For years, DTC media buyers spent their hours inside the audience panel: stacking interests, building lookalikes, carving out exclusions. In 2026 that work has largely moved into the ad itself. Creative is the new targeting — Meta's delivery system now reads your image, video, and copy to decide who sees the ad, so the asset you upload does the job that manual audience selection used to do. This post explains why that shift happened, how Meta's Andromeda algorithm processes your creative, and what it means for how a DTC brand should plan, produce, and test ads.
Why is creative the new targeting in 2026?
The short version: Meta moved the decision about who sees an ad out of the buyer's hands and into the model, then made the creative the main thing the model reads. Three changes drove this.
- Broad targeting became the default. Advantage+ audience and Advantage+ shopping campaigns ask you for a suggestion, not a hard boundary. The system tests well beyond whatever audience you name.
- Signal loss made manual targeting weaker. Post-ATT, the granular interest and behavior data that powered tight audiences degraded. The model leans on first-party conversion signal and on what it can infer from the creative instead.
- The ranking stack got better at reading creatives. Andromeda, Meta's retrieval and ranking model, can parse what is actually in a video or image and match it to users at the moment of the auction.
So the lever that used to live in the audience builder now lives in the asset. If you want to reach a different person, you change the creative, not the targeting spec.
How does Meta's Andromeda algorithm read your creative?
Andromeda is the retrieval layer that sits early in Meta's ad delivery pipeline. Its job is to scan a very large pool of candidate ads and pull the handful most likely to perform for a given user, before the final auction ranks them. To do that well it has to understand each ad's content, not just its metadata.
In practice the Andromeda algorithm in Meta ads uses computer vision and language models to extract signals from the asset itself:
- Visual content — what objects, scenes, faces, and on-screen text appear, via computer vision ad matching that reads frames the way a human reviewer would.
- Format and structure — static vs. video, aspect ratio, pacing, where the hook lands in the first seconds.
- Copy and concept — the headline, primary text, and the implied promise or persona the ad speaks to.
From those signals the model forms a hypothesis about who the ad fits. A clip that opens on a kitchen counter with a fast benefit-led hook gets matched to different people than a calm testimonial filmed in soft light, even if both sit in the same campaign with the same audience setting. The creative is the targeting instruction.
Does audience targeting still matter under Advantage+?
It matters less than it did, and in a narrower way. With Advantage+ audience in 2026, the audience field you fill in is read as a starting suggestion. The system honors it early, then expands once it finds pockets of performance the creative reaches on its own. So your targeting inputs still shape the first phase of learning, but they no longer cap who the ad can find.
Where manual control still earns its keep:
- Exclusions — suppressing existing customers or recent purchasers is still worth doing by hand.
- Hard constraints — geography, age and legal limits, and language remain real boundaries you set.
- Seeding — a sensible audience suggestion helps the model start in roughly the right place instead of cold.
What no longer pays off is spending an afternoon hand-tuning interest stacks. That time is better spent producing the next batch of creative, because that is the input the model is actually optimizing against.
How much does creative actually affect ROAS now?
Enough that most account-level performance differences between similar advertisers trace back to creative rather than to bidding or structure. When the delivery system is doing the targeting, the only large lever a buyer still pulls is the asset. Two brands with the same budget, the same offer, and the same Advantage+ setup will diverge mostly on the quality and variety of what they put in.
This is why a single winning ad does not hold its lead for long. The model finds the audience that creative fits, saturates it, and performance decays — the familiar fatigue curve. If you want a sober read on how fast that happens and when to refresh, see reading the ad fatigue curve. The practical takeaway is that creative is not a one-time deliverable; it is the thing you keep feeding so the system always has fresh material to match against new pockets of demand.
If creative drives delivery, why is volume the moat?
Because you cannot predict which asset the model will reward. When Andromeda decides who sees what, your job shifts from picking the winner to giving the system enough distinct candidates that it can find one. A brand shipping three ads a month gives the model three guesses. A brand shipping thirty gives it thirty, across more hooks, formats, and personas — and far more chances to land on a combination that opens a new audience.
That turns output rate into the durable advantage. The brands that win under creative-led delivery are not the ones with the single best ad; they are the ones that can sustain a high rate of genuinely different ads without burning out a design team. For a sizing exercise on weekly cadence by spend level, see how many ad creatives per week a DTC brand needs, and for the production side, how to scale ad creative production with AI.
How do you feed the algorithm without going generic?
The trap with high-volume creative is that more ads becomes more of the same ad — minor color swaps the model reads as one concept. Volume only helps if the variants are genuinely distinct in the dimensions Andromeda actually reads: hook, format, persona, and proof. A useful loop looks like this:
- Watch — pull what is live in your category from the Meta Ad Library and tag the creative DNA: which hooks, formats, and angles competitors are actually running.
- Create — turn those tagged signals into briefs and render distinct on-brand variants, not reskins of one layout. The point is to give the model real diversity to match against.
- Ship — push the batch into Ads Manager and let Advantage+ distribute it.
- Learn — read which concepts the system found audiences for, and feed that back into the next briefs.
Keeping the variants on-brand while still distinct is the hard part; turning competitor signal into winning creative walks through how to mine the library without copying it. The discipline is to vary the things the algorithm reads, not just the things a human notices.
This is the loop Uboros runs end to end. It scrapes competitor ads from the Meta Ad Library, tags their creative DNA, drafts briefs, renders on-brand static and video variants, and ships them to Meta Ads Manager — then learns from performance to shape the next batch. In a world where creative is the new targeting, the constraint is how fast you can produce distinct, on-brand assets for the model to match, and that is the constraint Uboros is built to remove. See how the Watch → Create → Ship → Learn loop turns competitor signal into a steady supply of creative.