Meta Andromeda is the name of the retrieval system inside Meta's ads engine that decides which ads are even considered for a given person before the final ranking happens. It matters to advertisers because it changed what the system rewards: instead of one polished ad shown to everyone, Andromeda does better when it has many genuinely different ads to match to many different people. For media buyers and creative strategists in 2026, that means volume and variety of creative are no longer a nice-to-have. They are how you get delivery at all.
Quick answer: Andromeda is Meta's ads retrieval system, the stage that picks candidate ads for each person from a huge pool before ranking. It performs best when an account supplies many distinct creatives, each suited to a different slice of the audience, so it rewards creative diversity. Practically: run more ads, make them genuinely different in angle, format and messenger, consolidate campaigns, keep targeting broad, and judge ads on their own results rather than forcing even spend.

What Is Meta Andromeda?
Every time someone opens Instagram or Facebook, Meta has a fraction of a second to pick an ad from millions. It does this in stages. Retrieval narrows the millions to a few thousand candidates that might fit this person. Ranking then scores those candidates in detail and picks the winner. Andromeda is Meta's upgraded retrieval stage, built on much larger models and faster hardware, so it can consider far more candidates and match them more precisely to each person's signals.
The practical consequence is that retrieval is now much better at personalization. It is looking for the specific ad that fits this specific person, not the ad that performs best on average. If your account only offers one or two creatives, there is nothing to personalize with. If it offers twenty genuinely different ones, Andromeda can match the skeptic's ad to skeptics and the price-sensitive ad to bargain hunters. Meta describes the system in general terms; the account-level behavior buyers observe is that diversity of creative is rewarded with more and cheaper delivery.
- Retrieval picks candidates. Ranking picks the winner. Andromeda is retrieval.
- It personalizes at the creative level, not just the audience level.
- It can only choose among what you give it. Variety is the input it needs.
- Near-duplicate ads are treated as roughly the same candidate, so they add little.
Why Creative Diversity Matters More Under Andromeda
Before Andromeda, a common approach was to find one winning ad and pour budget into it, sometimes with a few near-identical variants. That still works in the sense that a great ad is still a great ad. But it leaves most of the audience unmatched, because one message cannot be the best fit for a parent, a college student and a retiree at once. Under Andromeda, those unmatched people are reachable, but only if there is an ad in your account that fits them.
Diversity also changes the fatigue picture. When one ad carries the account, its audience slice gets exhausted and performance drops sharply. When twenty ads each carry a slice, no single one is overexposed and the account's aggregate performance is smoother. Buyers who moved from a handful of ads to a deep, varied bench in 2025 and 2026 generally report more stable CPAs and fewer sudden cliffs, though the specific numbers vary by account and nobody outside Meta has a precise formula.
| Approach | Pre-Andromeda logic | Under Andromeda |
|---|---|---|
| One hero ad, scale hard | Worked until fatigue | Leaves most of the audience unmatched |
| Many variations of the hero | Marginal gains | Treated as near-duplicates, little added reach |
| Many distinct angles and formats | Hard to manage | What the system is built to use |
| Narrow interest targeting | Found the audience for you | Limits the matching the system can do on its own |
| Many small campaigns | Control | Splits signal; consolidation usually helps |
What Creative Diversity Actually Means
Diversity is not twenty crops of one video. It means ads that differ in who they speak to, what problem they lead with, what proof they use, who delivers the message and how it looks. A useful mental model is that each ad should be the obvious best fit for some identifiable person. If you cannot name who a given ad is for and why it is the one they would stop for, it is probably a variation, not a new candidate.
The six dimensions below are where diversity comes from. A batch that varies on at least three of them is diverse. A batch that varies on one is a set of variations. For a deeper practitioner walkthrough of what creative diversity looks like in a live account, this video is a good explainer.
| Dimension | Low diversity | High diversity |
|---|---|---|
| Audience or persona | Everyone | New parent, shift worker, gift buyer, skeptic |
| Problem or angle | One benefit | Four different triggers from reviews |
| Proof | Product shot | Demo, review count, founder story, comparison, guarantee |
| Messenger | Brand voice only | Creator, founder, expert, customer, no person |
| Format | All 4:5 statics | UGC video, static, text-forward, carousel, short motion |
| Visual style | Brand template | Lo-fi phone footage, studio, meme-style, screenshot, editorial |
How Account Structure Changes Under Andromeda
Structure follows the same logic: give the system room to match. That usually means fewer campaigns with more ads inside them, broad targeting rather than stacked interests, and letting delivery concentrate where it wants instead of forcing even spend across ads. Many accounts now run one Advantage+ sales campaign carrying the proven bench, plus one testing campaign that introduces new concepts each week and feeds winners into the main campaign.
Uneven spend inside an ad set is not a bug to fix. If the system puts most spend on three of ten ads, those three are the ones it found good matches for. The other seven may still be valuable because they get delivery in pockets, or they may be genuine losers. Read them on their own metrics at their own spend level rather than killing them for being under-delivered, and give new ads a fair run in the testing campaign where they are not competing with incumbents.
- Consolidate: one main sales campaign with 10 to 20 live ads, broad targeting.
- Add a weekly testing campaign with one ad set and 4 to 8 new concepts.
- Promote winners from testing into the main campaign; do not just raise the test budget.
- Keep retargeting small and separate, since its audience is already matched.
- Resist duplicating ads across many ad sets; it fragments signal.
- Review the bench monthly: retire ads with falling leading metrics, keep diversity across the six dimensions.
A Creative Diversity Checklist for Your Next Batch
Run this before you launch a batch. If more than two answers are no, the batch is variations, not diversity, and the system will treat much of it as one ad.
- Can you name a different person each ad is the best fit for?
- Do at least three ads lead with different problems or triggers?
- Are there at least three proof types across the batch?
- Is there at least one ad with a person and one without?
- Are there at least two formats, for example video plus static?
- Does at least one ad look nothing like your brand template?
- Is at least one ad a long shot aimed at a persona you have never addressed?
- Do no two ads share the same first three seconds or the same headline?
Keep a running count of your live bench against the six dimensions. Most accounts find they are heavy on one or two, usually brand-template statics with product-shot proof, and light on messenger and visual-style variety. Filling those gaps is often the cheapest performance gain available.
Common Mistakes Advertisers Make Post-Andromeda
The most common mistake is hearing "volume" and uploading fifty near-identical ads. The system collapses them into a few candidates and you have spent production time for nothing. The second is keeping dozens of small campaigns and ad sets, each with its own tiny budget, which fragments learning. The third is forcing even spend, either by splitting ads into separate ad sets or by pausing the ads the system favours. The fourth is dropping the testing discipline because "Andromeda will figure it out"; it can only figure out what you give it.
| Mistake | Why it hurts | Fix |
|---|---|---|
| Fifty variations of one ad | Collapsed into a few candidates, wasted production | Fewer ads, each a distinct angle, format or messenger |
| Many small campaigns and ad sets | Fragmented signal, slow learning | Consolidate into one main campaign plus one testing campaign |
| Forcing even spend | Overrides the matching that is the whole point | Let spend concentrate; judge ads on their own metrics |
| Narrow interest stacks | Limits the matching the system can do | Go broad; let creative do the targeting |
| No weekly testing | Bench goes stale, diversity decays | Fixed weekly batch of 4 to 8 new concepts |
| Killing low-spend ads | Pocket performers get lost | Read them at their spend level before pausing |
How Loraloop Fits
The demand Andromeda creates is for more distinct creative every week, which is a production problem for most small teams. Loraloop's ads agent, Angie, generates ad creatives and copy from your Brand DNA at 10 credits per ad, so a brand on the Starter plan can draft roughly 30 ads a month and a Pro plan around 100, enough to keep a varied bench supplied. Because Brand DNA holds your audience segments and positioning, you can ask for ads aimed at different personas and proof types rather than variations of one. Angie also drafts the campaign and runs daily optimisation such as shifting budget to winners and pausing losers, with every launch and budget change waiting for your approval.
Frequently Asked Questions
What is Meta Andromeda and how does it affect ads?
Andromeda is the retrieval stage of Meta's ads system, the part that selects candidate ads for each person from millions before final ranking. It is built to personalize at the creative level, so it works best when an account supplies many genuinely different ads. For advertisers, it means creative diversity and volume drive delivery and cost more than they used to.
How many ads should I run under Meta Andromeda?
Most buyers keep 10 to 20 live ads in the main sales campaign and add 4 to 8 new concepts a week through a testing campaign. The number matters less than the variety: twenty near-duplicates behave like two ads. Aim for ads that differ in audience, problem, proof, messenger, format and visual style.
Does meta andromeda ads retrieval punish duplicate creatives?
Not as a penalty, but near-duplicate ads are treated as roughly the same candidate during retrieval, so they add little reach and compete with each other. Production time spent on minor variations is mostly wasted. Spend it on distinct angles and formats instead, and iterate winners with new hooks rather than new crops.
Should I still use interest targeting after Andromeda?
Broad targeting is the default for most accounts now. Andromeda matches at the creative level, so narrow interest stacks mostly limit what it can do. Keep targeting broad and let the diversity of your ads do the segmenting. Retargeting stays useful but should be small and separate.
Why is my spend concentrated on a few ads in one ad set?
Because the system found strong matches for those ads and is sending them the impressions they win. That is expected behavior, not a fault. Judge the under-delivered ads on their own metrics at their own spend level, and give new concepts a fair read in a separate testing campaign before deciding they are losers.
Feed the algorithm the variety it is asking for. Loraloop drafts distinct ads for different personas and proof points from your brand knowledge, and you approve each one before it spends.
Try Loraloop Free