Is Meta's AI Killing Your Ad Targeting?

Your audience targeting just died. Meta buried it, and the system that replaced it reads your ad creative before it ever looks at your targeting settings.

Meta calls it Andromeda. It is the AI retrieval system that runs the first stage of ad delivery: before the auction, before ranking, Andromeda narrows tens of millions of eligible ads down to what Meta describes as "a few thousand candidates" per user. It analyzes your creative first and predicts who will engage with it. Your interest targets, lookalikes, and audience definitions are now suggestions the AI may or may not use. In practical terms, creative is the targeting.

What happened to advertiser results during the rollout?

The independent numbers are sobering. Confect studied 3,014 e-commerce advertisers representing $834 million in ad spend through the Andromeda rollout and found average ROAS fell 7%, from 9.0 to 8.4. The pattern inside that average is the real story: the accounts hit hardest were the ones that had been winning on tight manual targeting. If your edge was a carefully segmented audience structure, Andromeda took it away.

The honest caveats: this is one vendor's study, not a controlled experiment, and correlation with a rollout is not the same as proof of causation. Meta's own claims about Andromeda are framed the other way: the company says the system improved retrieval recall by 6% and ad quality by 8% on selected segments. Both figures describe Meta's internal measures, not advertiser revenue. The gap between "the AI retrieves better candidates" and "your ROAS went up" is exactly where the work now sits.

How do you win when creative is the targeting?

Andromeda reads creative first, which makes creative diversity the lever. Media buyers have a name for the mechanism: what they call an Entity ID. Ads that look alike share one retrieval ticket, so 30 near-identical variations of the same concept count as one candidate, not 30. Meta itself shipped a Creative Diversity rating column in Ads Manager, which is the company's own confirmation that variety matters more than volume.

What this changes in practice: test genuinely different concepts, hooks, formats, and angles instead of 20 reskins of one winner. And give the system room to work: broad targeting with diverse creative now outperforms tight audiences with one hero ad. If you need to sanity-check what a real testing program costs before you commit, the Meta Creative Test Planner lays out the budget, timeline, and kill rules. And if your results already sagged, the Meta Conversion Optimizer helps diagnose whether creative diversity is the hole in your campaigns.

The Advantage+ era was the warning. Andromeda is the machine reading your creative and deciding who sees it. Advertisers who feed it variety will win the retrieval; advertisers who feed it the same ad 30 times will pay for one ticket and get one chance. The same lesson applies on the placement side: the machine decides more than ever, so your job is to give it good options, not to out-guess it.

How we can help

As your partner, we run paid social programs built for this exact reality. Our Meta management builds structured creative testing around diverse concepts, not reskins, so every test gives Andromeda something new to retrieve. Our audience strategy work reframes targeting as fuel for the AI rather than a gate around it. And our performance analysis keeps the scoreboard on revenue and ROAS, not platform vanity metrics.

What would you change first: your creative testing, or your audience strategy?

Data over opinions.