Meta's Andromeda Update: Why Your Targeting Stopped Working and Creative Became the Algorithm

Interest stacks and lookalike layers no longer move performance the way they used to. Here's what Andromeda changed inside Meta's ad delivery, and how to restructure around it.

What is Meta's Andromeda update?

Andromeda is Meta's ads retrieval system, rolled out across most objectives and placements by October 2025. It shortlists which ads are eligible for a user by reading the ad creative itself using computer vision and natural language processing, rather than relying primarily on targeting settings. Meta reports a 6 percent recall improvement at the retrieval stage from it.

Why did my Meta interest targeting stop working?

Because retrieval is now creative-led. Andromeda infers who an ad is for from its image and copy, so narrowing the audience adds constraint without adding signal. The creative has become the primary targeting instrument.

Should I still use lookalike audiences on Meta?

Rarely for prospecting, where broad targeting plus strong creative generally performs as well or better. Lookalikes and manual audiences still make sense for retargeting known customers, very small budgets, and genuinely niche markets.

How many ad sets should a Meta campaign have?

Few enough that each can reach roughly fifty conversions per week. Budget split across many small ad sets starves each one of the volume needed to exit the learning phase, so consolidation usually outperforms fragmentation.

Did Meta change conversion attribution?

Yes. In an announcement dated 3 March 2026, click-through began counting only link clicks, with other interactions moved to a one-day engage-through column, and the video engagement threshold dropped from ten seconds to five. Reported conversions can shift because of this without any change in actual performance.