If your Facebook and Instagram ads feel different from a few years ago, there is a reason. Meta has been rebuilding large parts of the artificial intelligence system that decides which ads are relevant enough to show to each person, and Meta Andromeda is an important part of that change.
Meta Andromeda is Meta’s AI-powered ads retrieval system. It sits near the beginning of Meta’s multi-stage ad recommendation process, narrowing tens of millions of eligible ads into a few thousand relevant candidates before more sophisticated ranking systems determine which ads are most valuable to show.
That definition matters because Andromeda is often described online as a new Facebook Ads algorithm that replaced audience targeting, made interests obsolete or turned creative into the only targeting mechanism that matters. The reality is more nuanced.
Andromeda is not a new campaign type. There is no Andromeda toggle inside Ads Manager. It does not replace Meta’s entire advertising system. And Meta has not said advertisers should abandon every audience control or upload a fixed number of creatives.
What Andromeda does represent is a major change in Meta’s ability to process, understand and retrieve relevant ads at enormous scale. For advertisers, that has important consequences for creative diversity, automation, campaign structure, targeting and the quality of signals being sent back to Meta.
This guide explains exactly what Meta Andromeda is, how it works, why Meta built it, what it changes for advertisers and how to adapt your Meta Ads strategy in 2026. For the technical source, read Meta’s official Andromeda engineering overview.
What is Meta Andromeda?
Meta Andromeda is a proprietary machine-learning system designed specifically for ads retrieval.
Retrieval is the first major filtering stage in Meta’s multi-stage ad recommendation system. According to Meta, this stage takes tens of millions of possible ad candidates and reduces them to a few thousand relevant candidates. Later and more sophisticated ranking models then estimate the value of those candidates for both the person seeing the ad and the advertiser.
In simpler terms, imagine Meta has millions of ads it could potentially show someone. It cannot run its most computationally expensive prediction models across every eligible ad each time someone opens Facebook or Instagram. Instead, it needs an extremely fast system capable of identifying a much smaller collection of promising ads.
That is where Andromeda comes in.
| Stage | What happens |
|---|---|
| Eligible ads | Meta starts with a very large pool of ads that could potentially be considered. |
| Andromeda retrieval | Andromeda identifies a much smaller group of ads that appear relevant to the individual and context. |
| Advanced ranking | More sophisticated recommendation models estimate likely value and relevance. |
| Delivery decisions | Meta’s broader ads system considers ranking, objectives, bids and other signals before determining what is ultimately shown. |
This distinction between retrieval and ranking is one of the most important concepts to understand about Meta Andromeda. Andromeda helps decide which ads deserve further consideration. It is not, by itself, the entire system responsible for deciding the final advertisement you see.
When did the Meta Andromeda update happen?
Meta publicly introduced the technical details of Andromeda on December 2, 2024 in a Meta Engineering article titled Meta Andromeda: Supercharging Advantage+ automation with the next-gen personalized ads retrieval engine.
It is better to think of this as an ongoing infrastructure evolution rather than a single Ads Manager update that appeared overnight.
Meta has continued developing other parts of its advertising recommendation infrastructure since then, including sequence-learning models, large-scale ranking systems and its Generative Ads Recommendation Model, or GEM. In other words, Andromeda is an important piece of Meta’s broader movement toward increasingly AI-driven ad recommendations, not the entirety of that transformation.
Why did Meta build Andromeda?
The simple answer is scale.
Meta’s retrieval system has an unusually difficult job. Meta says retrieval processes roughly three orders of magnitude more ads than subsequent recommendation stages. At the same time, the number of potential ads entering the system has been growing rapidly.
Several developments created this problem. Meta’s Advantage+ products automate more parts of campaign management. Advertisers can produce more asset combinations. Generative AI makes creating images and other advertising assets easier. Automated audience expansion creates additional possibilities for matching ads with people.
Meta reported in its original Andromeda announcement that more than one million advertisers had used its generative AI tools to produce more than 15 million ads in a single month. That statistic was not presented as an Andromeda performance result. It illustrated the creative-volume problem Meta’s retrieval infrastructure needs to handle.
More creatives can theoretically create better opportunities for personalisation, but only if Meta can efficiently identify which creative is relevant to which person. That is the problem Andromeda was designed to solve.
How does Meta Andromeda work?
The engineering behind Andromeda is considerably more complicated than anything advertisers need to understand to run campaigns effectively. However, a few components explain why the system is important.
At the core of Andromeda is a large deep neural network specifically designed for high-volume ad retrieval. Meta combined changes in machine-learning architecture with specialised hardware, new feature representations, hierarchical indexing and new inference methods.
1. A much larger retrieval model
Meta says Andromeda enabled a 10,000x increase in retrieval model capacity.
That does not mean advertisers receive 10,000 times better results. Model capacity describes the technical capability of Meta’s retrieval model, not an advertiser performance multiplier.
The significance is that Andromeda can model more complicated relationships between people, their interests and the products or services represented by ads. Meta says the system uses advanced interaction features and algorithms to capture latent relationships between people, interests, products and services.
2. Hierarchical indexing
One of Andromeda’s key innovations is a hierarchical ad index.
Instead of evaluating every possible advertisement in the same way, the system organises ads into multiple levels. It can progressively focus its computation on increasingly relevant parts of the index.
You can think of it like searching a huge library. An inefficient approach would examine every book individually whenever someone asked a question. A hierarchical system first identifies the relevant section, then the relevant shelf, then a much smaller group of books worth examining closely.
Andromeda applies a far more sophisticated machine-learning version of that idea to advertisements. Meta says this hierarchical architecture gives its retrieval models sub-linear inference costs and helps the system handle much larger volumes of advertising creative while maintaining retrieval accuracy.
3. Hardware designed for AI workloads
Andromeda was also built alongside specialised computing infrastructure. Meta specifically discusses NVIDIA’s Grace Hopper Superchip and its own Meta Training and Inference Accelerator, commonly called MTIA.
This matters because modern recommendation systems are not only constrained by the quality of their algorithms. They also need to make predictions extremely quickly while processing enormous amounts of data. An advertising recommendation that takes too long to calculate is not useful when someone is scrolling through Instagram.
Meta designed Andromeda’s machine-learning models, indexing system and hardware architecture together to increase both scale and speed.
4. Model elasticity
Not every advertising request requires exactly the same amount of computational effort. Andromeda uses what Meta describes as a segment-aware design, allowing different levels of model complexity to be applied depending on the situation and available resources.
Meta says this model elasticity, combined with its hierarchical neural network, improved model inference efficiency by 10x. Again, that is an infrastructure metric rather than a promise that an advertiser will receive 10x more efficient campaigns.
What performance improvements has Meta reported from Andromeda?
Meta reported several improvements following the deployment of Andromeda across Facebook and Instagram.
| Meta-reported metric | Improvement |
|---|---|
| Retrieval recall | +6% |
| Ad quality on selected segments | +8% |
Meta also reported more technical improvements, including more than a 100x gain in feature-extraction latency and throughput compared with previous CPU-based components and more than a 3x increase in end-to-end model inference queries per second.
These numbers require context. An 8% ad-quality improvement in selected segments does not mean every advertiser received an 8% improvement in ROAS. Similarly, the 6% retrieval-recall improvement is a measure of the retrieval system’s ability to identify useful candidates, not an advertiser conversion-rate guarantee.
How was Meta’s ad retrieval system different before Andromeda?
Before Andromeda, Meta says its retrieval architecture had more separate stages and relied on numerous rule-based heuristics. Those systems could still personalise advertising, but scaling model complexity across a huge number of potential ads was difficult.
| Before Andromeda | With Andromeda |
|---|---|
| More isolated retrieval components | More integrated ML-driven retrieval |
| Greater reliance on rule-based heuristics | Reduced reliance on manually defined rules |
| Lower retrieval-model capacity | Significantly larger model capacity |
| Harder to scale as creative volume increased | Hierarchical indexing built for larger creative pools |
| More infrastructure constraints | Greater AI hardware utilisation |
| More limited retrieval personalisation | Richer modelling of person-ad relationships |
Is Meta Andromeda the same as the Facebook Ads algorithm?
Not exactly.
Calling Andromeda an “algorithm” is understandable and matches the language many advertisers use when searching for the topic. However, Meta technically describes Andromeda as a machine-learning system for ad retrieval.
It is one important component inside a much larger advertising recommendation and delivery system. Meta Ads still involves campaign eligibility, optimisation objectives, retrieval, ranking, auctions, predictions, bids, creative, budgets, placements, conversion signals and other mechanisms.
Andromeda does not replace all of those processes.
Meta Andromeda vs GEM: what is the difference?
This distinction has become especially important as Meta has expanded its advertising AI stack. Andromeda and GEM are related to Meta’s broader ad-recommendation infrastructure, but they are not the same technology.
| Meta Andromeda | Meta GEM | |
|---|---|---|
| Primary role | Ads retrieval | Foundation model for ad recommendations |
| Main purpose | Reduce tens of millions of eligible ads to a smaller relevant candidate set | Learn broadly from Meta’s advertising data and improve recommendation intelligence across the ads stack |
| Advertiser-facing product? | No | No |
| Ads Manager toggle? | No | No |
The simplest way to remember the difference is this: Andromeda is specifically about retrieving promising ads from an enormous candidate pool. GEM is a broader ads recommendation foundation model intended to improve intelligence across recommendation systems.
You can read more in Meta’s official GEM engineering overview.
Is Meta Andromeda the same as Advantage+?
No.
Advantage+ is a collection of advertiser-facing automation features and campaign capabilities. Andromeda is underlying advertising infrastructure.
Advertisers interact with Advantage+. Advertisers do not directly interact with Andromeda.
Meta connects the two because Advantage+ automation can dramatically increase the number of possible combinations and eligible creative options its advertising system needs to process. Andromeda gives Meta a more scalable retrieval system for handling that increasing complexity.
Our broader guide to AI performance marketing looks at the same shift across paid media, where increasingly powerful automation makes the quality of the inputs marketers control more important rather than less important.
What does Meta Andromeda change for advertisers?
This is where official documentation ends and practical interpretation begins.
Meta has clearly built Andromeda to retrieve relevant ads from much larger and more diverse candidate pools. It has also linked the system to the rapid expansion of Advantage+ automation and AI-generated creative.
From that, several important implications follow. Advertisers have less reason to spend all of their effort trying to manually engineer tiny audience segments if Meta can increasingly identify relevant ads algorithmically. At the same time, creative inputs become more important because giving Meta meaningfully different messages gives its recommendation systems more options.
But this does not mean audience strategy no longer matters, nor does it mean every advertiser should copy the same broad-targeting campaign structure.
The most useful way to think about the change is: advertisers increasingly control the quality and diversity of the inputs, while Meta increasingly controls how those inputs are matched and distributed.
Is “creative is the new targeting” actually true?
“Creative is the new targeting” has become one of the most common ways advertisers explain Meta Andromeda. There is some truth behind the idea, but the phrase can easily be taken too far.
Meta’s official technical explanation says Andromeda models complex relationships between people, interests, products and services represented through ads. It also explicitly says the system is designed to support the rapidly increasing volume of creative entering Meta’s recommendation system.
What Meta does not say in the original Andromeda documentation is that advertiser-defined audiences have become completely irrelevant.
A better interpretation is: creative is becoming a stronger input into Meta’s ability to find relevant people, so the creative brief increasingly performs part of the job that advertisers once tried to accomplish mainly through audience segmentation.
For example, a SaaS company could create genuinely different concepts for a finance leader focused on reducing operational costs, a marketing leader focused on attribution, a sales leader focused on shortening the path from lead to opportunity and a RevOps leader focused on connecting fragmented CRM and advertising data.
Even if those ads eventually sit inside a broader audience, each communicates a clearer problem, persona and use case. That is meaningful creative diversification. Changing the background colour of the same advertisement ten times is not.
What does creative diversification mean after Andromeda?
One of the biggest mistakes advertisers can make is interpreting creative diversification as simply producing a larger number of ads. More ads are useful only when they create meaningfully different possibilities for Meta to test and match.
| Dimension | Example |
|---|---|
| Persona | Founder vs CMO vs Head of Growth |
| Problem | High CAC vs low lead quality vs poor attribution |
| Awareness | Problem-aware vs solution-aware vs product-aware |
| Value proposition | Save money vs save time vs improve revenue visibility |
| Proof | Customer testimonial vs data point vs product demonstration |
| Format | Static image vs short video vs carousel vs founder-led video |
| Offer | Guide vs webinar vs demo vs free audit |
| Use case | Prospecting vs retargeting vs reactivation |
The goal is not to manufacture arbitrary variation. The goal is to give Meta’s recommendation system genuinely different messages that may resonate with different people and buying contexts.
Our Facebook Ads for SaaS guide goes deeper into how creative, first-party data, targeting and campaign structure work together for SaaS acquisition.
How many Meta Ads creatives should you run after Andromeda?
There is no universal Meta rule saying that every advertiser should run 10, 20, 30 or 50 ads because of Andromeda.
Meta has clearly designed Andromeda to handle an expanding number of ad creatives. That does not mean simply uploading more ads automatically improves results.
An account spending $50 per day should not automatically use the same active creative count as one spending $50,000 per day. Creative quantity needs to reflect budget, audience size, conversion volume, testing methodology and how quickly individual assets accumulate useful data.
What matters more is concept diversity. Five fundamentally different creative concepts can provide more useful information than twenty near-identical ads with minor headline or colour changes.
Does Meta Andromeda make interest targeting obsolete?
No.
There is not enough evidence to support the blanket statement that Andromeda makes interest targeting obsolete. Meta has certainly moved toward more automated advertising, broader audience expansion and increasingly sophisticated recommendation systems. That makes overly restrictive audience structures less attractive in many situations.
But that is not the same as saying interest targeting can never be useful. The right targeting approach still depends on the advertiser, market, geography, conversion volume, objective, offer and quality of first-party data.
Advertisers should therefore test whether broader targeting improves performance rather than treating it as an ideological rule.
Do lookalike audiences still work after Meta Andromeda?
Andromeda does not automatically make lookalike audiences useless either.
What has changed is that advertisers should be more critical about whether additional audience constraints are genuinely helping Meta find better prospects or simply restricting its opportunity to optimise.
A high-quality lookalike based on valuable customers may still contain useful information. A lookalike created from every low-intent lead that ever completed a form may be considerably less useful.
Does broad targeting work better with Meta Andromeda?
Broader targeting can give Meta’s machine-learning systems a larger opportunity space in which to find people likely to complete an advertiser’s objective. That makes broad audiences increasingly attractive in many accounts.
But “broad” should not become a substitute for strategy. Advertisers may still need geographic restrictions, age restrictions, customer exclusions, regulatory controls or other constraints. Retargeting and first-party audiences can also serve very different purposes from pure broad prospecting.
The better question is not “Should every advertiser use broad targeting?” It is: Which targeting controls are genuinely necessary, and which ones are restricting Meta without improving business outcomes?
Should Meta Ads campaign structures become simpler?
In many accounts, yes, but again, there is no universal Andromeda campaign template.
Fragmenting a limited budget across large numbers of campaigns and ad sets can create several problems. Each segment receives less spend, learns more slowly and produces fewer conversion signals. A more consolidated structure can give Meta greater freedom to optimise across the available opportunity.
However, advertisers still need separation when there is a real strategic reason for it. Different countries, fundamentally different products, dramatically different funnel stages or different business objectives can justify separate structures.
The goal should therefore be intentional consolidation, not consolidation for its own sake.
Why conversion signals matter even more in an AI-driven Meta Ads system
As Meta automates more campaign decisions, advertisers need to pay more attention to what they are asking Meta to optimise for.
This is particularly important for lead-generation advertisers. Suppose your campaign sends Meta a “Lead” conversion whenever someone completes a form. Meta’s broader optimisation system will attempt to find more people likely to complete that event.
But your business may not actually want more form submissions. It wants more qualified opportunities and customers.
If most of those leads are students, competitors, vendors or people outside your ideal customer profile, an impressive cost per lead can hide poor business performance. Andromeda does not magically solve that measurement problem.
Advertisers still need better first-party tracking, Meta Pixel implementation, Conversions API where appropriate, CRM integration and downstream measurement. The smarter Meta becomes at achieving the objective you give it, the more important it becomes to give Meta the right objective and the right signals.
What does Meta Andromeda mean for B2B advertisers?
B2B advertising is where many simplistic Andromeda recommendations start to break down.
A direct-to-consumer company may generate thousands of purchases every month. Meta receives a relatively clear stream of feedback about which advertising interactions lead to revenue. A B2B SaaS company might generate fifty demo requests, fifteen qualified opportunities and three closed customers over the same period.
The actual purchase may happen weeks or months after the first ad interaction. That makes creative and conversion architecture especially important.
A B2B advertiser should think about creative as a mechanism for communicating who a message is intended for. An ad aimed at CFOs should communicate a CFO problem. An ad aimed at RevOps should communicate a RevOps problem. A generic “grow your business” ad creates much weaker differentiation.
B2B companies also need to consider which downstream events can realistically be returned to advertising platforms. CRM-qualified leads, opportunities, demos and other business events may provide much richer measurement than raw lead volume.
This deserves its own article, which is why we should treat Meta Andromeda for B2B Ads as a separate supporting guide rather than turning this broad explainer into a B2B-only article.
If you are already running paid social, OneMetrik’s Meta Ads agency services focus on creative testing, audience strategy, retargeting, tracking and CRM-connected optimisation rather than treating cheap leads as the final KPI.
What does Andromeda mean for retargeting?
Retargeting is not suddenly obsolete because Meta has better retrieval models.
People who have visited your website, interacted with your content, watched your videos, evaluated your product or entered your sales process represent different contexts from a completely new prospect.
The strategic question is whether those audiences are large enough and distinct enough to justify dedicated campaigns or whether Meta can efficiently handle some of that sequencing through broader automated structures.
For high-consideration purchases, retargeting can still be valuable because the creative objective changes. A prospecting ad might introduce the problem. A mid-funnel ad might provide a case study. A bottom-funnel ad might address objections, show the product, compare alternatives or present a demo offer.
What does Meta Andromeda mean for creative fatigue?
Andromeda does not eliminate creative fatigue. If anything, greater personalisation makes a strong creative-production system more important.
The solution is not to refresh every advertisement every seven days according to an arbitrary calendar. It is to watch actual performance signals and build a reliable pipeline of new concepts before the current winners deteriorate.
Advertisers should distinguish between creative iteration and creative diversification. Iteration improves an existing concept. Diversification creates a genuinely different concept. A mature Meta Ads programme needs both.
How does Meta Andromeda affect Advantage+ and AI-generated ads?
Andromeda was specifically designed in the context of Meta’s increasing use of Advantage+ automation and generative AI advertising tools.
More automation means more eligible combinations. Generative AI means advertisers can create larger creative libraries faster. Andromeda gives Meta infrastructure capable of searching those much larger spaces for relevant ads.
The strategic takeaway is straightforward: the cost of creating variations is falling, so the competitive advantage shifts toward knowing which variations are worth creating.
Common Meta Andromeda myths
| Claim | More accurate interpretation |
|---|---|
| “Andromeda replaced Meta’s advertising algorithm.” | Andromeda is specifically an ads retrieval system inside a much broader recommendation and delivery stack. |
| “Audience targeting is dead.” | Meta is increasingly automated, but advertiser controls, exclusions, first-party data and audience strategy can still matter. |
| “Interest targeting no longer works.” | There is no universal Meta rule saying interests are obsolete. Test them against broader structures. |
| “Lookalikes are dead.” | Lookalikes remain available. Their usefulness depends partly on source quality and account context. |
| “Creative is the only targeting.” | Creative increasingly contributes to relevance and personalisation, but it is one input within a much larger system. |
| “You need 20 ads per ad set.” | Meta does not prescribe one universal Andromeda creative count. Meaningful diversity matters more than arbitrary volume. |
| “Andromeda and Advantage+ are the same.” | Advantage+ is advertiser-facing automation. Andromeda is underlying retrieval infrastructure. |
| “Andromeda and GEM are the same.” | They are different components of Meta’s evolving AI advertising stack. |
| “There is an Andromeda setting in Ads Manager.” | There is no advertiser-facing Andromeda toggle. |
How advertisers should adapt to Meta Andromeda in 2026
- Build genuinely different creative concepts. Develop different personas, problems, hooks, proof points, offers, formats and stages of awareness rather than twenty cosmetic versions of one ad.
- Question unnecessary audience fragmentation. If multiple ad sets are pursuing essentially the same customer and business objective, test whether consolidation gives Meta better data and more room to optimise.
- Improve your conversion signals. Make sure Meta is not being rewarded for generating actions that look good in Ads Manager but have little business value.
- Use automation deliberately. Advantage+ and broad targeting can be powerful, but automation works best when the strategy, creative and measurement inputs are strong.
- Test broad targeting rather than assuming it is always better. Remove restrictions that do not improve performance, but retain controls that have legitimate strategic, geographical or regulatory purposes.
- Measure concept-level creative performance. Understand which problems, personas, messages and offers are producing results.
- Separate Meta-confirmed changes from industry folklore. Use advertiser experiments to validate claims about campaign structure instead of blindly applying another account’s Andromeda playbook.
- Optimise for business outcomes. The ultimate objective is not to make Andromeda happy. It is to use Meta’s increasingly sophisticated advertising system to create profitable customer acquisition and measurable business growth.
What should advertisers not change because of Andromeda?
Do not dismantle a profitable Meta Ads account simply because a new framework becomes popular on LinkedIn.
If an existing targeting method continues to generate profitable customers, retain it until controlled testing demonstrates that another approach works better. Do not dramatically multiply your creative count without sufficient budget to test it. Do not remove valuable first-party audiences simply because someone says broad is mandatory.
And do not assume that every decline in Facebook or Instagram performance is caused by Andromeda. Meta Ads performance still depends on the market, competition, offer, creative, landing page, tracking, conversion objective, product, seasonality and dozens of other factors.
Andromeda changes the infrastructure underneath advertising. It does not repeal the fundamentals of good marketing.
What comes after Meta Andromeda?
Andromeda should not be viewed as the final version of Meta’s ad recommendation technology.
Meta explicitly said its future Andromeda architecture could move toward an autoregressive loss function designed to retrieve a more diverse collection of candidate ads. The company also said future hardware integrations could enable another significant increase in model complexity.
The direction is clear: Meta is moving toward advertising systems where machine learning does more of the retrieval, prediction, recommendation and optimisation work.
That does not eliminate marketers. It changes where marketers create leverage. Audience micromanagement becomes less important in some situations. Creative strategy becomes more important. Signal quality becomes more important. Measurement becomes more important. Offers and landing pages remain important.
The bottom line on Meta Andromeda
Meta Andromeda is Meta’s AI-powered ad retrieval system, designed to identify a few thousand relevant ads from tens of millions of possible candidates before later recommendation systems determine which ads should ultimately be shown.
Its significance extends beyond a technical infrastructure upgrade. Andromeda allows Meta to process a much larger and more diverse universe of advertising creative while applying substantially more sophisticated personalisation at the retrieval stage.
For advertisers, that means the old playbook of endlessly slicing audiences while running a small collection of generic ads is becoming less reliable.
The better approach is to provide Meta with stronger inputs: genuinely diverse creative concepts, clear positioning, useful conversion signals, sensible campaign structures and measurement that connects paid media to actual business outcomes.
Andromeda does not mean targeting is dead. It means the quality of the advertising inputs you give Meta matters more as Meta takes greater control over how those inputs are distributed.
For companies that want to apply these principles to actual Facebook and Instagram campaigns, explore OneMetrik’s Meta Ads agency services.
Frequently Asked Questions About Meta Andromeda
What is Meta Andromeda?
Meta Andromeda is Meta’s machine-learning system for ads retrieval. It helps narrow tens of millions of eligible advertisements into a few thousand potentially relevant candidates before later ranking and recommendation systems decide which ads are most valuable to show.
What is Andromeda in Meta Ads?
In Meta Ads, Andromeda operates behind the scenes as part of Meta’s advertising recommendation infrastructure. Advertisers do not activate it manually. Its job is to improve the retrieval stage by finding relevant ad candidates efficiently from Meta’s enormous pool of eligible advertisements.
What is the Meta Andromeda update?
The term Meta Andromeda update generally refers to Meta’s move to its next-generation personalised ads retrieval architecture. Meta publicly detailed Andromeda in December 2024 and described improvements in model capacity, personalisation, retrieval recall, ad quality and infrastructure efficiency.
Is Meta Andromeda an algorithm?
Advertisers commonly call Andromeda an algorithm, but Meta more precisely describes it as a proprietary machine-learning system for ads retrieval. It combines neural networks, hierarchical indexing, machine-learning algorithms and specialised computing infrastructure.
Is Meta Andromeda the same as Advantage+?
No. Advantage+ refers to advertiser-facing automation products and features within Meta Ads. Andromeda is underlying machine-learning infrastructure used during the retrieval stage of Meta’s advertising recommendation system.
Is Meta Andromeda the same as GEM?
No. Andromeda is specifically an ads retrieval engine. GEM is Meta’s Generative Ads Recommendation Model, a large foundation model designed to improve recommendation intelligence across Meta’s advertising stack.
Does Meta Andromeda replace audience targeting?
No. Andromeda increases Meta’s ability to retrieve relevant advertisements and supports greater automation, but it does not mean audience targeting has completely disappeared. Advertisers can still use geographic controls, exclusions, custom audiences, lookalikes and other targeting options where they are strategically useful.
Does Meta Andromeda mean creative is the new targeting?
Creative is the new targeting is useful shorthand, but it oversimplifies Andromeda. Better creative diversity gives Meta more meaningful advertising options and signals to work with, but creative does not literally replace every targeting, optimisation and recommendation mechanism inside Meta Ads.
How many ads should I run with Meta Andromeda?
Meta does not specify one universal number of ads every advertiser should run because of Andromeda. The appropriate creative count depends on budget, audience size, conversion volume and testing capacity. Meaningfully different concepts generally matter more than producing a large number of near-identical variations.
Do I need to turn on Meta Andromeda?
No. There is no Andromeda setting or toggle in Meta Ads Manager. Andromeda operates as part of Meta’s advertising infrastructure behind the scenes.