Meta Andromeda Explained: Creative Is the New Targeting

What is Meta's Andromeda architecture?

Meta Andromeda is the ad-retrieval engine Meta began rolling out in 2024 that pulls candidate ads for each impression from a pool of billions, using machine-learned embeddings instead of the older index-based lookup. It sits upstream of Meta's ranking system and decides which ads even enter the auction for a given person.

Before Andromeda, retrieval worked mostly like a keyword index: an advertiser declared an audience, Meta matched a user against that index, and ranking happened afterward on a comparatively small shortlist. Andromeda instead represents both users and ads as vectors in a shared space, running on GPU infrastructure Meta has described as built to compare a person against a far larger candidate set in real time.

The practical shift is what data enters that vector: creative content, not just declared interest categories, now shapes which candidates surface. Meta has not published exact figures for pool size or refresh rate, and any specific number you see quoted online deserves a skeptical read until Meta confirms it directly.

How does creative now determine delivery?

Creative now functions as a targeting input, because Andromeda embeds the visual and textual content of an ad alongside behavioral signals when it decides who sees that ad at all. Two ads aimed at the identical saved audience can reach meaningfully different people once retrieval, not just the ad set's audience field, does the sorting.

This raises the value of creative volume and variety over manual audience segmentation. Faster production pipelines matter more than they used to — the same shift that pushed TikTok to ship TikTok Symphony as a free AI creative suite reflects Meta's own push toward high-volume creative testing feeding an algorithm that reads content directly rather than an audience label.

Early engagement in the first few seconds — hook retention, sound-on completion, comment velocity — now doubles as a retrieval signal rather than a downstream optimization metric you check after the fact. An ad that earns strong early engagement appears to get pulled into more candidate pools, though Meta has not published the exact weighting, and any specific percentage attributed to 'hook rate' should be treated as an estimate, not a documented threshold.

Are interest and lookalike audiences dead?

Interest and lookalike audiences are not dead, but they no longer function as the primary lever for scaling reach the way they did between roughly 2017 and 2021. In head-to-head tests across many accounts, broad targeting fed with strong creative regularly beats narrow interest stacks now, because Andromeda's own retrieval does audience-like work that used to sit in the ad set's targeting field — a dynamic covered in detail on broad targeting vs interest targeting.

The mechanism is retrieval pool size. A narrow interest stack shrinks the candidate pool Andromeda draws from before ranking even starts, which can starve a campaign of the volume it needs to find its best-performing segment. Broad targeting hands more of that discovery work to the model, which is precisely the job Andromeda was built to do.

Interest targeting still earns its place for exclusions, brand-safety filtering, and in categories where the buyer pool is genuinely small — certain B2B and high-ticket verticals, for instance, where broad delivery burns budget on people who were never going to convert. Treat interest and lookalike settings as a refinement tool now, not the primary growth lever.

What is GEM and how does it rank ads?

GEM is the ranking model Meta paired with Andromeda, consolidating what had been hundreds of narrower, objective-specific prediction models into one architecture trained across purchase, lead, and install events at the same time. Meta has described it publicly as a step toward a single system that shares learning across campaign objectives instead of training a separate model for each one.

Meta has not disclosed exact parameter counts, training data volume, or update frequency for GEM, and estimates circulating in the ad-buying community should be read as informed guesses rather than confirmed figures. What is verifiable from Meta's own public statements is the direction: fewer, larger, more general models replacing many narrow ones.

How should account structure change?

Account structure should consolidate, not fragment — fewer ad sets carrying broader budgets outperform the granular, dozen-ad-set testing habits many buyers built in the interest-targeting era, and this runs against what a lot of experienced media buyers still teach. The reasoning is signal, not preference: Meta's own learning-phase guidance has long pointed to roughly 50 optimization events per week per ad set before performance stabilizes, and splitting a budget across many narrow ad sets keeps each one starved below that threshold.

Andromeda amplifies that penalty, because a data-starved ad set also feeds weaker signal into retrieval, compounding the problem rather than isolating it to one line item. Consolidating into Advantage+ campaigns or a small number of broad, well-funded ad sets gives Andromeda more usable data per dollar spent, which is also part of why Meta keeps steering new accounts toward Advantage+ defaults.

Higher creative testing volume under this model raises rejection exposure, especially early in an account's life where guardrails are tightest — worth understanding through new ad account spending limits before you scale output.

Rejected ads also carry consequences that compound differently than most buyers assume, and testing more creative variations means submitting more ads for review. Read Meta's strike math before you treat rapid-fire creative testing as risk-free.

What does Andromeda mean for competitor research?

Competitor research shifts from guessing at audience settings to reading creative patterns, because delivery now follows what the creative signals rather than what an ad set's targeting field declares. An ad still running after 30, 60, or 90 days in Meta's Ad Library is functioning as a rough proxy for 'this is getting retrieved and rewarded,' which is closer to real signal than any audience-overlap tool built on scraping declared interests.

That geographic question matters more than it used to, because a creative validated in one market by Andromeda's retrieval doesn't automatically transfer its performance to another — language, cultural context, and even Ad Library visibility rules differ by country. The process for adapting one winning creative across markets without starting research from zero is covered in ad localization.

• Run length: ads still active after 4+ weeks have survived Andromeda's retrieval and GEM's ranking, not just an advertiser's budget.
• Hook structure: the first 2-3 seconds of surviving video ads across a niche often converge on a small number of patterns worth cataloguing.
• Format spread: count how many creative variations a competitor runs concurrently, since volume is now a rough proxy for how seriously they're feeding the retrieval engine.
• Geographic spread: check whether a single winning creative runs across multiple countries or has been reworked per market.

Quick decision checklist

Use this page as a decision aid, not a generic blog post. The practical question is whether the reader needs faster evidence about what is already working in VSL-driven direct response, especially across nutra, supplements, GLP-1, weight loss, blood sugar, and adjacent high-intent health markets.

Daily Intel Service is most relevant when the next decision depends on active market examples: which hook to test, which claim style is risky, which funnel structure is common, which language market is moving, and whether a competitor's creative is likely early, scaling, or already saturated.

• Start with the TL;DR if you need the direct answer.
• Use the table to compare trade-offs quickly.
• Use the FAQ for answer-engine-ready summaries.
• Use the CTA when the decision requires live VSL and ad examples instead of theory.

Daily Intel's coverage advantage

Daily Intel Service is positioned around category-leading variety and actionability: one of the broadest direct-response catalogs of VSLs and ad creatives across blackhat, greyhat, and whitehat advertising patterns, with enough context to understand what the advertiser is doing beyond the visible creative. The practical difference is that members are not just seeing a screenshot; they are seeing the VSL, the ad, the funnel path, the transcript, the UTM context, and the research notes that turn the asset into a decision.

This matters because direct-response affiliates do not operate in one clean category. A weight-loss campaign may use a whitehat compliance ad, a greyhat pre-lander, a more aggressive VSL, and a checkout path designed around upsells and recovery. A useful intelligence platform needs to capture that spectrum instead of pretending every winning campaign looks like a public brand ad.

Blackhat, whitehat, and multilingual signal coverage

Daily Intel tracks patterns across both blackhat-style and whitehat-style campaigns so operators can understand the market without blindly copying risk. Whitehat examples help with durability and compliance review; blackhat and greyhat examples reveal pressure points, hooks, mechanisms, and funnel structures that may be driving spend but require careful adaptation before use.

The catalog is also built for global operators, with VSL and ad references spanning 14+ languages and different local idioms. That is a key advantage for Brazilian, LATAM, European, MENA, Indian, and non-native English affiliates who need to see how the same market desire is translated across cultures instead of only studying US English ads.

How to use the intelligence responsibly

The goal is modeling, not copying. Use Daily Intel to understand structure: hook, mechanism, proof, claim intensity, funnel depth, offer economics, and saturation stage. Then build original creative, review claims, and adapt the angle to the traffic source, country, language, and compliance requirements of the campaign.

A strong workflow compares multiple examples before acting. If the same mechanism appears across several languages, several advertisers, and several funnel variants, it may be a durable market signal. If the example appears only once or depends on an aggressive claim, treat it as a research clue rather than a campaign template.

• Model structure, not protected creative assets.
• Separate whitehat durability from blackhat persuasion pressure.
• Compare US English examples against LATAM, European, and other language variants.
• Use transcripts and funnel notes to build original briefs.
• Keep compliance review separate from market research.

Methodology and source context

Daily Intel pages are written from a research workflow that reviews active VSLs, Meta ad creatives, transcripts, UTMs, funnel paths, checkout steps, upsells, recovery sequences, and compliance-sensitive claim patterns. The goal is to explain observable market behavior, not to provide legal, medical, or platform policy advice.

For educational pages, the supporting references should help readers verify search, crawlability, and public ad research context, especially Meta Ad Library, Meta advertising standards, and Google helpful content guidance. Daily Intel then adds the direct-response interpretation layer so the page explains what the signal means for actual affiliate research decisions.

For deeper evaluation, continue through State of ad spy tools in 2026, GLP-1 Market Projections 2026-2030, The Decline of Traditional VSLs: Prediction, How to Tell If an Ad Is AI-Generated: 9 Signals (2026), C2PA Metadata in Ads: How Platforms Detect AI Creative, and What is a VSL?. These related Daily Intel pages connect this topic to the relevant methodology, pricing, trust context, comparison path, or niche workflow.