Meta Andromeda Explained: How to Build Ads for Meta's AI (2026)

Meta Andromeda decides which of your ads Meta even considers. Learn how it works, what Andromeda 2.0 changed, and how to build static ads its AI rewards.

Updated 17 min readBy the Static Ad Templates team

Key takeaways

  • Meta Andromeda is the retrieval stage of Meta's ad system: it narrows tens of millions of eligible ads to a few thousand candidates before ranking and the auction.
  • "Andromeda 2.0" is an industry nickname for Meta's October 28, 2026 next-generation retrieval post, not an official product, and it added no new settings or creative requirements.
  • Meta says creative diversification is the best lever to find relevant audiences, so each new ad should change the persona, angle or format, not just the color or font.
  • Meta's only published creative count is at least 20 diversified ads per Advantage+ shopping campaign, plus 20 to 30 percent of budget for testing new ads.
  • Static and text-led ads are cheap to produce in variety, and Motion's 2026 benchmarks found text-only and product-with-text images had the highest hit rates.

Meta Andromeda is the AI retrieval engine that picks which ads compete for each person on Facebook and Instagram. It narrows tens of millions of eligible ads to a few thousand candidates before ranking and the auction run. For advertisers, the practical effect is that varied creative now matters more than narrow targeting.

This guide separates what Meta has actually published from what agencies infer, then turns it into a playbook you can run with static ads.

What is Meta Andromeda?

Andromeda is the first filter in Meta's ad delivery system. Meta announced it on December 2, 2024, on its Engineering at Meta blog, describing it as a "next-gen personalized ads retrieval engine" built to support Advantage+ automation.

On Meta's Q2 2025 earnings call, CFO Susan Li described its job plainly: Andromeda "powers the ads retrieval stage of our ads system, where we select the few thousand most relevant ads from tens of millions of potential candidates."

That is the whole job. Andromeda does not set your bid, decide your audience, or pick the final ad. It builds the shortlist.

Meta's stated reason for building it matters: the 2024 post says the number of ad creatives "is expected to grow significantly" because of generative AI tools, so retrieval had to handle far more candidates, more precisely.

How Andromeda fits into Meta's ad pipeline

Each time someone opens Facebook or Instagram, Meta runs a multi-stage process in milliseconds. Here is the simplified version.

Stage What happens Systems Meta has named Scale
1. Eligibility Your targeting, budget, schedule and policy status decide which ads could be shown to this person at all Targeting checks (co-located with retrieval in the 2026 system) Tens of millions of ads
2. Retrieval The system quickly scores eligible ads and keeps the most relevant Andromeda, plus generative retrieval (Meta Generative Recommender) since 2026 Down to a few thousand
3. Ranking Heavier models predict how likely this person is to click, convert or engage with each candidate GEM (Generative Ads Recommendation Model), Lattice Down to a handful
4. Auction Your bid is weighed against predicted results and ad quality to pick the winner Meta's ad auction One ad per slot

The key point for advertisers: if your ad doesn't survive retrieval, ranking never sees it. Andromeda's job is deciding which of your ads are even worth considering for a specific person, and it makes that call partly from what the ad is.

How Andromeda differs from GEM and Lattice

Andromeda is retrieval: wide and fast. GEM is a ranking-stage foundation model Meta described in November 2025, whose inputs include "ad format, and creative representation." Lattice is a unified ranking architecture that Meta for Business said in March 2025 "increased ad quality by almost 12%."

The lines have blurred a little. On the Q3 2025 call, Meta said it had combined "models across retrieval and early stage ranking into a single model," which it credited with a 14% increase in ads quality on Facebook surfaces.

Meta Andromeda timeline: 2024 to the October 2026 update

Here is what happened, with each claim tagged by who made it. All performance figures are Meta's own reported numbers, often measured "on selected segments," and none are independently audited.

Date Event Source
Nov 19, 2024 Meta publishes creative diversification guidance: at least 20 diversified ads in Advantage+ shopping campaigns, 20 to 30% of budget for testing Meta for Business
Dec 2, 2024 Andromeda announced. Meta reports +6% retrieval recall and +8% ads quality on selected segments Engineering at Meta
Apr 22, 2025 Meta's "Creative Advantage" post names creative diversification "the best lever to find the most relevant audiences" Meta for Business
Jul 2025 Andromeda expands to Facebook Reels; Meta reports "nearly 4% higher conversions on Facebook mobile Feed and Reels" Q2 2025 earnings call
Oct 2025 Retrieval and early-stage ranking merged; 14% ads quality increase on Facebook surfaces Q3 2025 earnings call
~Oct 2025 Global rollout "completed," according to agencies and vendors such as Confect Industry-reported, not confirmed by Meta
Jan 2026 Andromeda now runs on Nvidia, AMD and Meta's own MTIA chips, with "nearly triple" compute efficiency Q4 2025 earnings call
Jul 2026 Meta deploys its first generative model into ads retrieval (Meta Generative Recommender) Q2 2026 earnings call
Oct 21, 2026 Andromeda improvements "increased conversions by 1.4% on Facebook" Q3 2026 earnings call
Oct 28, 2026 Meta publishes the next generation of Andromeda, which the industry calls "Andromeda 2.0" Engineering at Meta

Meta itself only says Andromeda began rolling out "in the second half of 2024." You will see "the rollout finished in October 2025" stated as fact in many articles. It comes from practitioner and vendor reports, so treat it as approximate.

What "Andromeda 2.0" actually is

Meta has never used the name "Andromeda 2.0." It is shorthand for an October 28, 2026 Engineering at Meta post titled "Meta Andromeda: Powering the Next Generation of Ads Retrieval Through Model-System Co-Design."

The post's goal is to score every eligible ad with close to ranking-level model quality, but at retrieval scale. Meta's headline claims:

  • Production retrieval "surfaces the top candidates at over one billion ads per second," described as "a 100x lift compared to the 2024 Meta Andromeda system."
  • About "300x fewer inference-time FLOPs than the 2024 Meta Andromeda retrieval model," using a mixture-of-experts design.
  • Andromeda has "delivered a 2.4% increase in conversions on Facebook" since 2025.
  • User-side inputs now include "LLM-derived semantic features" alongside engagement history.

One caveat on the 100x figure. Coverage of the post points out that the 2024 baseline appears to be quoted per GPU card while some 2026 comparisons are framed per server, so the comparison may not be like-for-like. The direction is clear (much faster, much cheaper to run), but don't repeat "100x" as a precise number.

What the post does not include is just as important. It adds no new setting, no required account structure and no creative count. Meta presents the benefit as "better customer matches and more conversions," delivered automatically. Advice you see attached to "Andromeda 2.0," such as "fewer near-duplicates, more distinct concepts," is agency interpretation. It's sensible interpretation, but it isn't in Meta's post.

What changed for advertisers: confirmed, inferred, or myth

Most Andromeda advice mixes things Meta has published, reasonable inferences, and claims nobody has evidence for. Here is how the main ones sort out.

Claim Status Basis
Creative diversification is the main lever for reaching relevant audiences Confirmed Meta, Apr 22, 2025: "the best lever to find the most relevant audiences"
Creatives that are too similar may be treated as the same ad and limit reach Confirmed Meta's creative differentiation guide
Run at least 20 diversified ads in Advantage+ shopping campaigns Confirmed (for that campaign type only) Meta, Nov 19, 2024
Put 20 to 30% of budget toward testing new ads Confirmed (recommendation) Meta, Nov 19, 2024
Broad or Advantage+ targeting works better than narrow interest stacks Inferred Fits Meta's direction; evidence is mostly case studies
Consolidate into fewer ad sets with more budget each Inferred Fits Meta's direction; gains are mostly case studies
"Creative is the new targeting" Confirmed in substance Meta's wording is about creative as a lever; the slogan is practitioner shorthand
Every ad set now needs 20 or more ads Myth Meta's 20 figure applies to Advantage+ shopping campaigns, not every ad set
Andromeda replaced targeting or ranking Myth Andromeda is retrieval only

Creative is now the main lever

Meta's April 2025 post defines creative diversification as "creating a wide range of ad creatives with different themes, messages, and visuals to cater to diverse audience segments." It describes a shift from niche targeting toward that diversification.

If retrieval matches ads to people partly by what the ads contain, ads aimed at different needs can reach different people. Ten ads with the same message give the system one thing to match.

Similar ads compete for the same people

This is the most useful line Meta has published for creative teams. Its creative differentiation guide says: "When creative is too similar, our system may not recognize it as different and could deliver it to the same sets of audiences," and that this overlap "limits your reach."

Jon Loomer, who has tested Andromeda-era structures extensively, put it this way in October 2025: "Unless they're distinctly different enough, any variation is unlikely to make a meaningful impact on performance." Text changes, he added, should be "significant, not subtle."

Broad targeting and consolidation

Meta's direction since Advantage+ is fewer constraints and fewer, larger ad sets, so Andromeda can consider more people and ads together. Most consolidation wins you read about are single-account case studies, though. Broad targeting is a sensible default for most DTC accounts, but keep real exclusions (existing customers, regions you don't ship to).

How Meta's AI understands your creative

Meta has not said Andromeda runs OCR on ad images. Its two Andromeda engineering posts don't describe analyzing the visual or text content of creatives in detail.

Meta has confirmed that its wider ads system represents ads by their content, though:

  • GenRecs (October 19, 2026): "Pre-trained multimodal models read each ad, product, and page, including its text and images, and encode the items into an embedding." Meta says these are used for "both retrieval and ranking," and "similar items share tokens."
  • Meta Generative Recommender (Q2 2026 call): Meta is "using LLMs to reason about ad content and user preferences together, and predict the best ad for each person," and has deployed this into retrieval.
  • GEM (November 2025): uses "creative representation" as a ranking input.

So the safe statement is this: Meta's AI analyzes your ad's content, very likely including the words on your image, to decide who it's relevant for. That is a strong inference from Meta's statements, not a Meta quote about headline text specifically.

Two practical consequences follow. On-image copy written for one specific person is a relevance signal. And because "similar items share tokens," near-identical images with small text tweaks probably look the same to the system.

For the full evidence, including what Meta's old 20% text rule and its Rosetta text-recognition system do and don't tell us, read our deep dive: does Meta read the text in your ads?

The Andromeda playbook for static ads

The working rule: give Meta's AI genuinely different ads to match to different people. Static ads are the cheapest way to do that at volume.

Why static ads suit Andromeda

A video concept can take days or weeks to make. A static concept can take an afternoon, so you get more distinct concepts per dollar.

The data supports keeping statics in the mix. Motion's 2026 Creative Benchmarks analyzed 578,750 creatives across 6,015 Meta ad accounts and $1.29 billion in spend (September 2025 to January 2026). It ranked "Text only" first and "Product image with text" second of 15 asset types on hit rate. Plain product images with no text ranked 13th.

In a March 2026 talk on the same data, Motion gave hit rates of 11.60% for text-only ads and 8.75% for product images with text, against 6.87% for high-production creative. Motion defines a "winner" as an ad spending at least 10x its account's median creative spend (and at least $500), so this measures where Meta put budget, not ROAS. It is also vendor data from Motion's own customers.

Read it as evidence that text-led statics hold their own after Andromeda, not as proof they beat video. For the full format breakdown, see our static ads guide, and for the head-to-head see static vs video ads.

Step 1: Build a persona × angle × format matrix

Start from the people who buy, not from the design. List three or four buyer personas, two or three angles each, and pick a format that suits each angle. Every cell is one concept.

Here is an example for a hypothetical no-pull dog walking vest:

Concept Persona Angle Format On-image headline
1 Owner of a dog that pulls Problem + fix Problem + Solution "Walks shouldn't feel like a tug of war"
2 Owner of a dog that pulls Proof Testimonial Real customer quote about walks with a puller
3 First-time puppy owner Comparison Us vs Them Vest vs collar: pressure on the neck
4 Weekend hiker Features Feature/Benefit Callouts: padded chest, handle, reflective trim
5 City walker, small dog Candid voice Apple Notes "Things I wish I knew before my first walking vest"
6 Any (warm audience) Offer Offer/Sale "20% off this weekend + free returns"

Six concepts, six different messages for different people, with six different visual structures. That gives the system something real to work with.

If you're not a designer, you don't need to invent layouts. In our static ad template library, you can filter by format (Testimonial, Us vs Them, Apple Notes, Problem + Solution and 13 others) and by niche, then edit the template in Canva.

Step 2: Know what counts as "different"

Meta's own differentiation guide suggests angles such as testing urgency and exclusivity, promoting delivery and returns, giving a relatable perspective, using text overlays, dramatizing the benefit, and reframing the product as a conversation.

Likely reads as a new concept Likely reads as the same ad
New persona with a matching headline and image Same image in a new background color
New angle (proof, comparison, offer, problem) Same headline in a different font
New format or layout (testimonial vs us-vs-them) Swapping one word in the headline
New product context (on a hike vs on a city street) Same product shot cropped differently
New offer or reason to buy now Same ad with a new button color

Change the message and the visual together. Text alone may not be enough, and a new image with the same message may not be either. The small variations on the right are still useful for iterating on a proven winner. They just don't count as diversification.

Step 3: Decide how many ads to run

There is no Andromeda ad count. Meta's only published number is "at least 20 diversified ads in Advantage+ shopping campaigns," from November 2024, and it predates Andromeda's announcement.

Practitioners who have tested this are skeptical of big counts. Jon Loomer reported in November 2025 that he tested 25 and 50 ads per ad set without a clear advantage, and concluded that "Quality still beats quantity." A January 2026 AdExchanger column by Monica Shukla of Mile Marker called the idea that "20 or more creatives per ad set is universally beneficial" a misconception, arguing volume should match budget and funnel stage.

A reasonable starting point for small budgets is three to six distinct concepts per ad set, scaling up as spend grows. Flexible Format lets you put several assets in one ad if you want variety without fragmenting delivery. We cover the numbers by budget level in how many ads per ad set.

Step 4: Refresh on a weekly rhythm

Meta's November 2024 guidance says top advertisers "routinely launch new ads on a weekly basis." Motion's 2026 benchmarks put the average micro account (under $10K a month) at 2.8 new creatives a week and the top quarter of those accounts at 4.83.

Expect most ads to fail. In Motion's 2026 data, "Roughly five percent of creatives on Meta spend at least 10× their account median," and about half of creatives are switched off within 28 days. If you launch 20 concepts and one or two become winners, that's normal.

Watch Ads Manager's Delivery column for "Creative fatigue" or "Creative limited" labels, alongside falling CTR and rising frequency. Meta publishes no fixed frequency threshold, so judge fatigue against each ad's own history.

Step 5: Ring-fence a testing budget

Meta recommends "dedicating 20-30% of your overall ad budget towards testing new ads." On a $5,000 monthly budget, that's $1,000 to $1,500 for new concepts, with the rest behind proven winners.

Without a protected budget, spend tends to pile onto existing winners and new concepts never get enough delivery to prove themselves. Some advertisers run a separate testing campaign and graduate winners into Advantage+; others test inside the main campaign. Either works if you stay consistent.

Step 6: Write each headline for one person

Because Meta's models read ad content, a vague headline like "The best walking vest for your dog" matches everyone a little and no one strongly. "Walks shouldn't feel like a tug of war" speaks to one owner's daily frustration.

Specific copy gives the system a clearer signal about who the ad is for. Be careful with how you get specific, though. These choices carry real rejection or legal risk:

  • Implying personal attributes. Headlines like "Are you depressed?" or "Struggling with your weight?" can violate Meta's personal attributes policy. Describe the situation, not the viewer.
  • Notification, AirDrop or fake-button mockups. Meta restricts imagery that copies non-functional buttons or notifications. Design these formats so nothing looks clickable, and expect some rejections.
  • Weight-loss before and after images. Meta restricts these. Use other proof formats in that category.
  • Fake or AI-generated reviews. The FTC's rule banning fake reviews and testimonials, including AI-generated ones, has been in force since October 2024. Only use real customer quotes in testimonial ads.

Step 7: Find angles from ads that already run

Ads that competitors have kept running for months are usually still paying their way. Our Facebook Ads Library guide shows how to find them and pull out the angle, persona and format behind each.

Common Meta Andromeda myths

"Andromeda 2.0 is a new product with new rules"

It's an industry label for Meta's October 2026 engineering update, which made retrieval faster and more accurate without adding settings, structures or creative quotas.

"Meta assigns each ad an Entity ID, and ads over 60% similar get grouped"

Neither "Entity ID" nor a "60% similarity threshold" appears in Meta's published material. They come from vendor and agency blogs. The closest Meta statements are that too-similar creative "may not recognize it as different" and that in GenRecs "similar items share tokens." The underlying idea (near-duplicates compete with each other) is supported. The specific term and number are not.

"Andromeda reads the text on your image with OCR"

Meta has never said Andromeda performs OCR on ads. It has said its multimodal models "read each ad, product, and page, including its text and images." The practical advice is the same either way: write on-image copy that's clear, specific and legible.

"You need exactly 8, 15, 20 or 50 ads per ad set"

No Meta source gives an Andromeda ad count. The 20 figure applies to Advantage+ shopping campaigns and predates Andromeda's announcement.

"Andromeda increased ROAS by 22%"

That 22% figure appears in the same December 2024 Meta post, but it describes Advantage+ features, not Andromeda. Andromeda's reported launch numbers were +6% recall and +8% ads quality on selected segments.

"Andromeda does the targeting now"

Andromeda builds a shortlist. Targeting still defines eligibility, ranking still predicts outcomes, and the auction still picks the winner.

"Andromeda tanked everyone's ROAS"

Confect's March 2026 study of 3,014 ecommerce advertisers across 73 countries found average ROAS fell about 7% during the rollout period. It's a vendor dataset drawn from Confect's own users, and it's correlational, so it can't separate Andromeda from seasonality, auction prices or other changes. Meta, meanwhile, reports conversion gains with each update.

What to do next

Andromeda doesn't need a new setup. It rewards the same habits Meta has recommended since late 2024, applied with more discipline. This week:

  1. Audit your live ads. Group any that share the same message and visual structure, and count how many genuinely distinct concepts you actually have.
  2. Build a persona × angle × format matrix with at least six cells, and write a specific on-image headline for each.
  3. Produce the gaps as statics first, since they're the fastest to make. Static ad templates by industry is a quick way to see which formats suit your niche.
  4. Set aside 20 to 30% of budget for testing, and launch new concepts weekly.
  5. Judge results over weeks, not days, and expect only a few winners per batch.

If your account has one message in ten colors, fixing that will likely do more than any setting change.

Sources

  1. Engineering at Meta, "Meta Andromeda: Supercharging Advantage+ automation with the next-gen personalized ads retrieval engine" (Dec 2, 2024), https://engineering.fb.com/2024/12/02/production-engineering/meta-andromeda-advantage-automation-next-gen-personalized-ads-retrieval-engine/ (accessed Nov 2026)
  2. Engineering at Meta, "Meta Andromeda: Powering the Next Generation of Ads Retrieval Through Model-System Co-Design" (Oct 28, 2026), https://engineering.fb.com/2026/10/28/ml-applications/meta-andromeda-next-gen-ads-retrieval-gpu-co-design/ (accessed Nov 2026)
  3. Meta, "Q2 2025 Earnings Call Transcript", https://s21.q4cdn.com/399680738/files/doc_financials/2025/q2/META-Q2-2025-Earnings-Call-Transcript.pdf (accessed Nov 2026)
  4. Meta, "Q3 2025 Earnings Call Transcript", https://s21.q4cdn.com/399680738/files/doc_financials/2025/q3/META-Q3-2025-Earnings-Call-Transcript.pdf (accessed Nov 2026)
  5. Meta, "Q4 2025 Earnings Call Transcript", https://s21.q4cdn.com/399680738/files/doc_financials/2025/q4/META-Q4-2025-Earnings-Call-Transcript.pdf (accessed Nov 2026)
  6. Meta, "Q2 2026 Earnings Call Transcript", https://s21.q4cdn.com/399680738/files/doc_financials/2026/q2/META-Q2-2026-Earnings-Call-Transcript.pdf (accessed Nov 2026)
  7. Meta, "Q3 2026 Earnings Call Transcript" (Oct 21, 2026), https://s21.q4cdn.com/399680738/files/content_files/META-Q3-2026-Earnings-Call-Transcript.pdf (accessed Nov 2026)
  8. Meta for Business, "AI innovation in Meta's ads ranking driving advertiser performance" (Mar 27, 2025), https://www.facebook.com/business/news/ai-innovation-in-metas-ads-ranking-driving-advertiser-performance (accessed Nov 2026)
  9. Meta for Business, "The Creative Advantage: unlocking the power of diversification with Meta Andromeda" (Apr 22, 2025), https://www.facebook.com/business/news/the-creative-advantage-unlocking-the-power-of-diversification-with-meta-andromeda (accessed Nov 2026)
  10. Meta for Business, "Three steps to optimize your performance with creative diversification" (Nov 19, 2024), https://www.facebook.com/business/news/three-steps-to-optimize-your-performance-with-creative-diversification (accessed Nov 2026)
  11. Meta for Business, "Creative differentiation", https://www.facebook.com/business/m/small-business/creative-differentiation (accessed Nov 2026)
  12. Engineering at Meta, "Meta's Generative Ads Model (GEM): the central brain accelerating ads recommendation AI innovation" (Nov 10, 2025), https://engineering.fb.com/2025/11/10/ml-applications/metas-generative-ads-model-gem-the-central-brain-accelerating-ads-recommendation-ai-innovation/ (accessed Nov 2026)
  13. Engineering at Meta, "GenRecs Tokenization: Transforming Ads Recommendation at Meta with Stable Model Vocabulary" (Oct 19, 2026), https://engineering.fb.com/2026/10/19/ml-applications/genrecs-tokenization-transforming-ads-recommendation-at-meta-with-stable-model-vocabulary/ (accessed Nov 2026)
  14. Jon Loomer, article on Meta Andromeda and ad retrieval (Nov 10, 2025), https://www.jonloomer.com/meta-andromeda-ad-retrieval/ (accessed Nov 2026)
  15. Jon Loomer, article on Meta Andromeda and creative diversification (Oct 20, 2025), https://www.jonloomer.com/meta-andromeda-creative-diversification/ (accessed Nov 2026)
  16. AdExchanger, Monica Shukla, "What Meta's Andromeda Update Actually Changes, And What It Doesn't" (Jan 28, 2026), https://www.adexchanger.com/data-driven-thinking/what-metas-andromeda-update-actually-changes-and-what-it-doesnt/ (accessed Nov 2026)
  17. Motion (Runneth), "Creative Benchmarks 2026: Methodology", https://runneth.com/library/research/creative-benchmarks-2026/methodology (accessed Nov 2026)
  18. Motion (Runneth), "Creative Benchmarks 2026: Top asset types", https://runneth.com/library/research/creative-benchmarks-2026/top-asset-types (accessed Nov 2026)
  19. Motion (Runneth), "Meta ads in 2026: how many creatives do you actually need to launch?" (talk, Mar 2026), https://runneth.com/library/talk/meta-ads-in-2026-how-many-creatives-do-you-actually-need-to-launch/ (accessed Nov 2026)
  20. Motion (Runneth), "Creative Benchmarks 2026", https://motionapp.com/library/research/creative-benchmarks-2026/ (accessed Nov 2026)
  21. Motion (Runneth), "CB2026 key benchmarks and insights", https://runneth.com/thumbstop-pulse/cb2026-key-benchmarks-and-insights (accessed Nov 2026)
  22. Confect, Meta Andromeda 2026 study (Mar 10, 2026), https://confect.io/tactics/meta-andromeda-2026 (accessed Nov 2026)
  23. Storyboard18, "Meta's Andromeda rollout disrupts ad ecosystem" (Oct 27, 2025), https://www.storyboard18.com/advertising/metas-andromeda-rollout-disrupts-ad-ecosystem-advertisers-agencies-scramble-to-adapt-83155.htm (accessed Nov 2026)
  24. Federal Trade Commission, "Federal Trade Commission Announces Final Rule Banning Fake Reviews and Testimonials" (Aug 14, 2024), https://www.ftc.gov/news-events/news/press-releases/2024/08/federal-trade-commission-announces-final-rule-banning-fake-reviews-testimonials (accessed Nov 2026)

Frequently asked questions

What is Meta Andromeda in simple terms?

+

Meta Andromeda is the AI system that makes Meta's first cut of ads for each person. Out of tens of millions of eligible ads, it picks a few thousand likely to be relevant. Ranking models then score that shortlist and the auction chooses the winner. Meta announced Andromeda in December 2024 and released a next-generation version in October 2026.

Is Meta Andromeda 2.0 real?

+

Meta never used the name "Andromeda 2.0." Marketers use the label for Meta's October 28, 2026 engineering post about the next generation of Andromeda. That post describes faster, more accurate retrieval and a reported 2.4% conversion lift on Facebook since 2025. It does not add any new setting, campaign structure or creative requirement for advertisers.

Do I need to change my Ads Manager settings for Andromeda?

+

No. Andromeda runs automatically for every advertiser and has no toggle. What changes is what helps it: broader targeting, fewer ad sets with more budget each, and a wider range of distinct creative concepts. Meta's guidance focuses on creative diversification and testing budget, not on any Andromeda-specific setting.

Does Andromeda favor video ads over static image ads?

+

Meta has not said Andromeda prefers any format. It recommends a mix of formats as part of creative diversification. Static images remain viable, and in Motion's 2026 benchmarks text-only and product-with-text images had the highest hit rates of the asset types measured. The risk with statics is running near-identical product shots, which the system may treat as one ad.

Did the Andromeda update hurt ROAS?

+

The evidence is mixed. A March 2026 study by Confect of 3,014 ecommerce advertisers found average ROAS fell about 7% during the rollout, but it covers Confect's own customer base and cannot separate Andromeda from seasonality or auction costs. Meta reports conversion gains from each Andromeda update. Results depend heavily on account structure and creative variety.

Keep reading

Or browse all articles.

Put this into practice today

4,000+ static ad templates modeled on proven winners, organized by niche and format. Open any of them in Canva and launch in minutes.