Skip to content

Meta Andromeda Ad Retrieval System

What it does, where it fits, and what Meta does not publicly document

Last reviewed: 20 August 2026

Andromeda is Meta's personalized-ads retrieval system. It narrows a pool of tens of millions of eligible ads to a few thousand candidates for later recommendation, ranking, and auction stages. It is not the name for Meta's entire advertising algorithm and does not alone choose the final winning ad.

The documented pipeline

  1. 1.Eligibility. Meta first has a large pool of ads eligible for an opportunity based on campaign and delivery constraints.
  2. 2.Retrieval. Andromeda selects a much smaller candidate set that may be relevant to a person and the current context.
  3. 3.Ranking and recommendation. Later models score candidates. Meta's 2026 engineering material describes GEM as a central ads recommendation foundation model; it should not be conflated with the earlier Andromeda retrieval stage.
  4. 4.Auction. Meta says total ad value includes the bid, estimated action rate, and ad quality. Cost or delivery should not be attributed to a single undocumented “Andromeda premium” mechanism.

What the public sources do not establish

  • • An advertiser-visible “Entity ID optimization” score or control
  • • A universal requirement for 15, 25, or 50 creatives per account
  • • A universal weekly refresh rule or 7–14 day algorithmic penalty
  • • A guarantee that broad targeting always beats narrower targeting
  • • A causal rule that creative diversity automatically lowers CPM or CPA

Practical advertiser interpretation

Creative quality and diversity remain useful testing disciplines, but the reason is operational: different concepts create different hypotheses to measure. Do not present that practice as privileged access to Meta's embeddings or as a guaranteed auction advantage.

Choose a cadence from your own objective, budget, audience size, and signal volume. Compare concepts with consistent measurement, monitor account-level fatigue indicators, and treat broad-versus-narrow targeting as an experiment rather than a universal rule.

How AdRiseLab relates

AdRiseLab supports creative research, production, connected-account monitoring, and user-approved publishing for Meta workflows. Its hook, angle, and format-diversity tools are an advertiser testing framework; AdRiseLab does not claim access to undocumented Andromeda internals.

See the Creative Generation workflow →

Primary sources

Frequently Asked Questions

What is Meta Andromeda in simple terms?+
Andromeda is a personalized-ads retrieval system. Meta says it selects a few thousand relevant ad candidates from a pool of tens of millions. Those candidates still pass to later ranking and auction stages; Andromeda is not the entire ad-delivery algorithm.
Does Andromeda decide the final winning ad?+
No. Retrieval narrows the candidate set. Later recommendation or ranking systems score candidates, and Meta's auction evaluates total value using inputs that include bid, estimated action rate, and ad quality. Public documentation does not support describing Andromeda as the sole final decision-maker.
Does Meta document Andromeda Entity ID optimization for advertisers?+
Meta's public Andromeda engineering article does not document an advertiser-facing Entity ID optimization system, a way to inspect internal embeddings, or a guarantee that visually similar ads are collapsed into one auction entry. Treat such claims as unsupported unless Meta publishes evidence.
How many creatives does Andromeda require?+
Meta does not publish one universal number or weekly quota for every advertiser. Creative-testing cadence depends on objective, budget, audience, placement, production cost, and statistical signal. Product-specific examples should not be generalized into a rule for all accounts.
Does Andromeda create a universal 7–14 day fatigue penalty?+
Meta does not publicly document a universal 7–14 day Andromeda penalty. Evaluate creative fatigue from comparable account data, including frequency, CTR, CPM, CPA, ROAS, audience size, placement, and delivery trend.

Related concepts

Creative FatigueCreative DiversificationMeta Ads Auction
CM
Caner Moral

Founder & CEO, AdRiseLab

Performance marketer turned product builder focused on Meta advertising, creative workflows, and measurement. Founded AdRiseLab to reduce the research-to-publish bottleneck in Meta advertising.

XLinkedIn