The Meta Andromeda Update: What the New Ads Engine Changes for Your Campaigns (2026)

Meta Andromeda: what it is — and why it affects every Meta ads campaign

When the performance-marketing world talks about “Andromeda” right now, there is a lot of confusion. So let us start with a clear framing — because nothing hurts an advertising strategy more than measures built on false assumptions.

Important upfront, as of June 2026: “Andromeda” is a genuinely named system at Meta. Meta announced it officially via its engineering blog on 2 December 2024. It refers to the new, AI-powered ads retrieval engine — the stage at which Meta selects, from a vast pool of ads, the few thousand candidates that even make it into the auction. Over the course of 2025, Andromeda was rolled out step by step across Facebook, Instagram and Messenger; by October 2025 it was globally active across most campaign objectives and placements. Since early 2026 at the latest, it is the default behaviour behind your campaigns.

One important distinction right away: “Andromeda” has nothing to do with Google. In the SEO scene the same name circulates as an umbrella term for the AI rebuild of Google Search — that is purely an industry coinage with no official Google confirmation. The Andromeda covered here is the real, Meta-named advertising system. Conflating the two means optimising into the void.

Our aim in this guide: afterwards you should not only understand the buzzword, but know which of these shifts concretely affect your campaigns — and which practical steps follow from them. As a Meta Ads agency we navigate exactly this transition every day. We flag every figure Meta states explicitly as “according to Meta,” because it comes from Meta’s own communication and is not independently verifiable.

The short version in four sentences

Andromeda is the AI engine that decides, before the auction, which of your ads even get a chance to be delivered. In doing so it reads the creative itself heavily — image, video, text and format — rather than relying primarily on manually set audiences. This shifts the most important lever in the account from audience selection to creative diversity. Those who understand this and rebuild their campaigns accordingly give the engine better signals — and get more from the same budget.

Timeline: how Andromeda was rolled out

For orientation, the publicly communicated sequence:

• 2 December 2024 – official announcement: Meta introduces Andromeda via its engineering blog, described as a “next-gen personalized ads retrieval engine” to strengthen Advantage+ automation.
• Early/mid 2025 – gradual rollout: Andromeda is rolled out across campaign objectives and placements; advertisers observe that broad delivery increasingly beats tightly defined audiences and that the creative explains more of the performance variance.
• October 2025 – broad availability: by industry observation, Andromeda is globally active across most objectives and placements.
• Q4 2025 to early 2026 – default behaviour: in e-commerce and lead-gen accounts the creative-first logic is the default. Those still thinking in old structures leave performance on the table.

It is precisely this convergence — new retrieval architecture plus Advantage+ automation plus generative creative tools — that makes Meta ads in 2026 fundamentally different from 2023.

How the ads retrieval engine works technically

To understand why so much is changing about your creative and structure work, you need to understand where in the pipeline Andromeda actually operates. The decisive break with classic delivery is not in the auction itself, but before it — in how ad candidates are assembled in the first place.

The multi-stage Meta ads pipeline

When a user opens Facebook or Instagram, a multi-stage selection runs in milliseconds. Simplified, there are three phases: retrieval (filtering the relevant candidates out of a vast pool), ranking (scoring the candidates by expected value) and auction (deciding who gets the slot and at what price). Andromeda is the engine for the first phase — retrieval. It filters, according to Meta, “tens of millions” of ad candidates down to a few thousand that then proceed to ranking.

That sounds like a technical detail but is strategically decisive: what never makes it through retrieval cannot win in even the best ranking and the most expensive auction. If your creative is not even selected as a candidate for the relevant users, it is effectively invisible — no matter how high your bid.

Why Meta rebuilt the retrieval stage from scratch

The background is a volume problem. Through Advantage+ and generative AI tools the number of ads has exploded: according to Meta, over one million advertisers created more than 15 million ads per month using generative AI tools. The old retrieval stage could no longer keep pace with this exponentially growing volume of ads under strict latency requirements. Andromeda was built to resolve exactly this pressure — scoring more candidates more precisely without delivery becoming slower.

A deep neural network on specialised hardware

Technically, Andromeda is a custom-built deep neural network optimised for modern AI hardware. According to Meta, it runs on the NVIDIA Grace Hopper Superchip and Meta’s own inference accelerator, MTIA. This hardware combines CPU and GPU capabilities with very high memory bandwidth, which eases the classic scaling problem — the memory bottleneck. Meta states this enabled a 10,000x increase in model capacity (figure according to Meta). Simplified: per request the engine can account for vastly more features and relationships than its predecessor.

Another technical building block is, according to Meta, a hierarchical index trained jointly with the retrieval models and allowing “sub-linear” inference costs — so the system does not slow down proportionally as the ad volume grows. There is also a “model elasticity” that adjusts compute complexity to available resources. For you as an advertiser, the mechanics under the hood matter less than the consequence: the engine has become good enough to learn from the creative itself who it is relevant for.

Which improvements Meta states

Meta quantifies the effects with several figures. We report them explicitly as Meta’s own claims:

• +6% recall in the retrieval system (according to Meta) — the engine finds more of the actually relevant ad candidates.
• +8% ad quality on selected segments (according to Meta).
• Over 100x improvement in feature-extraction latency (according to Meta).
• A 3x increase in end-to-end inference throughput (queries per second, according to Meta).

These numbers are Meta’s own measurements from Meta’s own environment — treat them as an order of magnitude and a directional statement, not a guaranteed result for your account. The strategic message behind them, however, is undisputed: the retrieval stage has become significantly more capable, and that is exactly why what successful campaigns depend on is shifting.

What concretely changes for advertisers

The new architecture produces tangible shifts in day-to-day work. Important: not all the old levers are abolished — but their relative weight shifts, sometimes dramatically.

From audience-first to creative-first thinking

The biggest shift can be summed up in one sentence: your creative is now your targeting. For over a decade the craft of media buying lay in finding the right audience — interests, behaviours, lookalikes, custom audiences. Under Andromeda the engine increasingly reads from the creative itself who an ad is relevant for and matches it accordingly. Tightly defined, manually set audiences often act as a brake in this model: they constrain the search space in which the engine could otherwise play to its strength.

This does not mean audience knowledge becomes worthless — on the contrary, it now flows into the creative itself. Instead of setting an audience technically, you translate your knowledge about the audience into imagery, hook and message. Those who understand their customers build the creatives the engine assigns to the right people.

Broad delivery beats narrow targeting

In practice this means: broad targeting (often with only coarse guardrails such as country, language, age) frequently wins under Andromeda over detail-obsessed interest stacks. Classic audience constraints are increasingly treated as a “suggestion” rather than a hard rule. Give the engine room and in many cases it finds more profitable user segments than you could have defined manually.

Performance variance shifts into the creative

Because the engine weights the creative so heavily, the creative now explains a far larger share of the difference between winning and losing campaigns. Two accounts with identical setups but different creative quality and diversity can now be further apart than ever before. That is the uncomfortable but liberating truth of Andromeda: you can no longer disguise weak creative output with clever targeting.

Creative strategy under Andromeda: what counts now

When the creative becomes the most important lever, your creative production has to change — in volume, diversity and method. This is the actual working core of this guide.

Volume and diversity instead of one “perfect” creative

The engine can only match optimally if it has enough different ads to choose from. A single, supposedly perfect ad gives it almost no room. A bundle of many different creatives — different hooks, formats, visual worlds, messages — opens up many points of attachment to different user segments. The rule of thumb shifts from “find the one winning creative” to “supply a broad, diverse creative library and let the engine match.”

Creative diversity along the customer journey

Diversity does not mean “the same motif in ten colours.” It means covering genuine differences in message and approach — and ideally serving different stages of the customer journey:

• Awareness: attention-grabbing hooks, emotional imagery, raising problem awareness.
• Consideration: benefit arguments, comparison, trust-building, handling objections.
• Conversion: a clear offer, urgency, social proof, a concrete call to action.

The wider this spectrum is covered, the more user situations the engine can serve. Those who only run conversion creatives leave the upper funnel stages untapped, from which the engine would otherwise draw cheaper reach.

Formats that deliver the creative volume

To produce enough material economically, a modular mindset pays off: take a strong core concept and turn it into many variants — different hooks in the first three seconds of a video, different text openers, static and motion versions, vertical and square, UGC style alongside high production value. This turns a few ideas into many self-contained creatives without each starting from zero. How we systematise this production workflow we also show in our work as a social media agency in Hamburg.

What makes a good Andromeda creative

Because the engine also “reads” the creative substantively, ads that carry their message clearly win out: a strong hook in the first seconds, an unambiguous core statement, a recognisable offer, clean brand reference. Blurry, generic creatives without a clear message give the engine little signal — and are matched accordingly worse. Clarity beats arbitrariness.

What the engine infers from the creative

It helps to picture what the retrieval engine actually sees in a creative. It evaluates a range of signals: the visual motif and its imagery, the first seconds of a video, the ad copy, the format and aspect ratio, the tone, the product or service being advertised. From this overall picture it derives a hypothesis about which people, in which situation, this creative is likely relevant for — and matches it accordingly. That is why two superficially similar ads can reach completely different audiences: a different hook, a different visual world or a different tone shifts the engine’s hypothesis. In practical terms: every creative is not just a message but also a targeting signal. The clearer and more unambiguous this signal, the more precise the match.

This leads to a productive way of thinking: instead of asking “which audience do I set,” ask “which creative speaks to which situation.” An awareness creative that makes a concrete problem visible in the first two seconds signals a different user situation to the engine than a sober offer creative with price and call to action. Cover both deliberately and you give the engine the breadth to deliver across different phases of the customer journey.

Test systematically instead of guessing

Under Andromeda the logic of testing changes too. The classic A/B test of individual audiences loses importance because the engine handles delivery anyway. In its place comes the systematic testing of creative concepts: which hooks, which visual worlds, which messages and which formats work in your market? A structured approach makes sense, where per test wave you run several clearly distinguishable concept directions against each other — not twenty mini-variants of the same motif, but genuine alternatives in idea and approach. Scale the winning directions by producing further variants in the same vein; replace the losers with new concepts.

Patience with significance matters here: give each concept enough budget and time to deliver robust signals out of the learning phase. Switching a creative off after a few hundred impressions produces noise, not insight. Document your test results in a structured way so that, over the months, a proprietary, data-backed creative map emerges — this becomes a genuine competitive advantage no rival can copy, because it rests on your data.

Generative AI as a production lever — with judgement

One reason for the sheer volume of ads that made Andromeda necessary in the first place is generative AI tools: they drastically lower the production cost per creative variant. That is both opportunity and trap. Opportunity, because it lets you reach the diversity Andromeda needs economically. Trap, because pure mass without an idea quickly leads to interchangeable, arbitrary creatives — and those give the engine weak signals. The productive path is to use generative tools to scale good concepts, not as a replacement for the concept work itself. The human contribution thus shifts from manual single production to strategy, the hook idea and quality control.

Campaign structure and Advantage+

The new engine logic is reflected directly in the recommended account structure. The era of fragmented accounts with dozens of ad sets per audience is ending.

Consolidate instead of fragment

Many small ad sets, each with its own narrow audience, slice the budget and the learning signal into tiny portions — exactly the opposite of what Andromeda needs. The logic proven under Andromeda points toward consolidation: few, broadly delivering ad sets with budget at the campaign level and a large, diverse creative library within them. This way the engine gathers robust signals faster and can optimise more cleanly.

Scaling under Andromeda: calm, not abrupt

Scaling, too, follows a different logic under Andromeda. Those who have found a successful setup tend to crank the budget up fast and hard — which tears the learning phase and often causes a short-term performance drop. The calmer path is usually superior: raise budget in moderate steps and with spacing, so the engine can process the additional reach stably. In parallel, under Andromeda you scale not only via budget but above all via creative: more profitable reach comes more often from fresh, diverse creatives than from merely cranking the daily budget while the ad material stays the same. So the question is less “how much budget do I add” and more “with which new creative do I give the engine additional room.”

Advantage+ as the standard, not an experiment

Advantage+ is the automation layer Andromeda was built for — Meta describes Andromeda explicitly as an engine to strengthen Advantage+ automation. Advantage+ campaigns handle audience finding and placement in an automated way and thus fit the creative-first logic. In many accounts Advantage+ is now the default for sales, lead and app objectives. Your lever then no longer lies in audience tuning but in supplying the automation with excellent, diverse creative and giving it enough budget consistency for stable learning.

Discipline remains important here: automation does not replace your strategy. You still decide on the offer, the message, the creative direction, exclusion lists and clean conversion tracking. How we set up Advantage+ campaigns for lead objectives we go deeper on in our article on lead generation via Meta ads.

Patience with the learning phase

Consolidated, broadly delivering campaigns need a stable environment to learn. Constant intervention — daily budget swings, endless on/off switching, ongoing audience changes — keeps resetting the learning phase and prevents the engine from realising its potential. Under Andromeda, “intervene less, supply more creative” is often the more productive stance.

Clean data signals as the foundation

However capable the engine, it only ever optimises on the signals you feed it. An AI retrieval engine learning on patchy or distorted conversion data reliably optimises toward the wrong outcomes. Especially since the loss of many browser-side tracking signals, server-side data transfer via the Conversions API has become mandatory: it closes gaps the pixel-only approach leaves and gives the engine a more complete picture of which delivery actually led to value. Also watch for clean event definitions and consistent values — a double-counted or mis-valued event distorts optimisation across the entire account.

The link is direct: the better your data foundation, the more precisely Andromeda can identify valuable users and deliver your creative to them. Tracking under Andromeda is therefore not a technical afterthought but part of performance strategy. Skimp here and you give away exactly the signal quality the whole AI delivery lives on.

What to do NOW: the checklist

The order below is deliberately prioritised — from stocktaking to scaling.

• Creative audit: how many genuinely different creatives currently run per campaign? Do they cover different hooks, formats and funnel stages — or are they variants of the same motif?
• Build a creative pipeline: establish a fixed cadence for new creatives instead of sporadic one-off uploads. The goal is a continuous, plannable supply.
• Slim down audiences: dissolve unnecessary interest constraints and give the engine broad delivery. Keep sensible guardrails (country, language, age range, exclusions).
• Move to Advantage+: make Advantage+ the standard for your conversion and lead objectives, rather than only testing it.
• Consolidate campaigns: merge fragmented ad sets into few, broad structures with budget at the campaign level.
• Secure tracking: clean conversion tracking via pixel and Conversions API is the data basis the engine optimises on. Without reliable signals even the best engine runs empty.
• Respect the learning phase: reduce hectic intervention and give campaigns time to gather stable signals.
• Change how you measure: evaluate at the account and campaign level instead of micro-optimising individual ad sets.

Common mistakes under Andromeda

From practice, the patterns that most often cost performance:

• Still thinking audience-first: detail-obsessed interest stacks and narrow lookalikes brake the engine rather than help it. Over-steering the audience constrains the search space.
• Too few creatives: with two or three ads per campaign the engine has almost no material to match. Creative scarcity is the most common drag.
• Pseudo-diversity: ten variants of the same motif are not real diversity. You need different hooks, messages and funnel stages.
• Over-segmenting the structure: many small ad sets slice budget and learning signal. Consolidation is usually superior.
• Constant intervention: daily budget swings and endless on/off switching reset the learning phase and prevent stable optimisation.
• Neglecting tracking: patchy or double-counted conversion tracking feeds the engine wrong signals — it then optimises toward the wrong outcomes.
• Micro-evaluating success at ad-set level: under Andromeda the overall result at account and campaign level counts, not the isolated performance of individual ad sets.

Measurement and KPIs: what matters now

As the logic shifts, so does how you should measure success. Those who still look at individual ad sets like it is 2022 draw the wrong conclusions.

Account and campaign level instead of a micro view: evaluate performance holistically. The engine shifts delivery dynamically between creatives and segments; a single “weak” ad set can still contribute to the overall result by supplying signal. So judge the system, not the individual building block.

Take creative metrics seriously: since the creative is the central lever, hook rate (how many stay in the first seconds), hold rate, click-through rate and conversion quality per creative type belong in standard reporting. This shows which creative directions the engine rewards — so you can deliberately produce more of them.

Steer toward the business result: the most robust metric remains return on ad spend, or cost per acquisition against your target margin — not surface metrics like CPM. More expensive delivery that finds more profitable buyers beats cheaper delivery.

Keep incrementality in view: where possible, use lift tests to check what real additional revenue the campaigns generate, rather than relying on platform-native attribution alone. With broad delivery in particular this helps separate apparent wins from genuine uplift.

What Andromeda means for different advertisers

The effects are not the same everywhere. A differentiated view helps you set the right priorities.

E-commerce and DTC

Here Andromeda plays to its strength especially clearly: many products, clear conversion goals, large creative latitude. The lever lies in industrialised creative production and consistent Advantage+ usage. Those who regularly add fresh, diverse creatives give the engine what it needs.

Lead generation and service providers

The creative-first logic applies to lead objectives too — with the addition that lead quality depends strongly on the clarity of the message. Vague creatives attract vague enquiries. Those who sharpen offer and qualification logic in the creative itself get better leads. More on this in our article on lead generation via Meta ads.

B2B and advisory-intensive services

In B2B the audience is often narrow, which makes some hesitate to deliver broadly. The better route is usually to do the narrowing through the creative: building message, imagery and hook so they speak to exactly the right role. How to set up social campaigns in B2B we cover in the article on social media marketing in B2B.

Local advertisers and smaller budgets

With small budgets, consolidation is especially important, because fragmented structures dilute the already scarce learning signal. Few, broad campaigns with a manageable but genuinely diverse creative library are usually the most sensible path here.

Andromeda in the wider AI advertising context

Andromeda does not stand alone. It is part of a broader movement in which Meta bases the entire ad delivery more heavily on AI and automation — from audience finding through delivery to generative creative aids. For advertisers the tendency is: the platform takes over ever more of the operational fine-tuning, while the human contribution shifts toward strategy, the creative idea and a clean data and offer foundation. Those who accept this division of labour work with the platform rather than against it. How this shift fits into the whole of marketing work we frame in our article on AI in marketing. The strategic frame for all of this is provided by our work as a marketing agency in Hamburg.

Frequently asked questions

Is Meta Andromeda the same as the “Google Andromeda update”?

No. Andromeda is a genuinely named system at Meta — the ads retrieval engine behind Facebook, Instagram and Messenger ads, announced officially on 2 December 2024. “Google Andromeda,” by contrast, is not an official Google product but an umbrella term coined by the SEO industry for the AI rebuild of Google Search. The two have nothing to do with each other.

What exactly does the Andromeda engine do?

Andromeda is responsible for the retrieval stage — the step before ranking and auction take place. It filters, according to Meta, tens of millions of ad candidates down to a few thousand relevant ones that then proceed to ranking. In doing so it weights the creative heavily, rather than relying primarily on manually set audiences.

Do I have to delete my audiences now?

Not delete across the board, but slim down. Narrow interest stacks and granular lookalikes often brake the engine. Keep sensible guardrails (country, language, age range, exclusions) and give the engine broad delivery beyond that. Under Andromeda your audience knowledge flows above all into the creative.

How many creatives do I need per campaign?

There is no fixed mandatory number, but the logic is clear: the more genuinely different creatives the engine has to choose from, the better it can match. A broad, diverse creative library with different hooks, formats and funnel stages is the central lever — not the one supposedly perfect motif.

What does Andromeda mean for Advantage+?

Meta describes Andromeda explicitly as an engine to strengthen Advantage+ automation. The two belong together: Advantage+ handles audience finding and placement in an automated way, while Andromeda provides the more capable candidate selection beneath it. Your lever thus shifts from audience tuning to the quality and diversity of your creative.

Are the performance figures Meta states reliable?

The figures — such as +6% recall, +8% ad quality or the 10,000x model capacity — come from Meta’s own communication and measurement environment. We report them as “according to Meta.” Treat them as a directional and order-of-magnitude statement, not a guaranteed result for your specific account.

How do I measure success under Andromeda correctly?

Evaluate at account and campaign level instead of micro-optimising individual ad sets, watch creative metrics like hook and hold rate per creative type, and steer toward the business result (ROAS, or CPA against your target margin). Where possible, add lift tests to separate genuine additional revenue from platform-side attribution.

Does this also apply to small budgets and local advertisers?

Yes. With small budgets in particular, consolidation matters, because fragmented structures dilute the scarce learning signal. Few, broadly delivering campaigns with a manageable but genuinely diverse creative library are usually the most sensible path here.

Conclusion

Meta Andromeda is not a cosmetic update but a structural change in the logic of Facebook and Instagram advertising: the AI retrieval engine decides before the auction which ads even get a chance — and it reads the creative itself heavily to do so. This shifts the most important lever in the account from audience selection to creative diversity, from fragmented structures to consolidated, broadly delivering campaigns with Advantage+. Those who accept this and industrialise their creative production give the engine better signals and get more from the same budget. The uncomfortable truth: weak creative can no longer be disguised by clever targeting. The good news: those who master message, imagery and offer win under Andromeda more predictably than ever.