Creative Is the New Targeting: How to Build Ads for Andromeda
For most of the last decade, a good Meta media buyer earned their fee in the ad set. Interest stacks, lookalike percentages, exclusions, layered demographics. Creative mattered, but targeting decided who got the chance to see it.
That order has flipped. Meta Andromeda, the retrieval engine that now sits at the front of ad delivery on Facebook and Instagram, reads your ad itself to work out who it should go to. The visual, the hook, the problem it names, the person it shows. Those are now the strongest signals you send.
So when an account plateaus in 2026, the fix is rarely a smarter audience. It is usually a creative library that is too narrow for the system to work with. This guide covers what Andromeda actually does, why creative has become your targeting, and how to build, structure and test ads that the system can genuinely use.
If you want the short version: stop making more versions of the same ad and start making different ads.
What Meta Andromeda actually is
Andromeda is Meta’s personalised ads retrieval engine. Meta announced it on its engineering blog in December 2024, and by late 2025 it was running globally across Facebook and Instagram.
To understand why it matters, you need to know where it sits. Every time someone opens their feed, Meta works through roughly three stages:
- Retrieval. Tens of millions of eligible ads get narrowed to a few thousand candidates for that one person. This is Andromeda’s job.
- Ranking. Those candidates are scored on how likely this person is to click, convert or engage.
- Auction. The ranked ads compete on bid and predicted value, and one wins the slot.
Here is the part most advertisers miss. If your ad does not make it through retrieval, your bid never gets a chance to matter. Meta says retrieval handles about three orders of magnitude more ads than the later stages, which makes it the real gatekeeper.
The old retrieval system was a patchwork of smaller models and rule-based filters, and it leaned heavily on the targeting you set. Andromeda replaced that with a single deep neural network that Meta says supports a 10,000x jump in model complexity. Meta reports a 6% improvement in retrieval recall and an 8% lift in ad quality on selected segments.
Those are modest-sounding numbers at Meta’s scale, but the bigger change is philosophical. The system now looks closely at what your ad is and matches it to people who are likely to respond to that specific message.
Why creative is now your targeting
Andromeda groups ads by meaning. It reads the visuals, the copy, the format and the angle, then places similar ads close together in its index. When it retrieves candidates for a person, it pulls from the clusters that sit closest to that person’s behaviour.
That has one huge consequence. Ads that look and say the same thing are treated as roughly the same ad.
Picture a skincare brand running 30 ads. Five headline variants, three background colours, two CTAs, all built on one product shot and one claim about hydration. In the old world that looked like a rich test. To Andromeda it looks like one idea wearing 30 outfits, so it earns something close to one retrieval opportunity.
Now picture the same brand running eight ads that are genuinely different:
- A dermatologist explaining the ingredient science
- A 45-year-old customer talking about fine lines
- A gym-goer showing it under sweat
- A price comparison against a premium competitor
- A 7-day before-and-after
- A founder story about why the brand exists
- A routine video for oily skin
- A gifting angle for the holidays
Each of those speaks to a different person with a different reason to buy. Andromeda can place each one in a different part of its index and match it to a different slice of the market. That is targeting, done by the creative.
This is why broad audiences now tend to beat tightly built ones on Meta. When you narrow the audience manually, you limit how far Andromeda can carry each concept. When you go broad and supply real variety, the system finds the buyers for you. We covered the wider shift in where your budget wins between Google Ads and Meta Ads, and this is a big part of why Meta now behaves so differently from search.
How to build creative for Andromeda
The goal is not volume. It is distance between ideas. A useful test: if you put two ads side by side, would a stranger say they are selling to the same person for the same reason? If yes, they are one concept.
We plan creative across six dimensions. Change at least two of them between concepts and you are usually creating something Andromeda will treat as new.
| Dimension | What you change | Example for a meal kit brand |
|---|---|---|
| Persona | Who the ad is clearly for | Busy parents, gym regulars, couples who hate washing up |
| Motivation | The reason to buy | Save time, eat healthier, cut takeaway spend |
| Awareness stage | How much they already know | Problem-unaware hook vs a direct offer for people comparing brands |
| Messenger | Who is speaking | Founder, chef, real customer, creator, no person at all |
| Format | How it is built | UGC video, static, carousel, talking head, product demo |
| Visual world | Setting, style, pacing | Messy family kitchen vs clean studio vs on the go |
Start with angles, not assets
Most teams brief creative by format. “We need three videos and four statics this week.” That produces assets that look different but say the same thing.
Brief by angle instead. Pull your angles from real sources: customer reviews, sales call objections, support tickets, competitor comments, search queries. Every distinct reason someone buys or hesitates is a potential concept. Write each as one line, for example “For parents who feel guilty about another frozen pizza night.” Then decide the best format for that angle.
Build in concepts, then iterate within them
Think of your library in two layers:
- Concepts. A new angle, persona or motivation. These are what earn new retrieval entries. Aim for 3 to 6 new concepts a month for a mid-sized account, more if you spend heavily.
- Iterations. New hooks, edits, thumbnails or copy on a concept that is already working. These extend a winner’s life but do not open new audiences.
The common mistake is spending 90% of production on iterations. Flip that ratio early on until you find two or three concepts that carry the account, then iterate harder on those.
Make the first three seconds do the targeting
The hook is where the persona and the problem should be obvious. If the opening frame shows a new mum at 6am with a crying baby, Andromeda and the viewer both know who this is for. A generic logo intro tells neither of them anything.
Use AI to scale production, not to replace ideas
Meta’s Advantage+ creative tools can generate background swaps, text variations and image expansions. They are useful for format coverage across placements. They are not a source of new concepts, because they mostly produce close variations of what you already have. Our AI marketing team uses these tools for speed, but the angles still come from humans who understand the customer.
Account structure that suits Andromeda
Great creative still underperforms inside a fragmented account. Andromeda learns faster when data is concentrated, so the structure that worked in 2021 now actively works against you.
Consolidate hard
If you are running separate ad sets for every interest, age band and lookalike, you are splitting your learning across groups that the system would rather treat as one. For most brands spending under roughly $50,000 a month, a simple setup does the job:
- One main scaling campaign using Advantage+ sales or a broad manual campaign with Advantage+ audience switched on
- One testing campaign for new concepts
- One retention or retargeting layer only if your data shows it adds incremental sales, not just cheap attributed ones
Broad targeting with clean exclusions (existing customers if you are prospecting, for example) is the default. Age, gender and location only where the business genuinely requires it.
Give the system enough room
In the scaling ad set, run roughly 8 to 15 live ads across several genuinely different concepts. Fewer than that and Andromeda has little to choose from. Far more than that, especially if they are near-duplicates, and spend spreads too thin for anything to learn.
Test concepts, not tweaks
Use the testing campaign to answer one question: does this new concept find buyers? A practical approach:
- Put each new concept in its own ad set, with two or three executions of it
- Give each ad set enough budget to reach meaningful conversion volume within 5 to 7 days
- Judge on cost per acquisition and hook rate, not on CTR alone
- Move winners into the scaling campaign as new ads and leave them alone to learn
Do not duplicate a winning ad into ten ad sets to “scale” it. Under Andromeda that mostly makes your ads compete with each other.
Feed it clean conversion data
Creative tells Andromeda who might care. Conversion data tells it who actually bought. If your pixel and Conversions API are patchy, the system is learning from guesswork. Make sure server-side events are deduplicated and that purchase values are passed correctly. Our paid marketing team treats tracking as step one in any Meta audit, before a single creative is changed.
Mistakes we see in almost every account audit
Mistaking variations for diversity. Forty ads built from four ideas is still four ideas. Count concepts, not ads.
Killing new concepts too early. A fresh angle often looks expensive for its first few days while the system works out who it is for. Set a decision threshold before launch, such as a set spend or number of conversions, and stick to it.
Refreshing by recycling. Pausing an ad and relaunching it, or changing a button colour, is not a refresh. When frequency climbs above 3 to 4 and cost per result rises, the answer is a new concept.
Over-polishing. Studio-perfect ads often lose to rougher, native-feeling content because they look like ads. Production budget is better spent on more distinct ideas than on more gloss per idea.
Reading the wrong numbers. Cheap CTR from a clickbait hook can drag your cost per purchase up. Judge creative on the metric that pays the bills.
How to measure creative under Andromeda
Track these at concept level, not just ad level:
| Metric | What it tells you | Rough signal to watch |
|---|---|---|
| Hook rate (3-second views ÷ impressions) | Does the opening stop the right people | Under 25% usually means a weak first frame |
| Hold rate (ThruPlays ÷ 3-second views) | Does the message keep them | A big drop after the hook means a mismatch |
| Cost per acquisition | Does the concept find buyers | Compare against your target, not the account average |
| Spend share | Is Andromeda choosing it | A concept that gets no spend is not being retrieved |
| Frequency | Is the concept tiring | Rising frequency with rising CPA means refresh |
| New customer rate | Is it reaching new people | Low rates suggest it is mostly hitting existing buyers |
Spend share is the one most teams ignore. If Meta keeps refusing to spend on a concept even with budget available, that is the system telling you it cannot find a home for the idea.
Pair platform data with your own analytics so you are not relying only on Meta’s attribution. Our guide on how to use Google Analytics 4 walks through setting that up. And if acquisition costs are climbing, our breakdown of how to lower cost per lead without cutting ad spend pairs well with everything here.
The brands that win on Meta now are the ones with the most ideas
Andromeda has not made media buying irrelevant. It has moved the skill. Structure, tracking and budget discipline still matter, but the biggest gains now come from understanding your customers well enough to speak to five or ten different versions of them.
That is a strategy problem before it is a production problem. Know your angles, brief by motivation, build concepts that are genuinely far apart, and give the system a clean, consolidated account to work in. Do that consistently and Andromeda stops being a black box and starts doing your prospecting for you.
If your Meta results have stalled since the rollout, the creative library is the first place to look. At Social Schnell we run creative audits that map your current ads into concepts, find the gaps, and build a testing roadmap around them. You can see how we approach social media marketing or book a free audit with our team.
Frequently asked questions
What is Meta Andromeda?
Andromeda is Meta’s AI-powered ad retrieval engine. It narrows tens of millions of eligible ads down to a few thousand candidates for each person before ranking and the auction take over. It uses the content of your ads as a major signal for deciding who should see them.
Does detailed targeting still work after Andromeda?
It still works, but it rarely beats broad targeting with strong, diverse creative. Narrow audiences limit how far the system can carry each concept. Keep targeting restrictions for genuine business needs such as location or legal age limits.
How many ads should I run in an ad set?
For most accounts, 8 to 15 live ads in a scaling ad set is a sensible range, provided they represent several genuinely different concepts. Running 50 near-identical ads does not give the system more to work with.
What counts as creative diversity on Meta?
Diversity means a change in the idea, not the decoration. Different personas, motivations, awareness stages, messengers, formats and visual settings all count. New headlines or background colours on the same concept do not.
How often should I launch new creative?
Most mid-sized accounts benefit from 3 to 6 new concepts a month, with ongoing iterations on proven winners. Refresh when frequency rises above 3 to 4 and cost per result starts climbing.
Is Advantage+ creative enough to create diversity?
No. It is helpful for adapting ads across placements and formats, but it mostly produces variations of an existing ad. New concepts still need human insight into your customers.
