Meta's Andromeda and GEM AI Engines Just Got 15.7% More Powerful: What

Meta Andromeda and GEM AI ad delivery system visualization

On October 28, 2026, Meta published engineering details on its two AI ad delivery engines, Andromeda and GEM (Generative Ads Recommendation Model), revealing that recent upgrades drove a 15.7% increase in conversions and 8.3% more ad clicks across Facebook and Instagram. For ecommerce brands running Advantage+ Shopping campaigns, this means the creative you upload is now the primary targeting signal, not the audience settings you build around it.

What Does Meta's Andromeda and GEM Upgrade Mean for Your Ad Campaigns?

Andromeda handles the first stage: narrowing tens of millions of candidate ads down to a shortlist in milliseconds, reading your creative visuals and copy to match your ad to receptive users. GEM then ranks that shortlist using deep engagement history and purchase-intent signals. Together, they explain why broad targeting in Advantage+ consistently outperforms narrow interest stacking, the system finds your buyers by reading your creative, not your audience settings.

How Do Andromeda and GEM Actually Work Inside Meta's Ad Auction?

Meta's ad delivery runs in two stages. Andromeda operates at retrieval: it processes over 1 billion ad candidates per second using NVIDIA Grace Hopper and Meta's own MTIA silicon. It replaced rule-based heuristics with AI that reads creative elements, including visual composition, copy hooks, and product imagery, to filter which ads get to the ranking stage. GEM, the foundation model, then scores the surviving candidates against each user's engagement history and predicted purchase intent. The incremental Andromeda update alone added a 1.4% conversion lift in Q3 2026. The broader GEM upgrade delivered up to 18.9% conversion improvement in controlled tests.

For 7-9 figure ecommerce brands, the operational implication is direct: your creative quality is now your targeting mechanism. A mediocre creative in a broad Advantage+ campaign will underperform not because you targeted incorrectly but because Andromeda could not identify a strong match signal in the asset itself.

Why Does Creative Quality Matter More Than Audience Targeting in 2026?

The shift is structural, not strategic. When Andromeda reads your creative to find matching users, it processes visual hooks, opening frames, product presentation, and copy clarity. Interest buckets and lookalike audiences inform the system, but they no longer dominate the retrieval decision. Brands that still rely on narrow audience stacking to compensate for weak creative are fighting the system's design. Brands that build high-signal creative (clear product, fast hook, strong visual contrast) and give GEM a broad population to work with are aligned with how Meta actually delivers ads in 2026.

This also explains the performance gap between Advantage+ Shopping campaigns and manually structured campaigns. ASC gives Andromeda and GEM the widest possible pool. Manual campaigns with tight audiences restrict the retrieval stage before it can do its job. See our full Meta Ads 2026 update tracker for context on how this fits the year's broader shift toward AI-driven delivery.

STAY CURRENT

Ad platforms change weekly. Your inbox should not miss it.

Get CTC's weekly breakdown of the platform changes that actually matter for ecommerce brands, no noise, just what affects your campaigns.

What Should You Do Now with Meta's Andromeda and GEM Update?

Audit your top 10 active creatives against Andromeda's reading criteria. Pull your highest-spend assets and evaluate the first 3 seconds. Is the product visible within frame 1? Is there a clear visual hook? If Andromeda cannot extract a clear signal from your opening frame, your creative is starting at a disadvantage before the auction begins.
Shift campaign structure toward Advantage+ Shopping with broad targeting. Consolidate any manually structured campaigns with tight interest stacks into ASC campaigns. Give Andromeda and GEM a wide enough audience pool to retrieve and rank effectively. Tight audiences reduce the system's ability to find high-intent matches outside your assumed customer profile.
Increase creative volume and variation. GEM improves predictions for new products with no prior interaction data, but it still needs multiple creative variations to identify which signals drive engagement for your specific catalog. Run at minimum 3 to 5 distinct creative concepts per product category and let GEM determine which combinations perform.
Do not strip Meta's AI watermarks from Advantage+ creative. Under Meta's new consolidated ad terms effective October 30, 2026, advertisers are responsible for all AI-generated assets and prohibited from removing AI authentication metadata. Review your creative production workflow before any Advantage+ AI creative runs in November.
Check your Product Browsing opt-out status before Black Friday. With Meta's auto-enabled Product Browsing active as of October 28, users tapping your ads may land on an in-app product page rather than your website. If your attribution model depends on landing page tracking, disable Product Browsing under Advantage+ Creative and Destination settings to restore direct-to-site traffic.

Frequently Asked Questions

What is Meta Andromeda and how does it affect my ad delivery?

Andromeda is Meta's first-stage AI retrieval engine. When a user opens Facebook or Instagram, Meta has tens of millions of candidate ads eligible for that impression. Andromeda narrows that pool to a few thousand candidates in milliseconds by reading your creative, including visuals, copy, and context, to identify which ads match the user's likely intent. Upgraded in late October 2026 to process over 1 billion candidates per second, it now operates at near-ranking accuracy in the retrieval phase.

What is Meta GEM and how is it different from Andromeda?

GEM (Generative Ads Recommendation Model) is Meta's foundation ranking model. It operates after Andromeda's retrieval stage, evaluating the shortlisted candidates against each user's engagement history, purchase signals, and contextual intent. In controlled tests, GEM updates delivered up to 18.9% conversion improvement. Combined with Andromeda, the two systems produced 15.7% more conversions and 8.3% more clicks on Facebook in the most recent measurement period.

Does this mean interest-based targeting no longer works on Meta?

Interest targeting still functions, but its role has diminished relative to creative quality. Andromeda reads creative elements to find user matches independent of stated interest buckets. Tight interest stacking restricts the population Andromeda can retrieve from, limiting the system's ability to find high-intent buyers outside your assumed audience profile. Broad targeting in Advantage+ Shopping campaigns gives the AI the widest retrieval pool.

How does this change how I should think about creative testing on Meta?

Creative testing becomes a data-collection exercise for Andromeda and GEM. Each distinct creative variation teaches the system which visual signals, copy hooks, and product presentations resonate with different user segments. More variation gives GEM more signal to work with. Run at minimum 3 to 5 conceptually distinct creatives per product category and give the system sufficient budget and time to learn before drawing conclusions.

Ready to Build a Meta Ad Strategy That Works With the Algorithm?

CTC's team runs Meta campaigns for 7-9 figure ecommerce brands with creative and structure designed for how Andromeda and GEM actually deliver ads. Talk to us about your Meta performance.

Common Thread Collective is the leading source of strategy and insight serving DTC ecommerce businesses. From agency services to educational resources for eccomerce leaders and marketers, CTC is committed to helping you do your job better.

For more content like this, sign up for our newsletter, listen to our podcast, or follow us on YouTube or Twitter.

Meta's Andromeda and GEM AI Engines Just Got 15.7% More Powerful: What Ecommerce Brands Must Know Before BFCM