways that aren't the winner

Take your best-performing ad and rebuild the same claim five ways: a testimonial, a demonstration, a text-led static, a founder piece, a side-by-side comparison. Same message, five different experiences of it. This is the highest-yield hour in the whole process because you already know the message works — you are only varying the delivery.
‍

Cut variants, not concepts

When capacity runs out, three genuinely different concepts beat twelve variants of one. If you audit your live ads honestly, most accounts find they have around three concepts and twenty exports, and have been reporting that as twenty ads.

‍
The part that isn't creative at all

Andromeda is making a prediction about who will respond to a given creative. That prediction runs on your conversion signal, so signal quality is not a separate workstream from creative performance — it is a precondition for it.

‍
Have the Pixel and the Conversions API running together, and check event match quality rather than assuming it. A degraded signal produces a degraded prediction no matter how good the creative is, and it will look exactly like a creative problem in your reporting. Teams burn entire quarters solving the wrong one.

‍
Structure matters in the same direction. Fragmenting the same creatives across many ad sets splits the learning without adding a single new concept. Consolidation is not laziness here; it is what gives the system enough data per creative to tell them apart.

‍
The timing point

Creative lead times for Q4 close in September. Studio availability, creator schedules and retouching queues all fill before the campaigns they serve go live. If the plan involves any new production at all, the booking is this month's job — the shoot itself can happen later, the slot cannot.

‍
What to do this week

• Check the creative diversity rating on your main ad sets in Ads Manager. It is imperfect and it is still a free second opinion.
  ‍
• Audit live ads against the eight dimensions and count concepts, not files. Be honest about how many you actually have.
  ‍
• Pull ten distinct pain points from reviews, support tickets and returns reasons. That is ten messaging angles at no production cost.
  ‍
• Brief your next creator session for five hooks rather than one video.
  ‍
• Verify that the Pixel and Conversions API are both firing and check event match quality before committing more budget to production.
  ‍
• Book Q4 creative capacity now, before the calendar closes and the decision gets made for you

‍

‍