Meta Andromeda is the system that decides which ads are even considered for each person on Facebook and Instagram. Since it arrived, delivery has rewarded accounts that feed it genuinely different creative, so the work of targeting has moved out of ad set settings and into the ads themselves. Here is what changed and what to do.
Most of the confusion about the Andromeda update comes from treating it as a new algorithm that punishes certain tactics. It is better understood as a much larger front door. Meta can now consider far more ads for each impression, which makes the variety of your creative more important than the precision of your audience settings. That has practical consequences for account structure, creative production and how you judge results.
What Meta Andromeda Is, And What It Is Not
Meta’s ad delivery works in stages. Retrieval comes first: out of tens of millions of eligible ads, the system pulls a shortlist of a few thousand candidates for one person in one moment. Ranking and the auction then decide which of those candidates is shown. Andromeda is Meta’s name for the retrieval stage, described in a December 2024 Engineering at Meta post.
According to that post, Andromeda runs on NVIDIA Grace Hopper hardware and Meta’s own training and inference chips, supports a 10,000x increase in model capacity for retrieval, and delivered a 6% improvement in recall and an 8% improvement in ads quality on selected segments. Meta built it because its automation and generative AI tools were multiplying the number of ads in the system faster than the old retrieval stage could handle. The same post says more than a million advertisers used Meta’s generative AI tools to create over 15 million ads in a single month.
What Andromeda is not: a ranking model that picks winners on its own, a penalty for any specific setting, or a reason every account’s results changed. If performance dropped, Andromeda is one possible cause among several, and the others are usually easier to check.
What The Andromeda Update Changed For Advertisers
Because retrieval can now consider far more candidates and match them more finely to each person, the inputs that matter have shifted. Meta’s own guidance, as quoted by Jon Loomer, says “the focus has shifted from niche targeting to creative diversification as the best lever”. Meta also dropped its earlier recommendation of six ads per ad set, which Loomer links to the same change.
| Lever | Before Andromeda | After Andromeda |
|---|---|---|
| Who sees the ad | Set mainly by interests, lookalikes and ad set splits | Increasingly set by what the ad itself shows and says |
| Account structure | Many ad sets, each a targeting hypothesis | Fewer campaigns and broad ad sets with more creative inside |
| Creative testing | Variations of one winning ad | Distinct concepts, formats and personas |
| Number of ads | Two to six per ad set | Enough distinct ideas to cover different buyers |
| Main risk | Audiences too narrow | Creative too similar to give the system real choices |
The cost of getting this wrong has also risen. Meta’s third quarter 2026 results reported ad impressions up 10% and the average price per ad up 14% year over year. When each impression costs more, an ad that the system cannot place well is a more expensive mistake.
Why Near-Identical Creative Now Underperforms
Twenty versions of the same lifestyle image with different headlines look like twenty ads in Ads Manager, but to a system matching ads to people they offer one idea. The person who responds to a sizing explainer is rarely the same person who responds to a founder story or a bundle offer, so an account that only varies headlines limits who it can reach, however many ads it runs. Loomer recommends diversity across concepts and angles, formats and personas rather than small copy or visual tweaks, and notes that volume alone does not guarantee results.
| Dimension | Genuinely different | Not really different |
|---|---|---|
| Concept | Problem solved, desire, objection handled, social proof | The same claim with a new headline |
| Format | Creator video, static, carousel, catalog, founder to camera | The same video in a new crop |
| Persona | First-time buyer, gift buyer, loyal customer, a different age group | The same person with a new background color |
| Hook | Opening on the product, the problem, the result or a person | The same opening with different on-screen text |
Iterations still have a place: once a concept works, new hooks and edits extend its life. The mistake is treating iterations as the testing program.
Not sure whether your account is built for Andromeda? A Meta ads audit checks structure, creative diversity and tracking together. If you would rather talk it through, a free growth audit covers the same ground live in half an hour.
How To Structure An eCommerce Meta Account After Andromeda
For most ecommerce brands the structure gets simpler while the creative work gets deeper. A setup that suits Andromeda looks like this:
| Campaign | Purpose | Judge it on |
|---|---|---|
| Prospecting (Advantage+ sales, broad) | Find new buyers at scale with a deliberate cap on spend to existing customers | New customer CPA and new customer ROAS |
| Creative testing | Introduce new concepts weekly against the current best ads | Cost per purchase against the control, plus hold rate and click rate |
| Existing customers | Launches and replenishment to past buyers | Revenue per returning customer |
| Catalog retargeting (optional) | Viewed and carted products | Incremental revenue, not reported ROAS |
Three rules keep it honest. First, separate new and returning customers in reporting, because automated campaigns find the cheapest conversions and those are often people who would have bought anyway; our guide to NC-ROAS explains the metric. Second, consolidate rather than split: fewer ad sets give each one more data. Third, keep tracking clean, because better retrieval cannot help if Purchase events arrive late, twice or not at all.
This is the structure we run in our Meta ads management, and it is the same thinking behind our guide to Facebook ads for clothing brands.
Creative Volume Without Chaos: A Production System
Andromeda raises the bar on creative output, which is where most teams struggle. The answer is a system rather than a sprint. Start every month with a short list of concepts, each written as one sentence about a buyer: what they want, what worries them, or what they need to believe. Then assign each concept a format and a persona so the batch covers genuinely different ground.
As a working rule, we plan new concepts every week in proportion to spend, and we retire ads on written criteria rather than instinct. Brands with modest budgets can cover more ground by mixing formats: a founder video, a creator review, a product static and a catalog ad can come from one shoot. Generative tools help with variants and translations, but they need rules; our guide to AI UGC ads covers where they help and where they create legal risk. For apparel brands specifically, our breakdown of clothing ads that sell lists twelve formats worth testing.
Production is half the job. The other half is reading results by concept, not by individual ad, so the team learns which ideas work. Our performance creative service runs that testing calendar for brands that want it handled.
What Andromeda Does Not Change
- Your offer and margin. Better delivery cannot fix a product people do not want at your price, or a ROAS target your margin cannot support.
- Tracking. Retrieval and ranking learn from the events you send. Missing or duplicated Purchase events still mislead the system.
- Attribution. Meta still credits conversions by its own window and rules, which is why we judge accounts on store data as well as Ads Manager. Our guide to the Meta attribution window explains the settings.
- The basics of diagnosis. If results fall, check tracking, offer, landing page and creative fatigue before blaming the algorithm. Our five-step diagnosis walks through it in order.
In other words, Andromeda changes how ads find people, not whether the business behind them works. Fix the fundamentals first and the creative work pays off faster.
Signs Your Account Is Fighting Andromeda
Most accounts that struggled after the change show the same symptoms. If three or more of these describe yours, structure and creative are the first places to look, before bids or budgets.
- Spend piles onto one or two ads. The system keeps choosing the same creative because the alternatives are too similar to be worth testing.
- Frequency climbs while reach stalls. The account keeps finding the same people because the creative only speaks to one type of buyer.
- Many small ad sets never exit learning. Budget is spread too thin for any of them to gather enough conversions.
- New ads rarely beat the old winner. Tests are iterations of the control rather than new concepts.
- Reported ROAS is steady but new customer revenue is falling. Delivery has drifted toward past buyers who were going to purchase anyway.
Each symptom points to a specific fix: consolidate ad sets, widen the concept list, separate new and returning customer reporting, and give tests a fair budget against the control.
Want a second opinion before Q1? In a 30-minute growth audit, at no cost, Jason Lu reviews your account structure and creative mix with you and names the first three changes to make.
Meta Andromeda FAQs
What is Meta Andromeda?
Andromeda is Meta’s ad retrieval system, the first stage of ad delivery. It narrows tens of millions of eligible ads to a few thousand candidates for each person before ranking and the auction decide what is shown.
When did the Andromeda update happen?
Meta described Andromeda publicly in December 2024. Advertiser guidance shifted through 2025, when Meta began emphasizing creative diversification over niche targeting and dropped its six-ads-per-ad-set recommendation.
Do I need more ads because of Andromeda?
You need more distinct ideas, not simply more ads. Variety across concept, format, persona and hook gives the system real choices; dozens of near-identical variations do not.
Should I still use interest targeting?
For most ecommerce prospecting, broad targeting with strong creative now performs as well or better. Narrow audiences can still make sense for specific exclusions, launches to past buyers or small markets.
Does Andromeda affect Advantage+ campaigns?
Yes. Andromeda was built partly to handle the volume of ads that Advantage+ automation and generative tools create, and Advantage+ sales campaigns rely on it to match many creatives to many people.
