Meta Andromeda and GEM Ranking Model Explained: The 2026 Playbook for Performance Marketers

Quick answer: Meta Andromeda is a two-stage ML system deployed 2024 to 2026. Andromeda retrieves candidate ads, GEM (Generative Ads Recommendation Model) ranks them. For advertisers, this shifts the playbook from bid tuning to creative volume (20 plus concepts per month), signal quality (CAPI EMQ 8.0 plus), and catalog completeness (500 plus SKUs where relevant).

Meta Andromeda is the two-stage ML retrieval and ranking system that replaced Meta's legacy DLRM ad stack across 2024 to 2026. Andromeda handles retrieval, GEM (Generative Ads Recommendation Model) handles ranking, and together they shift the winning playbook from bid tuning to creative volume, signal quality, and catalog completeness. This guide covers what changed, what to do, and a 30 day rebuild plan.

By Nand Patel, Founder, New Public Media.

What Is Meta Andromeda and How Is It Different From GEM?

Meta Andromeda is the retrieval layer of Meta's ads system, deployed in production during 2024 and scaled across 2025 and 2026. GEM (Generative Ads Recommendation Model) is the ranking layer that sits above Andromeda, published as a Meta engineering paper in November 2025. Together they form a two-stage pipeline: Andromeda scans billions of ads and returns thousands of candidates per impression, then GEM ranks those candidates against user context, creative embeddings, and predicted conversion probability.

Key takeaway: Andromeda finds the ads. GEM chooses the winner. Neither layer cares about your manual bid.

Why Meta Rebuilt the Ranking Stack in 2025 to 2026

Meta rebuilt the stack because the legacy DLRM ad system could no longer absorb the scale of catalog inventory, creative variants, and signal loss from iOS 26. On the Q1 2026 earnings call, Meta credited GEM and Andromeda with a 5 percent conversion lift on Instagram and a 3 percent lift on Facebook Feed. The old system was tuned by hand with heuristics. The new stack is a foundation-scale ML system that ingests billions of signals per day, most of which no human operator could reason about.

iOS 26 signal loss forced the shift. Advantage+ Creative Enhancements generate 5 to 15 variants per source asset, and DTC catalogs cross 500 to 50,000 SKUs. The old stack ranked in milliseconds with compact models. GEM uses transformer-scale ranking with attention over creative embeddings, which is why Meta credits it for the 2026 conversion lift.

Key takeaway: Signal loss did not kill Meta ads. It killed the old ranking model. GEM is the replacement.

Andromeda vs GEM vs the Legacy DLRM Stack: A Comparison

The three stacks are not different Ads Manager UIs. They are different backends Meta has run in sequence.

Key takeaway: Every year you run the account like it is 2022, you leave 20 to 40 percent of Meta's algorithm gains on the table.

What Andromeda and GEM Changed for Advertisers in 2026

Five things shifted materially for advertisers in 2026.

Advantage+ Shopping now beats manual on almost every DTC account

Meta's Q1 2026 earnings deck showed Advantage+ Shopping delivering 17 percent lower CPA than manually structured campaigns on comparable spend. On our client accounts running Advantage+ Shopping with a clean catalog, the same shift shows up as 12 to 22 percent lower CPA.

Catalog Ads print outsized returns

Catalog-heavy accounts (60 to 100 percent of spend in Catalog Ads with a clean feed) see 44 percent higher ROAS and 68 percent lower CPA than accounts running mostly non-catalog creative. GEM ranks catalog impressions per-user per-impression, which no manual DPA setup could match.

Creative volume is the primary lever

Brands running 20 or more new ads per month see 65 percent higher ROAS than brands testing under 10 new ads per month. This is the single biggest shift. Under DLRM, one hero creative could carry an account. Under GEM, the ranker needs a fresh, diverse candidate pool weekly.

Signal quality gets weighted heavier

Server-side signals via CAPI, Enhanced Conversions, and Offline Conversions are weighted higher than browser pixel events. EMQ below 6.0 shrinks attribution. 8.0 plus is the target.

Bids and budgets are near-irrelevant knobs

Manual bid and cost caps constrain the ranker without improving performance. Nearly every account we audit in 2026 performs better on budget-only pacing with Advantage+ Shopping and Advantage+ audiences.

Key takeaway: If you are still tuning bids and rotating one hero video every quarter, GEM is punishing you.

How to Feed Andromeda and GEM: The Four Inputs That Matter

There are four levers advertisers actually control under Andromeda and GEM. Every meaningful account rebuild we run in 2026 revolves around these four.

1. Creative volume. Ship 20 or more new concepts per month. Each variant is a candidate in the GEM ranker. Concept cost typically runs USD 100 to 300 per variant with UGC, USD 300 to 800 for founder-led studio, and near-zero for Advantage+ Creative Enhancement variants generated from a clean source asset. Our Meta Advantage+ Creative Enhancements guide covers the AI variant setup.
2. Signal quality. Deploy Meta Conversions API alongside the browser pixel with dedup keys, Enhanced Conversions where supported, and Offline Conversions for lead-based businesses. EMQ target 8.0 plus. Below 6.0, attribution and optimisation both suffer.
3. Catalog completeness. For DTC ecommerce, a catalog with 500 plus SKUs, complete product images, correct pricing, and populated custom labels is the single biggest catalog-side lever. See our Advantage+ Shopping campaigns guide for the full catalog setup.
4. Audience seeding via first-party data. Upload customer lists, high-value customer segments, and 180 day site visitors as seeds. Advantage+ audience uses these as directional signals, not as hard constraints, and lifts new-customer CPA 10 to 25 percent on our client accounts.

Key takeaway: You do not tune GEM. You feed it. Volume, signal, catalog, and first-party seeds are the four inputs it wants.

Campaign Structure That Works With Andromeda (Not Against It)

The correct 2026 Meta account structure is aggressively consolidated. Two to four ad sets per campaign is the ceiling, not the floor.

The lean five-item structure we deploy on every account:

1. One Advantage+ Shopping campaign for prospecting, budget USD 100 to 250 per day (GBP 80 to 200, AUD 150 to 380) at the entry tier, scaling with performance.
2. One Advantage+ Retention or manually structured retargeting campaign, budget capped at 15 to 25 percent of prospecting spend.
3. One creative testing campaign with a hard daily budget (USD 100 to 250 / GBP 80 to 200 / AUD 150 to 380), one ad set, five to eight ads per launch, running for 5 to 7 days before promotion decisions.
4. Zero manual bid caps or cost caps at launch. Add cost caps only after 90 days of stable data, and only if unit economics genuinely require them.
5. Weekly, not daily, decisions. Under GEM, day-over-day noise is louder than signal for anything below 100+ daily conversions.

For accounts scaling past USD 100K per month, see our full Meta scaling playbook for the additional layers.

Key takeaway: Fewer campaigns, fewer ad sets, more ads. That is the entire 2026 structural shift.

Creative Volume, Diversity, and the New 20 Per Month Floor

Creative volume compounds under GEM because the ranker treats every variant as an independent candidate with its own predicted conversion probability. Twenty is the floor, forty is a healthier median. Format diversity matters as much as raw count: UGC, founder POV, static comparison, catalog dynamic, and motion graphics each cover different user contexts the ranker matches against.

Key takeaway: The 20 per month floor is not aspirational. Below it, GEM literally does not have enough candidates to rank your account well.

Measurement Under Andromeda: Why Blended CAC Beats Platform ROAS

Platform ROAS in Ads Manager understates and overstates depending on signal quality. GEM optimises for signals GA4 and Shopify cannot fully see, so the platform number diverges from ledger truth. Blended CAC (total paid spend divided by new customers) is the number that maps to unit economics. Every serious 2026 account we run has a blended CAC target ratified with finance, a rolling 4 week MMM view, and a quarterly geo holdout.

For the dashboard structure, see our CMO dashboards for paid media guide. For the incrementality tests that ground the numbers, see our incrementality testing for paid media walk through.

Key takeaway: Ads Manager ROAS is a diagnostic. Blended CAC and geo holdouts are the truth.

A 30 Day Plan to Rebuild a Meta Account for Andromeda

The 30 day rebuild is the fastest path to move a legacy account onto the Andromeda + GEM playbook.

Week 1: audit, CAPI, and catalog cleanup

Audit all events for EMQ score. Ship CAPI via Shopify native, Stape, or direct server. Clean the catalog: images, prices, custom labels, feed rules.

Week 2: consolidate campaign structure

Collapse to one Advantage+ Shopping, one retention, one testing campaign. Kill legacy ABO / interest-stacked ad sets. Remove manual bid caps.

Week 3: ship 20 plus creative concepts

Brief 15 UGC + founder concepts. Generate 5 to 10 Advantage+ Creative Enhancement variants from clean source assets. Launch inside the testing campaign.

Week 4: measure with blended CAC and geo holdout

Reconcile Ads Manager against GA4, Shopify, and finance ledger. Run a two week geo holdout for incrementality. Adjust Advantage+ Shopping budget on the blended CAC number, not the platform ROAS.

Key takeaway: The single highest-leverage move in the first 30 days is doubling your monthly creative concept count. Everything else is secondary.

If you want our team to run this on your account, see our Meta Ads agency service page or book a free audit.

Frequently Asked Questions

Does Andromeda replace Advantage+ Shopping?

No. Advantage+ Shopping is a campaign type in Ads Manager that you can pick as an advertiser. Andromeda and GEM are the backend retrieval and ranking layers that power every campaign type, Advantage+ or manual. Advantage+ Shopping happens to give GEM the most latitude, which is why it outperforms manual on 80 plus percent of the accounts we audit in 2026.

How much does creative volume actually matter under GEM?

It is the single biggest lever. Brands running 20 or more new ads per month see 65 percent higher ROAS than brands testing under 10, per Meta's own 2025 to 2026 aggregate data. GEM ranks per-impression against creative embeddings, so a fresh candidate pool directly changes what the ranker can serve. One hero creative rotated quarterly starves the model.

Can small budgets under USD 5K per month still benefit from Andromeda?

Yes, with caveats. Under USD 5K per month, you cannot afford 20 concept UGC production, but you can generate 15 to 20 monthly variants from 4 to 6 source assets using Advantage+ Creative Enhancements. Signal quality (CAPI, EMQ 8.0 plus) matters more, not less, at small budgets because there is less raw data to work with. Structure stays the same: one Advantage+ Shopping, one retention, one testing campaign.

Further reading: Meta's engineering post on Andromeda retrieval, Meta's GEM technical paper, Meta's Q1 2026 earnings call, and the Common Thread Collective analysis of Andromeda for DTC brands.