Meta’s Andromeda Update: Everything Advertisers Need to Know About the Biggest Shift in Facebook and Instagram Advertising

What Is Meta’s Andromeda Update?

Meta’s Andromeda update is a complete overhaul of the ad delivery system that powers Facebook and Instagram advertising. It replaced the legacy ad retrieval engine that advertisers had relied on for years with an entirely new AI-driven system built on advanced machine learning, deep neural networks, and custom hardware including the NVIDIA Grace Hopper Superchip and Meta’s own Training and Inference Accelerator (MTIA).

Andromeda began rolling out in late 2024 and completed its global deployment by October 2025. Industry experts and Meta itself have described it as the most significant change to the platform’s advertising infrastructure since the introduction of Advantage+ campaigns in 2022.

At its core, Andromeda is a personalized ads retrieval engine. It operates at the very first stage of Meta’s ad delivery process — the retrieval stage — where the system filters tens of millions of candidate ads down to a few thousand eligible ads for each individual impression opportunity. Think of it as the gatekeeper that decides which ads even get a chance to compete in the auction. If your ad doesn’t pass through Andromeda’s retrieval filter, it never gets seen — regardless of your budget, your bid, or your targeting settings.

The previous retrieval system relied heavily on rule-based logic, isolated model stages, and manual heuristics to manage the ad selection process. It worked well enough when advertisers were uploading a handful of ads per campaign. But the explosion of AI-powered creative tools — Advantage+ creative, generative AI, dynamic creative optimization — meant advertisers were suddenly producing 50, 100, or even hundreds of ad variations per campaign. The old system simply couldn’t process that volume efficiently. Retrieval had become the bottleneck.

Andromeda was built to solve that bottleneck. Meta’s engineering team described it as a 10,000x increase in the complexity of models used for ad retrieval. The system uses deep neural networks with massive parallelism to scan millions of ads in milliseconds, evaluating not just basic demographic matches but hundreds of behavioral signals — engagement patterns, watch times, purchase history, content interactions, save behaviors, and more — to predict which specific ad each individual user is most likely to respond to.

The result is a system that doesn’t just ask “who should see this ad?” It asks “which ad should this person see?” — and it makes that determination based on real-time behavioral intelligence rather than the static audience definitions advertisers have traditionally provided.

How Andromeda Changed the Way Ad Delivery Works

To understand why Andromeda matters so much, it helps to understand how Meta’s ad delivery system worked before — and what specifically changed at each stage.

Meta’s ad delivery process has always operated in three stages. First, retrieval — where the system narrows millions of candidate ads to a manageable shortlist. Second, ranking — where the shortlisted ads are evaluated and scored based on predicted engagement, relevance, and conversion likelihood. Third, the auction — where the highest-ranked ads compete based on bid amount, relevance score, and estimated action rates, and the winner gets displayed to the user.

Andromeda replaced the first stage. It didn’t fundamentally change how the auction works or how ads are ranked once they make it to the final competition. What it changed is which ads get to compete in the first place — and it changed that dramatically.

Under the old system, retrieval was relatively constrained. The system evaluated a limited pool of eligible ads based primarily on the targeting inputs advertisers provided — demographics, interests, behaviors, lookalike audiences, and geographic parameters. If you set up your targeting correctly, your ads got into the consideration set. If you didn’t, they didn’t. The advertiser controlled the experience.

Under Andromeda, the system controls the experience. Instead of starting with the advertiser’s audience definition and finding ads that match, Andromeda works in reverse. It starts by evaluating the ad itself — its creative content, visual elements, messaging, format, historical engagement data — and then uses AI to determine which users are most likely to find that ad relevant and take action on it.

This is the fundamental shift. Targeting used to be the primary lever advertisers controlled to drive performance. Now, creative is the primary lever. Your targeting settings still exist, but they function more as loose signal inputs than hard boundaries. Meta’s own recommendation as of 2025 is for advertisers to adopt broad targeting, use Advantage+ placements, and allow the algorithm to make its own delivery optimizations.

The ads that make it through Andromeda’s retrieval filter and into the auction are already more relevant to the users they’re being shown to. That means the entire system performs better end-to-end — but only when advertisers provide the creative diversity and quality that the system needs to work.

What Andromeda Means for Organic Content

Andromeda’s changes extend beyond paid advertising. Meta’s updated ranking systems now prioritize meaningful engagement across both paid and organic content. Under the new system, content that generates saves, shares, comments, extended watch times, and repeat interactions is rewarded with greater reach and organic discovery.

This is significant because Andromeda’s algorithm evaluates engagement depth — not just surface-level metrics like likes or impressions. Content that makes someone stop scrolling, spend time reading or watching, and interact meaningfully sends strong signals to the system. Those signals influence not just how that individual piece of content performs, but how the algorithm treats future content from the same account.

For businesses, this means organic content strategy and paid advertising strategy are no longer separate disciplines. They work together within Andromeda’s ecosystem. Strong organic engagement reinforces paid campaign performance, and vice versa. Brands that invest in authentic, engagement-driven organic content alongside their paid campaigns create a compounding advantage that brands relying solely on paid advertising cannot match.

The types of organic content that perform best under Andromeda include original storytelling, behind-the-scenes content, tutorials and educational posts, content that invites conversation and audience participation, and short-form video — particularly Reels. Visual content, especially photos and Reels, has seen the strongest performance gains, with some publishers reporting referral traffic increases of up to four times compared to the previous year.

The Role of Creative Diversity Under Andromeda

If there is one concept that defines advertising success under Andromeda, it is creative diversity — and Meta has been explicit about this.

In their April 2025 blog, Meta stated that the focus has shifted from niche targeting to creative diversification as the best lever to find the most relevant audiences. This isn’t a suggestion — it’s the operating reality of how the platform now works.

Here’s why creative diversity matters so much under the new system. Andromeda thrives on having options. When you provide the algorithm with a range of genuinely different creative assets — different messages, different visual styles, different emotional appeals, different formats, different calls to action — it can test and learn which specific creative resonates with which specific user segment. The more meaningful variation you provide, the more effectively the system can personalize delivery at the individual level.

But the definition of “diversity” under Andromeda is stricter than many advertisers realize. The system doesn’t just analyze words and headlines — it pattern-matches visual cues, video structure, narrative tone, frame dynamics, and creative composition. If you upload 12 ads that look and feel similar — even if the headlines are different — Andromeda may group them together and treat them as a single creative option. If the system determines that the core concept of that cluster isn’t performing well, it suppresses all variations simultaneously.

This is why so many advertisers saw their “winning” ads suddenly stop delivering in early 2025. The algorithm had evaluated the entire creative cluster as fatigued or irrelevant, and there wasn’t enough meaningful diversity to give it alternative options.

True creative diversity under Andromeda means producing ads that represent fundamentally different concepts, not just minor variations of the same idea. A campaign might explore entirely different themes — value versus aspiration versus urgency versus social proof versus transformation — each expressed through different visual approaches, different talent, different formats, and different storytelling structures. The goal is to prevent the algorithm from clustering your ads together by ensuring each asset targets a genuinely different psychological angle.

Previously, Meta recommended running three to six ads per ad set. That recommendation was quietly removed in early 2025 — almost certainly because of Andromeda. Advertisers are now finding success with far more creative variations, with some running 15, 20, or even 50 ads within a single campaign. The key isn’t hitting a specific number — it’s providing enough genuine diversity, paired with quality, that the algorithm has meaningful options to work with.

How Andromeda Works With GEM and Lattice

Andromeda doesn’t operate in isolation. It’s part of a broader AI infrastructure that Meta has been building throughout 2025, including two companion systems: GEM (Generative Ads Recommendation Model) and Lattice.

Understanding how these systems interact helps explain why the advertising landscape has shifted so dramatically.

Andromeda handles retrieval — it decides which ads are eligible to be shown by filtering millions of candidates down to a shortlist. Lattice handles ranking — it sorts and scores the shortlisted ads to determine which ones should appear first. GEM is the foundation model that sits underneath both systems, providing the predictive intelligence that powers their decisions.

GEM is Meta’s most advanced advertising AI model, built at the same scale as large language models. Meta officially introduced GEM in November 2025, though it had been running in the background since the second quarter of 2025. Unlike previous systems that optimized each placement separately — Instagram in one silo, Facebook Feed in another — GEM creates unified intelligence across Meta’s entire ecosystem. It learns from both paid ad interactions and organic content engagement across Facebook, Instagram, Messenger, and other Meta properties, and shares those learnings everywhere.

The practical impact for advertisers has been measurable. Meta reported that GEM delivered a 5% increase in ad conversions on Instagram and a 3% increase on Facebook Feed, with those improvements doubling in the third quarter of 2025 as the model was refined. These gains happened automatically — no opt-in required, no campaign restructuring needed. Every advertiser running on Meta’s platforms is already benefiting from (or being affected by) GEM’s algorithms.

The relationship between these systems works like this: Andromeda decides which ads make it onto the shelf. GEM learns what shoppers actually buy and shapes what gets featured next. Lattice determines the final ranking order. Together, they create an advertising delivery system that is dramatically more intelligent, more personalized, and more dependent on creative quality than anything Meta has deployed before.

Why Your Facebook Ads Stopped Working in 2025

If your Facebook or Instagram ad performance declined noticeably between late 2024 and mid-2025, Andromeda is almost certainly the reason. The timing of the rollout coincided precisely with the performance disruptions that thousands of advertisers experienced — rising CPMs, inconsistent delivery, campaigns that would surge one week and stall the next, and “winning” ads that suddenly stopped delivering.

The core issue was that most advertisers’ campaign structures were designed for the old system. They were optimized for precise targeting, limited creative variation, and manual control over audience selection — exactly the approach that Andromeda deprioritizes or actively penalizes.

Several specific patterns caused the most problems under the new system. Running limited creative — three to five similar ad variations — gave Andromeda almost nothing to work with. The algorithm grouped similar ads together and treated them as a single option, severely limiting its ability to personalize delivery. Detailed targeting restrictions — selecting specific interests, demographics, and behaviors — actually limited the algorithm’s ability to find high-performing audiences. Andromeda performs best with broad targeting because it needs the widest possible data space to learn and optimize. Fragmented account structures — running many separate ad sets for different audiences, placements, or minor creative variations — split learning across too many campaigns and prevented the system from optimizing effectively. And infrequent creative refreshes — letting the same ads run for months — created fatigue faster under Andromeda, because the system cycles through creative faster and top performers can drop off within two to four weeks.

Advertisers who recognized these patterns and adapted early saw performance recover and often improve beyond pre-Andromeda levels. Those who continued running legacy strategies continued to struggle — and many are still struggling heading into 2026.

Andromeda’s Impact on Ad Costs and Performance Metrics

Andromeda has had a measurable impact on key advertising metrics, though the effects vary significantly depending on how well campaigns are structured for the new system.

Meta’s own data indicates an 8% improvement in ad quality on selected segments and reports that the system is 4x more efficient at driving performance gains compared to the previous delivery engine. Advertisers who turned on Advantage+ creative features experienced a 22% increase in return on ad spend. Campaigns using Advantage+ with broad targeting showed up to 10% lower cost per lead compared to traditional manual campaigns. And more than a million advertisers used Meta’s generative AI creative tools to create over 15 million ads in a single month — with Meta estimating a 7% increase in conversions for businesses using AI-generated images.

However, these improvements are averages across millions of advertisers and are not guaranteed for any individual account. Some campaigns saw dramatic improvements. Others saw minimal change or continued declines — particularly those that didn’t adjust their creative strategy or campaign structure.

One important change that Andromeda introduced is a shift in which metrics matter most. Surface-level metrics like click-through rate and cost per click are less meaningful under the new system. The algorithm optimizes for deeper engagement signals — watch time, saves, shares, comments, conversion events — and evaluates performance holistically at the campaign level rather than at the individual ad set level. Advertisers who focus too heavily on individual ad performance or short-term fluctuations may make premature optimization decisions that actually hurt long-term results.

Attribution has also become less precise at the individual ad set level under Andromeda’s broad delivery approach. Meta recommends evaluating overall campaign performance and creative performance rather than trying to track results at granular audience segment levels.

What Andromeda Means for Local Businesses

Andromeda’s shift toward broad targeting raises a particular concern for local businesses — and it’s one that requires careful navigation.

For most advertisers, Meta now recommends going as broad as possible: targeting an entire country with no age selection, no gender selection, and no interest targeting, and letting the algorithm find the right audience. For national brands and e-commerce businesses, this approach works. The algorithm has enough data and geographic flexibility to optimize delivery across the entire user base.

For local businesses — a medical spa in San Diego, a VR entertainment venue in Fort Worth, an automotive dealership in Jacksonville — broad national targeting doesn’t work. A spa in one city doesn’t want to generate leads from customers three states away. Geographic restrictions are still necessary and still effective.

The recommendation for local businesses under Andromeda is to use geographic targeting as a boundary — restricting delivery to your service area — but to go broad on everything else within that geography. Remove interest targeting, age restrictions, and gender selections unless your product or service absolutely requires them. Let Andromeda handle audience selection within your geographic area, and focus your efforts on producing diverse, high-quality creative that gives the algorithm enough variation to find your ideal local customers.