Meta Andromeda needs 15 creative concepts. Here's how to research them.

Every Andromeda article tells you that you need 8-15 diverse creatives per ad set, but almost none tell you where those 15 ideas are supposed to come from. This one does.

Andromeda punishes sameness, not volume

Meta rolled out Andromeda and the old playbook broke in one update: find one winner, clone it six times, change the thumbnail, let Advantage+ sort it out. That's dead now.

Andromeda's ranking model punishes creative repetition inside the same ad set. Ads that look alike get treated as near-duplicates, and spend collapses into two of them while the rest starve.

So every agency deck says the same thing: ship 15 diverse concepts. The guides stop there, never showing you how to find 15 ideas that have already been validated by a competitor's ad budget.

The benchmark working right now is 8 distinct concepts, 2 variations each, across 3 formats (static, short-form UGC, longer VSL). That's around 15 assets, but only 8 ideas at the conceptual level.

Eight real concepts is harder than it sounds. Sit a team in a room and ask for 15 different angles on the same product, and by angle 6 you're describing the same benefit with a different background color.

The fix is to stop inventing and start copying from operators who've already paid to test.

Guessing is what used to work

When CPMs were lower and the auction was slower, you could afford to guess: make six creatives, run them for a week, kill four, scale two. A bad test cost $200.

That math is broken now. A single weak concept eats around $500 in learning spend before Meta decides it's dead, and six weak concepts out of eight burns $3,000 getting back to the two you should have led with.

Andromeda scores the concept, not the variation. Swapping thumbnails doesn't save you because if the angle is wrong, every variation is wrong.

So the real question isn't "how do I make 15 creatives faster." It's "how do I walk into production week already knowing which 15 angles are working in my niche right now."

The answer is sitting in your competitors' ad accounts.

Step 1: find the brands that are actually scaling

Not every competitor is worth copying, as most brands running ads are losing money. You want to learn from the ones the algorithm already validated.

Open Brandsearch Discovery, set platform to Meta, format to video, and apply Phase: Winning. Add Running Days: 25+ on top.

Phase: Winning means the ad made it through Meta's delivery test, and 25+ running days means the brand kept paying for it. That's the only proof of profitability that matters.

Layer your niche. You want 10-15 brands in your shortlist, not 100.

Save them to a swipe file folder called "Andromeda research - [niche]." Everything you pull from these brands has already passed the Meta delivery test, so you're not brainstorming but reading off an answer key someone else paid to write.

Step 2: open each brand's Creative Tests tab

Scrolling Meta Ad Library gives you 200 ads per brand and no way to tell which ones matter. Nine out of ten never spent a dollar.

Click into any brand from your shortlist and open Brandsearch Brand Analysis. Go to the Creative Tests tab.

Creative Tests clusters every ad the brand has run, groups them by visual and copy similarity, and ranks each cluster by run length and reach. You see the brand's actual portfolio of tested ideas, not a chronological upload feed.

For a brand running 120 active ads, you'll typically see 6-10 clusters. Each cluster is one concept the brand bet money on.

Write down three things per cluster:

1. The angle: what problem or desire is this concept hitting?
2. The format: UGC? Static comparison? Voice-over demo? Split-screen?
3. The verdict: is the cluster scaling or dying?

A dying cluster is just as useful: it tells you what the brand tried and walked away from. You just saved yourself the same test.

Do this for all 10-15 brands and you'll end up with roughly 60-100 clusters. Maybe 40 will be winners, and that's your raw material.

Step 3: pull the hook structures from the Scripts tab

Concepts aren't enough because Andromeda is a hook-obsessed algorithm. The first two seconds of a video decide whether the rest gets served.

Stay inside Brand Analysis and switch to the Scripts tab. This shows the full AI transcript of every video ad the brand has run, with the opening line isolated at the top.

For each winning cluster from step two, pull the top 3 hooks. That's your hook library.

You're looking for the structure, not the exact words. A hook like "I tried 14 whey proteins this year, here's the only one I'm still drinking" is a pattern (specificity plus personal test plus contrarian pick) that you can rebuild for your product.

After an hour in the Scripts tab across your shortlist, you'll have 30-50 hook structures. Group them into families: pain-point opener, numerical claim, founder story, anti-category rant, before/after reveal.

Pick the top 15 structures across families. That's your brief.

What the brief looks like when you're done

You walk into production week with eight distinct creative concepts, each pulled from a cluster that's actively scaling in a competitor's account. You also have fifteen hook structures grouped by family, each tied to a real ad you can show the editor, plus a list of 3 formats the clusters most commonly used.

And a kill list of dying clusters you now know not to chase.

Compare that to the brief you had last year: two good ideas, four guesses, and a moodboard. Andromeda will eat that alive.

Why research beats brainstorming in the Andromeda era

One brand I work with does around $800k/month in supplements. Before this workflow, their production week was a Monday brainstorm, Tuesday-Thursday shoots, and six ads uploaded to Ads Manager on Friday.

Their hit rate was roughly 1 in 6: one concept would scale while five would die in learning.

After one quarter on this workflow, the hit rate moved to around 3 in 8. Same production budget, same editor: the only change was where the ideas came from.

Every concept in the brief has evidence attached: the brand, the cluster, the run length. The editor gets a reference ad for each idea instead of a written description, and the kill list cuts the 10 ideas that look fine on paper but already failed in someone else's account.

Production stops debating creative and starts executing it.

Summary

Andromeda made creative diversity the new scaling lever. Nobody scales on 15 guessed concepts; they scale on 15 concepts already validated inside someone else's ad account.

The workflow:

1. Find the scaling brands in Brandsearch Discovery. Filter to Meta video, Phase: Winning, Running Days: 25+, and your niche. Save 10-15 brands.
2. Map the portfolio in Brandsearch Brand Analysis, Creative Tests tab. Read the clusters. Mark the winners and the dying ones.
3. Pull the hooks from the Scripts tab. Top 3 hooks per winning cluster. Group into families.
4. Build the brief. Eight concepts, 15 hook structures, a kill list of angles not to touch.
5. Repeat next month. The list of scaling brands changes. Re-run the workflow and your brief stays current.

Stop brainstorming your way into Andromeda. Start reading off the answer key your competitors paid to write.