Targeting matters less. Bidding matters less. Creative quality is now doing the work the algorithm used to do for you. Here is what Meta Andromeda actually is, and what your campaigns need to look like because of it.
TL;DR
Meta’s Andromeda AI system has fundamentally changed how paid social works. Targeting matters less. Bidding matters less. Creative quality and diversity are now the primary performance drivers. Advertisers adapting to Andromeda are seeing 22% ROAS increases. Those who are not adapting are watching their CPAs creep up despite unchanged bids and budgets. This article walks through exactly what Andromeda is, what your creative strategy needs to look like now, and the specific changes to make to your campaigns this week.
Andromeda is Meta’s AI-driven ads retrieval system, and it operates as the first stage of Meta’s multi-stage ad recommendation process. Before an ad even reaches the auction stage, Andromeda determines which ads from the advertiser pool are eligible to compete. [1]
Here is the part that changes everything: Andromeda does not optimize at the audience level. It optimizes at the individual user level. That means the old model, where you define a tight audience and trust the algorithm to find the right person within it, is fundamentally broken. Andromeda’s job is to find the right creative for each person, not the right person for your creative.
The system is powered by NVIDIA GH200 Grace Hopper Superchips and delivers a 6% improvement in retrieval recall along with, in certain segments, an 8% lift in ad quality. [2] Those are Meta’s numbers. What they mean in practice is that the algorithm is significantly better at identifying which creative will resonate with a specific person at a specific moment. The catch: it can only work with the creative signals you give it. Feed it narrow, repetitive creative and it has nothing to work with. Feed it diverse, high-quality creative and it has everything it needs.
Three things that used to matter a lot in paid social now matter much less with Andromeda:
Three things that now matter most:
The shift in plain language: Andromeda optimizes for semantic meaning. A video and a static image with the same concept read as one signal, not two. True creative diversity means different angles, hooks, emotional appeals, and formats, not the same ad with a different background colour.
The accounts seeing the strongest Andromeda results are running what I call a concept-first creative strategy. They start with 3 to 5 distinct creative concepts and then build assets across formats for each concept. This is fundamentally different from starting with a format and varying execution.
Say you are selling a project management tool for marketing teams. Your five concepts might be:
Each of those is a genuinely different semantic concept. Andromeda will read them as distinct signals. Then for each concept, you build assets: a 30-second video, a static image, a carousel, and maybe a UGC clip. That gives you 20 assets across 5 concepts. More importantly, it gives Andromeda 5 genuine creative directions to test against each user.
One documented case: an ad set with 25 diverse creatives beat a traditional 5-ad-set structure by 17% on conversions and 16% on cost per conversion. [4] That is not a marginal improvement. That is a structural advantage.
Accounts providing 3 to 5 distinct creative concepts across multiple formats consistently outperform those focusing on a single creative type, regardless of how many variations exist within that type. [5] The recommended format diversity looks like this:
What does not work: launching 15 versions of the same static ad with different background colours. Andromeda reads these as one signal with minor surface variation. You have not given it more to work with. You have given it the same thing 15 times.
The campaign structure that makes most sense now is simpler than what many marketers were running 18 months ago. This might feel counterintuitive after years of granular ad set segmentation, but Andromeda needs consolidation to work properly.
One Advantage+ campaign per primary objective (acquisition, retargeting, upsell). Within that campaign, one to two ad sets maximum, with broad audience targeting or Advantage+ Audience. Within each ad set, 10 to 15 distinct creative assets across your concept and format matrix.
The creative rotation handles what ad set segmentation used to handle. Andromeda allocates impressions based on which creative performs best for each individual user, dynamically. Splitting this across five ad sets with separate budgets does not help Andromeda learn faster. It slows the learning phase and fragments your signal.
Practical note: If you are running retargeting, keep it as a separate campaign with a small budget (10-15% of total spend). Retargeting audiences are small enough that Andromeda’s individual-level optimization is less relevant. Tight targeting still makes sense for retargeting. It does not make sense for cold acquisition anymore.
The volume of creative Andromeda demands is genuinely challenging for small teams. This is where AI production tools become operationally essential, not just nice to have.
For concept ideation: ChatGPT or Claude with a structured brief about your product, audience, and top-performing angles. Ask for 10 distinct creative concepts with different emotional hooks and proof points. Use the output as a starting point, not a finished script.
For UGC at scale: Tools like Arcads, Creatify, or Captions.ai generate UGC-style video content with AI-generated presenters. Not perfect for premium brand contexts, but effective for testing concept-level hypotheses before investing in real creator partnerships.
For static creative iteration: Canva’s AI tools, Adobe Firefly, or Midjourney for visual concepts. The key is using these to produce genuinely different visual directions, not just colour variants of a single layout.
For copy across assets: Build a prompt that includes your concept, your audience’s top objection, and the specific placement (Feed vs. Stories vs. Reels). Copy length and tone should vary by placement, and a good prompt structure makes this fast. See our guide on AI ad copy that converts for exact prompt templates.
Here is what to do in the next two weeks:
This week: Audit your current Advantage+ campaigns. Count how many conceptually distinct creative concepts you are running. If the answer is fewer than four, that is your bottleneck. Start building a creative brief for three to five new concepts based on different angles (pain point, social proof, feature, comparison, aspiration).
Next week: Consolidate your ad set structure. If you are running more than three ad sets within a single acquisition campaign, merge them. Increase the creative count per ad set using your new concepts across multiple formats. Turn on Advantage+ Audience if you are not already using it, and widen any audience restrictions you have in place.
Track the learning phase carefully. After consolidation, expect a 7 to 14 day learning period before performance stabilizes. Do not panic and make changes mid-learning phase. Give Andromeda the signals and the time to optimize. For more on how AI is reshaping the full marketing funnel, see our piece on agentic marketing in 2026.
What is Meta Andromeda?
Meta Andromeda is an AI-driven ads retrieval system that serves as the first stage in Meta’s multi-stage ad recommendation process. Powered by NVIDIA GH200 Grace Hopper Superchips, it optimizes at the individual user level rather than the audience level, making creative quality the primary determinant of which ads get shown.
How does Meta Andromeda change paid social advertising?
Andromeda shifts the performance lever from audience targeting to creative quality and diversity. Traditional tactics like narrow audience targeting, aggressive bidding, and multiple segmented ad sets matter significantly less. Creative diversity across concepts and formats matters most. The algorithm needs distinct semantic signals to match the right creative to the right person.
How many creatives should I launch per Advantage+ campaign?
Experts recommend launching 10 to 15 conceptually distinct assets per Advantage+ campaign. Minor variations like colour changes read as identical to the algorithm. Each asset needs to be a genuinely different creative concept, not just a visual variation of the same idea.
What creative formats work best with Meta Andromeda?
Accounts providing 3 to 5 distinct creative concepts across multiple formats consistently outperform single-format approaches. The recommended mix includes static images, carousel ads, short-form video (15-30 seconds), and UGC-style content. The algorithm values semantic diversity across formats, not just format quantity.
What ROAS improvements are Meta advertisers seeing with Andromeda?
Advertisers who understand and adapt to Andromeda are seeing an average 22% increase in ROAS. One documented case showed an ad set with 25 diverse creatives beating a traditional 5-ad-set structure by 17% on conversions and 16% on cost per conversion. Results vary based on how effectively the new creative strategy is implemented.
About This Guide
Written by Hina Mian, co-founder of Future Factors AI and marketing strategist with 10+ years of experience running paid social campaigns. This article draws on published performance data, industry research, and documented case studies from advertisers adapting to Meta’s Andromeda system.
Sources
Weekly tips. Real examples. Practical help for busy professionals.
We care about your data, check out our privacy policy.
