Your ads didn't break. Meta's algorithm did. If your ROAS dropped, your CPAs crept up, or campaigns that ran reliably for months suddenly stalled, the Meta Andromeda update is almost certainly to blame.
The Meta Andromeda update is Meta's next-generation personalized ads retrieval engine, announced in December 2024 and fully rolled out globally by October 2025.
Built on NVIDIA Grace Hopper Superchips with a 10,000x increase in model capacity, it replaced the previous retrieval system entirely. Instead of matching ads to predefined audience segments, Andromeda evaluates tens of millions of ad candidates in real time and selects the creative most likely to resonate with each individual user.
This guide covers what changed, why it matters for campaign performance, and what your creative team needs to do in this new era.
For the expert perspective on winning under Andromeda, see the winning creative playbook for 2026 from three top Meta practitioners.
What Is the Meta Andromeda Update?
Image source: Unsplash
Andromeda is Meta's proprietary machine learning system for ad retrieval: the stage in Meta's ad delivery process where the system narrows tens of millions of active ads down to a shortlist of candidates before ranking and auction.
The practical results are measurable.
Meta's internal testing showed a 6% improvement in recall and an 8% improvement in ad quality on selected segments. Advertisers using Advantage+ Creative saw a 22% uplift in ROAS.
Those numbers don't happen automatically. They depend entirely on how well you feed the new system.
The old system vs the new one
Before Andromeda, Meta's retrieval system matched ads to users through advertiser-defined targeting: interest stacks, demographics, lookalike audiences, and manual exclusions.
The system evaluated a constrained set of eligible ads based on predefined targeting and passed them into the auction. Advertisers controlled who saw each ad.
Andromeda changes what gets evaluated.
The new brain reads the creative itself: its visual elements, hooks, copy themes, format, and semantic meaning. It then matches those signals to individual user behavior patterns. Your creative content determines your audience, not the other way around.
| - | Before Andromeda | After Andromeda |
|---|---|---|
| Ad selection method | Audience targeting rules | Creative signal matching |
| What drives delivery | Who you targeted | What your ad communicates |
| Creative lifespan | Several months | 2-4 weeks before fatigue |
| Ads per set | 3-6 | 10-20 distinct concepts |
| Campaign structure | Multiple segmented ad sets | One broad ad set, more creatives |
| Testing lever | Audience variations | Creative concept variations |
How Andromeda Changed What Creative Teams Must Deliver
Andromeda didn't raise the bar on creative quality. It raised the bar on creative volume, creative velocity, and creative variety. One strong ad used to run for months. Under Andromeda, the same ad fatigues in two to four weeks.
A creative similarity score above 60% risks suppression before your ad even enters the auction. Similar creatives don't share the load: they cannibalize each other.
Creative diversity is now the primary performance lever. The more conceptually distinct your ad set is, the more behavioral signals Andromeda can collect, and the better it matches your assets to users most likely to convert.
Creative fatigue arrives faster now
Andromeda's retrieval system cycles through creative assets at a higher velocity than the old system did. It burns through impressions faster, which means your best-performing ad reaches saturation much sooner.
What once lasted a quarter now needs replacing within the month.
Ads that held for six to eight weeks are now fading in two to three. You'll see rising CPMs and declining CTRs. Don't wait for campaign performance to collapse before refreshing. By then, you've wasted budget on a fatigued asset, and Andromeda has moved on.
Creative similarity score and suppression
Andromeda doesn't just evaluate individual ads. It compares your ads. Two ads scoring above 60% on creative similarity risk suppression before the auction even starts. Similarity isn't just pixels. Andromeda reads semantic meaning, visual structure, hook patterns, and copy themes.
Two videos with different actors, scripts, and settings can still be considered similar if they share a format, tone, or narrative arc. That's where creative burnout compounds the problem: teams produce multiple creative variations of the same idea, and the system struggles to distinguish them.
The new production math
Under the previous system, Meta recommended three to six ads per ad set. That guidance no longer applies in this new environment.
Most practitioners now run 10 to 20 semantically distinct concepts per ad set. Each concept should represent a meaningfully different angle: a different persona, a different emotional trigger, a different awareness stage, or a different format.
Small tweaks don't count as multiple creative variations.
Swapping a headline or changing a background color produces high similarity scores. Concept diversity is what Andromeda rewards: different creative directions built around different ideas, not different executions of the same idea.
Reframe Ad Strategy: From Targeting to Creative Strategy
Andromeda represents a structural break from a decade of Meta strategy. The skill that drove performance for many advertisers: building precise audience segments through predefined targeting, is now actively hurting results.
Over-segmentation fragments the budget, starves the algorithm of learning signals, and cuts off the audience behaviour data that Andromeda needs to optimize.
The pain points are consistent across accounts: sudden CPA spikes, delivery instability, and scaling that stalls without explanation. The cause isn't the offer or the landing page. It's the structure.
Why audience stacks work against you
Multiple segmented ad sets with stacked interests, overlapping lookalikes, and heavy exclusions were once standard practice. They told Meta exactly who to target. Now they tell Meta where not to look, and that limitation costs you.
Andromeda's retrieval architecture learns from behavioral signals at the individual level. Narrow targeting cuts off the signal pool it needs to learn from. Brands running tight audience stacks report sharper delivery instability and weaker scaling after the rollout.
How creative matching replaces demographic filters
Your ad's hook determines who pays attention. Its value proposition determines who clicks. Its visual approach determines who identifies with it. Each of those choices sends a signal to Andromeda about which users are most likely to respond, without you needing to define those users manually.
Creative matching is now the mechanism.
You're no longer buying access to a defined audience. You're letting the creative self-select its audience through Andromeda's retrieval signals. The more distinct concepts you provide, the more micro-segments Andromeda can identify and serve.
For a step-by-step approach to mining competitor creatives for concept ideas across multiple angles, see how to spy on competitor ads on Meta and TikTok.
What strategic clarity means in the Andromeda era
Strategic clarity now means knowing which concept directions to pursue before you build anything. That requires two inputs most teams don't have: what's winning in the market right now, and what your competitors are running. Without those inputs, the briefing is guesswork.
Build one concept per distinct audience pain point or motivation. Not one ad per motivation. One concept, with multiple creative variations branching from it.
Each concept needs to be semantically distinct from all others. If you're working from a brief, stay ahead by starting with market intelligence rather than internal assumptions.
If your creative strategy needs a reset, start a free Atria trial and let Raya find what's winning in your market right now.
Campaign Structure Recommendations for Facebook and Instagram Ads
Image source: Unsplash
Simplified structures outperform complex ones under Andromeda. One campaign, one broad ad set, 10-20 distinct creative concepts. That's the new baseline. The system continues to reward consolidation over complexity at every level.
The recommended structure
Run a single campaign with a single broad ad set. Enable Advantage+ placements. Remove manual exclusions and interest stacks. Feed the ad set 10 to 20 semantically distinct concepts and let Andromeda decide who sees what at the ad level.
Every additional ad set you create splits your data. Split data means slower learning. Slower learning means worse delivery. Budget allocation across multiple ad sets starves each one of the conversion volume it needs to exit the learning phase.
Andromeda performs better with broader audiences and fewer structural constraints, and that only comes from consolidation. Fewer, simpler structures improve performance because the algorithm has the signal volume it needs to optimize.
What to do with legacy segmented ad sets
Don't kill them immediately. If existing segmented campaigns are still generating results, let them run while you build a parallel Andromeda-structured campaign alongside. Shift budget incrementally as the new structure proves itself, usually over two to three weeks.
Watch for the crossover point: when the new campaign's CPA stability matches or beats the legacy structure, migrate the full budget. Abrupt shutdowns disrupt meaningful learning and can spike costs during the transition.
Advantage+ settings to enable for new targeting
Turn on Advantage+ audience targeting, which allows Meta to expand beyond your seed audience when the algorithm identifies similar behavioral signals.
Enable Advantage+ placements so the system distributes across Facebook feed, Instagram feed, Reels, Stories, and the Audience Network based on real-time performance.
Set your campaign objective before running ads and don't change it. Andromeda optimizes strictly around the objective it was given. Switching mid-campaign resets the learning phase and costs you the signal accumulation you've already built.
Creative Production and Refresh Cadence
Most teams can't produce 10 to 20 distinct concepts per month on their current workflow. That's the real operational challenge Andromeda created. The brief-to-launch cycle is too slow, the briefing process produces too many similar concepts, and there's no reliable system for knowing which concepts to double down on before the budget is wasted.
The production loop: brief, variant, launch, grade, repeat
Atria, the AI-powered creative intelligence platform, responds to this new era with a production loop built for creative velocity: brief, variant, launch, grade, repeat. Raya, Atria's AI creative strategist, trained on over $5 billion in ad spend data, runs each stage.
The loop works like this.
Raya analyzes your top-performing creatives alongside competitor ad intelligence to generate new concept briefs. Those briefs go into variant production: not a single execution, but multiple formats for each concept.
Approved variants are launched directly to Meta from Atria. Raya grades live creative performance and finds winners before you've burned a meaningful budget on losers. Meta sees a steady stream of fresh, distinct creative signals.
Then the cycle repeats with fresh concepts informed by what just worked.
Most creative analysis tools only show you your own performance data. Raya layers your performance data with $5B+ in market spend data and competitor ad intelligence. You're not briefing in a vacuum. You know what's winning in your market before you build, which is exactly what meaningful learning at scale requires.
Refresh triggers: when to replace, not just add
Watch for three signals that a creative needs replacing: rising CPMs on a previously stable asset, a CTR decline of more than 20% week-over-week, and a high creative similarity score surfacing in your Meta dashboard.
Don't wait for all three. One sustained signal is enough. Adding a new concept is cheap. Continuing to run a fatigued ad that Andromeda has deprioritized is not.
How often to brief new concepts
For accounts spending $10,000 or more per month on Meta, a weekly concept review is the right cadence. That doesn't mean launching more ads every week. It means evaluating your creative library weekly and replacing any asset showing fatigue signals.
Smaller accounts can operate on a two-week review cycle. Monthly refreshes are no longer sufficient under Andromeda's faster delivery pace.
Want to see how this works with your actual creative library?
Book a demo, and we'll walk through the production loop.
Tracking, Data, and Measurement for Meta Ads
Andromeda learns from your conversion signals. If your tracking is unreliable, the algorithm optimizes against bad data, and your campaign performance reflects that. Clean data isn't optional. It's what determines whether Andromeda can do its job.
Conversions API setup priorities
Implement server-side event tracking through Meta's Conversions API. This sends conversion data directly from your server to Meta, bypassing browser-based tracking limitations.
Deduplicate events to avoid double-counting, verify your domain in Meta Business Manager, and confirm that all purchase, lead, and add-to-cart events are firing correctly.
Aim for at least 50 conversions per ad set per week. That's the threshold Meta's system needs to exit the learning phase and optimize delivery efficiently. Below that threshold, Andromeda can't learn fast enough to make reliable matching decisions.
What metrics to watch at the campaign and ad level
Shift your analysis from individual ad ROAS to campaign-level creative group performance. Andromeda's delivery is fluid. It cycles through creative options based on real-time signals. An ad that looks weak at the ad level in isolation may be serving an important role in the full creative mix.
Track these at the campaign level: creative fatigue rate, CTR trend by concept group, CPM trajectory, and frequency by creative. For automated creative scoring across your full library, see how to automatically score and turn underperforming ads into winners).
Atria's visual performance reports surface these signals in a single view. For a comparison of what's available, see our review of creative reporting tools for Meta ads.
Operational Playbook: Roles, Creative Libraries, and Workflow
The creative production requirements Andromeda demands are beyond what most human teams can sustain manually. The teams winning right now have restructured: they've defined what humans decide and what AI handles. That division is where the efficiency gains come from.
See how to achieve it with Atria's Raya.
Who does what in an Andromeda-ready team
| Task | Human creative team | Raya (Atria AI) |
|---|---|---|
| Creative concept approval | Reviews and approves | Generates options based on $5B+ data |
| Competitor intelligence | Reviews insights | Pulls and analyzes competitor creatives |
| New concept briefing | Provides brand direction | Drafts briefs using winning signals |
| Variant production | Reviews for brand fit | Generates multiple creative variations |
| Creative grading | Makes final scaling calls | Scores live creative assets before budget scales |
| Performance reporting | Interprets and acts | Builds visual reports on demand |
The vision is to cut creative workflow from roughly 40 hours per week to four. The research, the competitive intel, the briefing, the grading: Raya handles all of it.
Your team focuses on the calls that require judgment, brand knowledge, and creative instinct.
Building a living creative library
A creative library isn't a folder of assets. It's a structured, tagged repository that makes it fast to identify what's working, what's fatiguing, and what gaps need filling. Every creative asset should be tagged by concept type, persona, format, and hook: not just labeled by campaign.
Atria handles tagging automatically. Your library stays current without manual effort. That tagging is also what enables Raya to identify patterns across winning ads: she learns from what's tagged, not just what's uploaded.
Meta's algorithm then receives cleaner signals because your internal classification maps directly to the concept diversity it needs.
For more on building a scalable tagging system, read how to automate creative tagging.
How Raya cuts the operational overhead
Most creative strategy work is operational: pulling competitor ads, reading customer reviews, identifying hook patterns, building briefs, generating variants. That work used to take a full creative team a week. Raya does it before your morning standup.
The output isn't generic AI content. It's strategy built on $5B+ in verified ad spend data. Raya knows which hooks are performing in your vertical right now, which competitor angles are gaining traction, and which concept directions your account hasn't tried yet.
That's the data advantage competitors without Atria don't have. Most creative analysis tools show you your own performance. Atria adds the market layer: what's winning across the industry, not just in your account.
Book a demo to learn more.
Your Andromeda Readiness Checklist
Andromeda changed the required output from creative teams. It didn't change the fundamentals. Great creative still wins. It just raised the creative velocity and concept diversity required to stay competitive.
Use this checklist before your next campaign launch:
- Targeting broadened to the broad or Advantage+ audience, with manual exclusions and interest stacks removed
- Campaign consolidated to one campaign, one broad ad set
- Creative library contains 10 to 20 semantically distinct concepts: not multiple creative variations of the same idea
- Creative similarity score verified below 60% across all active ads in the set
- Weekly creative review cycle in place, with refresh triggers defined: rising CPMs, CTR decline, similarity score
- Conversions API implemented with server-side events, deduplication, and domain verification
- Campaign performance analysis shifted to campaign-level creative group metrics, not individual ad ROAS
- New concept briefs informed by competitor intelligence and market spend data, not just internal history
Atria's production loop (brief, variant, launch, grade, repeat) handles all of this systematically. Start a free trial and see what Raya surfaces from your creative library on day one.
Or book a demo, and we'll walk you through the full workflow.
Conclusion About Meta Andromeda Update
Andromeda isn't an algorithm tweak you can patch around. It's a new environment with different rules, and many advertisers are still running the old playbook.
The shift is structural. Meta's system no longer cares about the audience you defined. It cares about the creative you gave it. That means creative diversity, creative volume, and creative velocity are now the levers that determine whether your ad spend works or doesn't.
Budget allocation, campaign structure, and data quality all feed into that system. Get them right, and the algorithm amplifies your best work. Get them wrong, and it exposes every weakness faster than the old system ever did.
The teams pulling ahead in this new era aren't producing more ads. They're running a tighter production loop, briefing from market intelligence rather than gut instinct, and replacing fatigued creative assets before performance drops.
That's what Raya, Atria's AI creative strategist trained on $5B+ in ad spend data, is built to do: give your team the data advantage and the operational speed that Andromeda now demands.
Start a free trial and see what Raya finds in your creative library. Or book a demo for a live walkthrough.
FAQs About Meta Andromeda Update
What is the Meta Andromeda update?
The Meta Andromeda update is Meta's next-generation personalized ads retrieval engine, introduced in December 2024 and rolled out globally by October 2025. Built on NVIDIA Grace Hopper Superchips with a 10,000x increase in model capacity, it replaced Meta's previous retrieval system. Instead of matching ads to predefined audience segments, Andromeda reads creative signals and matches each user to the ad most likely to resonate with their individual behavior patterns.
How does Andromeda affect Facebook ad performance?
Andromeda shifts ad delivery from audience-based targeting to creative matching. Ads are now selected based on how well their creative signals (hooks, formats, value propositions, visual style) align with each user's individual intent and audience behaviour patterns. Campaigns with limited creative variety, over-segmented ad sets, or static hero ads tend to see declining campaign performance. Campaigns with broad targeting and 10 to 20 distinct creative concepts benefit from stronger ROAS.
Why did my Meta ads stop working after the update?
The most common causes are limited creative variety (Andromeda needs conceptually distinct ideas, not visual tweaks of similar creatives), over-segmented campaign structure (multiple segmented ad sets split learning signals), creative burnout accelerating faster under Andromeda's higher delivery velocity, or creative similarity scores above 60% causing suppression before ads enter the auction.
How many creatives do I need per ad set under Andromeda?
Most practitioners recommend 10 to 20 semantically distinct creative concepts per ad set. These should cover multiple angles: different value propositions, formats, personas, and awareness stages. Not multiple creative variations of the same visual or headline. Andromeda needs a conceptual range to match creative assets to different micro-segments effectively.
What is a creative similarity score, and how does it affect delivery?
Creative similarity score measures how semantically similar two ads are according to Andromeda's retrieval models. Ads scoring above 60% risk suppression: they may not enter the delivery auction at all, even if individually strong. Andromeda evaluates visual cues, hook patterns, copy themes, and narrative structure. Two ads with different actors and scripts can still score as similar if they share a format and tone.
How often should I refresh creatives after the Andromeda update?
Creative fatigue now typically hits in two to four weeks under Andromeda, compared to several months under the previous system. For accounts spending $10,000 or more per month, a weekly creative review is the right cadence. Smaller accounts can review every two weeks. Replace assets showing any sustained fatigue signal: rising CPMs, declining CTR, or increasing creative similarity scores. Don't wait for campaign performance to collapse before acting.
