For years, the advice for Meta ads was to find the right audience and then test ads into it. That order has flipped. Meta now says the creative does much of the targeting, and it has started treating ads that look alike as a single ad. Thirty files built from three ideas still give you a narrow range of messages to test.
This piece explains what changed, what Meta means by a genuinely different ad, and how to audit your own account in about an hour.
Benchmark context: Motion’s “winner” definition measures spend allocation relative to an account’s median, with a minimum spend floor. It does not establish profit, incremental sales or the results a new client will achieve. Use the findings to choose tests, then judge your account against its own economics.
What Andromeda is
Andromeda is the retrieval system in Meta’s ads stack. When someone opens Instagram or Facebook, it scans “tens of millions of ads and selects a few thousand strong candidates,” and later models rank those candidates for the person (Meta, Demystifying Creative Diversification). Meta’s engineering team described it in December 2024, reporting a 6% improvement in retrieval recall and an 8% improvement in ads quality on selected segments (Engineering at Meta).
You don’t need the engineering detail. You need the consequence: retrieval is choosing which ads get a chance, person by person. An ad that speaks to one kind of buyer clearly gives the system something to match. Ten near-copies give it one thing to match, ten times.
Meta groups lookalike ads
This is the sentence from Meta that every advertiser should read twice:
“Our ads system also analyzes creatives and groups together ads that share key visual and thematic attributes. 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.” (Meta, December 2025)
So twenty colour swaps of one ad add up to a single bet with twenty file names.
Iteration versus diversification
Meta separates the two explicitly. Iteration “might produce two ads with identical visuals, but different text CTAs.” Diversification is “two distinctly different pieces of creative,” such as “a Reels ad tailored for GenZ and a separate partnership ad featuring a creator who connects with new moms.” Iteration can inform new concepts, Meta says, but “should not replace true creative diversification.”
Both have a job. Iteration improves an idea that’s already working. Diversification finds the next idea and the next audience. Trouble starts when a team only iterates, usually because the first ad was expensive to make.
What “different” means in practice
Meta’s best-practice list asks for ads that are “truly different in look, feel, storyline, and message,” and suggests building them around different motivations, for example one ad on emotional appeal and another on product features. We use five levers as a briefing framework. Changing at least two is our practical heuristic, not a published rule for how Meta classifies an ad:
| Lever | Question it answers | Example (a pillow brand) |
|---|---|---|
| Person | Who is this for? | Side sleepers, new parents, people with neck pain |
| Reason to buy | Why would they act? | Wake up less, stop neck ache, finally a pillow that stays cool |
| Format | What shape is the ad? | Creator video, comparison static, notes-app post, demo |
| Visual world | What does it look like? | Dark bedroom at 3am, bright product studio, phone selfie |
| Proof | Why believe it? | A real review, a demo, a founder explaining the design |
Changing only the headline, the button text or the background colour is iteration. It’s useful later. It does not explore a substantially different buying reason.
Meta’s own persona prompt is a good starting point: it suggests younger shoppers may respond more to deals and value messages, and older shoppers more to reviews and demonstrations (Meta, Three steps to optimize your performance).
Production value is a weak lever
Diversity comes from ideas. Production budget matters much less. Motion’s 2026 benchmark, built on 578,750 creatives, ranks text-only ads and product-image-with-text ads at the top of its asset-type hit-rate list, with high-production assets in the middle of the pack (Motion, Top asset types). Motion’s suggested role for high production is to upgrade angles that have already proven themselves.
That’s good news for smaller teams. Cheap, clear, genuinely different ads are a better way to explore than one expensive shoot.
How many ads, and how often
Meta’s November 2024 guidance gives concrete numbers for Advantage+ shopping campaigns: launch new ads every week, maintain “at least 20 diversified ads,” and dedicate “20-30% of your overall ad budget towards testing new ads.” It also recommends a weekly meeting where the ad buyer and the creative team review what’s working and write the next brief (Meta).
There’s a counterweight for smaller accounts. Meta’s learning-phase guidance warns against high ad volumes, because “the delivery system learns less about each ad and ad set” when you create many of them (Meta, About the learning phase). If your budget is modest, a handful of truly different concepts that each get enough spend will teach you more than twenty that each get a trickle.
A one-hour diversity audit
Export your live ads into a spreadsheet and tag each one. Be strict.
- Person: who is the ad clearly for? If you can’t say, write “everyone.”
- Reason to buy: saving money, status or identity, solving a problem, social proof, curiosity.
- Format: UGC video, product static, comparison, organic-looking static, demo, founder, offer.
- Visual world: describe it in three words.
- Spend share over the last 14 days.
Then look for clusters. If most of your ads share the same person and the same reason to buy, you have a diversity problem, whatever the file count says. If spend has piled onto one cluster, that’s where fatigue will hit first.
The gaps are your next brief. Missing a persona that shows up in your reviews? Missing a format entirely? Those are the concepts to make next.
Checklist
- Count ideas, not ads: group live ads that share person, reason and look
- Every new concept changes at least two levers: person, reason, format, visual world, proof
- Keep iteration for concepts that have already shown signal
- Put a fixed share of budget on new ads each week
- On small budgets, fewer concepts with enough spend each
- Use your reviews to find personas your ads don’t speak to yet
Sources
- Meta for Business, Demystifying Creative Diversification (December 2025)
- Engineering at Meta, Meta Andromeda: Supercharging Advantage+ automation (December 2024)
- Meta for Business, Three steps to optimize your performance with creative diversification (November 2024)
- Meta Business Help Center, About the learning phase
- Motion, Top asset types on Meta in 2026