This post quotes Meta sources dated 16 December 2025 and 29 July 2026, and is re-checked against them quarterly.
Search "Meta Andromeda" and the top pages disagree on when it launched, what it does, and which numbers to trust.
This post sorts those claims against Meta's own words, then gives you a count you can do by hand in about twenty minutes: how many distinct ideas sit behind the 40 ads you got live. Creative diversification, as Meta defines it, is about distinct ideas, and most accounts have never counted theirs.
What is Meta Andromeda?
Meta Andromeda is Meta's ads retrieval system. Meta says it was introduced in the second half of 2024 and narrows a pool of tens of millions of ads to the few thousand it considers showing one person (Meta Q4 2024 earnings call, 29 January 2025).
Retrieval is the first cut. Meta CFO Susan Li described it on that call as "the part of the ranking process where we narrow down a pool of tens of millions of ads to the few thousand we consider showing someone." An ad that doesn't make that cut isn't considered for that person at all.
The launch date comes from the same call: "In the second half of 2024, we introduced an innovative new machine learning system in partnership with Nvidia, called Andromeda." Meta hasn't published a month, or a date when the rollout finished. A page that gives you one is going past Meta's sources.
Meta later said it had "extended our Andromeda ads retrieval engine so it can now run on Nvidia, AMD, and MTIA" (Q4 2025 earnings call, 28 January 2026), and on 29 July 2026 that it had "deployed the first generative model into our ads retrieval system", a passage that doesn't name Andromeda.
For your creative, the sentence that matters came in December 2025. Meta wrote that Andromeda "quickly scans tens of millions of ads and selects a few thousand strong candidates so only the most relevant and distinct creatives move forward" (Meta for Business, "Demystifying Creative Diversification", December 16, 2025).
Distinct is Meta's word. The rest of this post is about checking whether your ads earn it.
Which Andromeda claims are sourced?
Most of what you'll read about Andromeda comes from vendor and agency blogs. Here are ten widely repeated claims, set against what Meta has published.
| Claim you will read | What Meta has published | Verdict |
|---|---|---|
| Andromeda's rollout "finished in October 2025", or in "early 2025" or "mid-2025". | "In the second half of 2024, we introduced..." (Q4 2024 earnings call, 29 January 2025); "introduced in 2024" (Meta for Business, "AI Innovations in Meta's Ads Ranking Driving Advertiser Performance", March 27, 2025). | Untraceable. No source found for a completion date (checked October 2026). Use Meta's date. |
| A creative similarity score above 60% suppresses delivery, so keep ads under 40% similar. | Describes grouping ads that "look or feel alike", with no number (Meta for Business, "Demystifying Creative Diversification", December 16, 2025). | No source found (checked October 2026). Meta has published no threshold. |
| Similar ads collapse into a single "Entity ID". | Ads that look or feel alike "are seen as variations of the same creative", with learnings shared at the creative level (Meta for Business, "Demystifying Creative Diversification", December 16, 2025). | Partly sourced. The grouping has a Meta source; the "Entity ID" mechanism has none (checked October 2026). |
| Andromeda brought a 10,000x increase in ad volume. | "a 10,000x increase in the complexity of models we use for ads retrieval" (Q4 2024 earnings call, 29 January 2025). | Contradicted. Meta's source says model complexity, not the number of ads. |
| Andromeda lifted ROAS 300% while CPMs fell 73%. | Nothing. | No source found for either number (checked October 2026). |
| Meta made creative diversification mandatory in October 2024. | "we recommend maintaining at least 20 diversified ads in Advantage+ shopping campaigns" (Meta for Business, "Three steps to optimize your performance with creative diversification", November 19, 2024). | Contradicted. Meta's source is a recommendation for one campaign type, in its November 2024 wording. |
| "Creative is the new targeting": Andromeda replaced audience targeting. | Andromeda passes forward "only the most relevant and distinct creatives" (Meta for Business, "Demystifying Creative Diversification", December 16, 2025); the GEM write-up lists "creative representation" among ad attributes (Engineering at Meta, "Meta's Generative Ads Model (GEM): The Central Brain Accelerating Ads Recommendation AI Innovation", November 10, 2025). | The direction has a Meta source. "Replaced audience targeting" has none (checked October 2026). |
| Andromeda improved ad quality 8%. | "+8% ads quality improvement on selected segments" (Engineering at Meta, "Meta Andromeda: Supercharging Advantage+ automation with the next-gen personalized ads retrieval engine", December 2, 2024). | Sourced, for selected segments only. Quote Meta's source with its qualifier. |
| "Andromeda 2.0" launched in 2026. | "We deployed the first generative model into our ads retrieval system" (Q2 2026 earnings call, 29 July 2026). | Not a term in the Meta sources checked (October 2026); cite Meta's sentence instead. |
| Accounts adapted to Andromeda see 20-35% or 65% higher ROAS. | Nothing. | No source found. Vendor analyses, no original study located (checked October 2026). |
How these verdicts were reached: each claim was read against Meta-owned pages (Meta for Business News and Engineering at Meta) and Meta's earnings call transcripts in October 2026. "No source found" means no Meta source and no original study turned up.
Meta has not published a creative similarity threshold, so the circulating "60% similarity score" has no Meta source (checked October 2026).
What is creative diversification in Meta ads?
Meta's December 2025 article, "Demystifying Creative Diversification" (Meta for Business, December 16, 2025), opens by separating two things marketers treat as one: "Creative diversification and creative iteration are two separate concepts."
Creative diversification is the creation of distinct campaign assets tailored to different personas or use cases, while creative iteration is manual A/B testing to refine assets, formats, and messages (Meta for Business, "Demystifying Creative Diversification", December 16, 2025).
Meta gives the test in one sentence: "Creative iteration might produce two ads with identical visuals, but different text CTAs, while creative diversification would generate two distinctly different pieces of creative" (Meta for Business, "Demystifying Creative Diversification", December 16, 2025).
Read that with your own account in mind. Iteration changes the wrapper: same footage, new caption, new call to action. Diversification changes what's inside it: a different person, a different problem, a different reason to believe you.
Both are legitimate work, and they aren't interchangeable. If your account is short on ideas, iteration won't fix it, because iteration works on ideas you already have.
Does Meta actually group your similar ads?
Yes. The precise version is less dramatic than the one circulating, and more useful.
In the same December 2025 article, "Demystifying Creative Diversification" (Meta for Business, December 16, 2025), Meta writes: "Our ads system also analyzes creatives and groups together ads that share key visual and thematic attributes. When multiple ads look or feel alike, these are seen as variations of the same creative, meaning learnings and delivery optimizations are shared at the creative level."
Meta says its ads system groups ads that share key visual and thematic attributes, and shares learnings and delivery optimizations at the creative level, not only the individual ad level (Meta for Business, "Demystifying Creative Diversification", December 16, 2025).
The direction is clear enough to act on. The system reads your creative, decides which ads resemble each other, and treats that as information.
Creative similarity vs creative fatigue
Meta names two failure modes in the same article, and most teams only diagnose one of them.
Meta defines creative fatigue as overexposure to the same creative, and creative similarity as ad images that are visually too alike even if they are technically different assets; it adds that high similarity can also drive fatigue (Meta for Business, "Demystifying Creative Diversification", December 16, 2025).
That last part is easy to miss. In Meta's words: "High similarity can also drive fatigue, as users perceive ads as repetitive, reducing engagement" (Meta for Business, "Demystifying Creative Diversification", December 16, 2025). The two are separate problems, and one feeds the other.
You can have a fresh account and still have a similarity problem. Ten new files shipped this week, none of them tired, all of them variations on the same idea. Nothing is worn out, and nothing is new either.
There's a name for that pattern: creative fatigue is a supply problem, not a detection problem.
Why "just launch more" stopped working
The volume advice wasn't wrong when you first heard it. Making an ad used to be slow and expensive, so ad count was a decent proxy for effort, and effort a decent proxy for supply.
That proxy broke, and the platform broke it.
More than 4 million advertisers were using at least one of Meta's generative AI ad creative tools in January 2025, up from one million six months earlier, according to Meta CFO Susan Li on the Q4 2024 earnings call (29 January 2025).
On the Q1 2026 call (29 April 2026), Meta put the figure at more than 8 million advertisers. Both figures count advertisers, and setting them side by side is our pairing, not a growth rate Meta published. When millions of advertisers have the same tools, more files stop setting you apart, because every one of them enters retrieval, where Meta narrows tens of millions of ads to the few thousand it considers showing one person (Meta Q4 2024 earnings call, 29 January 2025; Meta for Business, "Demystifying Creative Diversification", December 16, 2025). The fortieth version of an idea already in that pool is a weak position. The only ad naming a problem nobody else names is a much better one.
The concept census: count what you actually have
This is the part you can do today, by hand. Open Ads Manager, pull up your live ads, and open a spreadsheet.
One row per live ad, five columns.
| Column | What to look for | Example (illustrative) |
|---|---|---|
| 1. Problem or desire | The single thing the ad leads with in its first three seconds or its headline | "My skin looks tired by 3pm" |
| 2. Buyer and awareness stage | Who it addresses, and how much they already know about the problem and your solution | First-time buyer, problem aware, solution unaware |
| 3. Proof type | Demo, testimonial, data, authority, before and after, or none | Founder authority plus a demo |
| 4. Format and structure | Talking head, UGC unboxing, static, carousel, editorial | UGC talking head, 22 seconds |
| 5. Offer or ask | What the ad asks for, and on what terms | Shop now, no discount |
The single thing the ad leads with in its first three seconds or its headline
Who it addresses, and how much they already know about the problem and your solution
Demo, testimonial, data, authority, before and after, or none
Talking head, UGC unboxing, static, carousel, editorial
What the ad asks for, and on what terms
Column definitions operationalise Meta’s stated distinction between creative iteration and creative diversification. The example column is illustrative, invented for this post. Not a real ad account and not customer data.
Then apply one rule.
Two ads are the same creative concept, not two different ones, if the problem they lead with, the buyer they address, and the proof type they use all match, regardless of thumbnail, caption, hook line, or edit length.
Columns 4 and 5 are still worth recording, because they tell you where you are over-invested in one format and where an offer is doing all the heavy lifting. They just do not create a new concept on their own.
Collapse the duplicates. Count the rows that are left. That number, not your ad count, is the honest measure of your creative supply.
One CreatStrat customer described the same line between real concepts and cosmetic ones:
We gave CreatStrat one product and within the first week had more genuinely different concepts to test than we'd produced internally in the previous month. The biggest difference wasn't just volume, the ads were based on different customer problems, objections and awareness levels rather than being cosmetic variations of the same concept.
Claire M. · CreatStrat customerHere is what a collapse looks like on four synthetic ads.
| Ad | Problem or desire | Buyer and stage | Proof type | Concept |
|---|---|---|---|---|
| Ad 1: UGC, 20s | Skin looks tired by 3pm | First-time buyer, problem aware | Before and after | A |
| Ad 2: same footage, new hook line | Skin looks tired by 3pm | First-time buyer, problem aware | Before and after | A |
| Ad 3: static carousel, new photography | Skin looks tired by 3pm | First-time buyer, problem aware | Before and after | A |
| Ad 4: founder to camera | Worried the formula is harsh | Repeat buyer, solution aware | Ingredient authority | B |
Four ads. Two concepts. Ad 3 feels like the biggest change of the set, because it required a new shoot, new photography, and a new format. It is still concept A. Ad 4 cost the least to make and is the only genuinely new idea in the group.
That is the uncomfortable arithmetic this exercise produces, and it is why the count is worth doing by hand rather than assuming. Run the census on your own account and see what you get. A lot of accounts land in the single digits.
Once you have your number, check it against concepts you haven't tried yet.
Free. 3 researched opportunities, 1 customer-avatar gap, and 1 example concept within 24 hours.
A note on where the five columns come from, because a framework you cannot trace is just an opinion with a table around it. Columns 1 through 3 operationalise Meta's own distinction between iteration and diversification (Meta for Business, "Demystifying Creative Diversification", December 16, 2025) into attributes you can read off an ad without a tool. The underlying idea is older than the platform: Schumann, Petty and Clemons (1990), writing in the Journal of Consumer Research, separated cosmetic variation, meaning changes to nonsubstantive features, from substantive variation, meaning changes to the actual product attributes and arguments, and found that the two do not behave the same way, with the difference depending on how motivated the audience is to process the message. You can read Meta's iteration-versus-diversification split as a parallel to that 35-year-old distinction, though neither source connects the two. The census just gives you a way to apply it on a Tuesday.
What makes two concepts actually different
A new hook line on the same footage is iteration, a better way to express an idea you already have, while a new buyer with a new proof type is a new concept. The pattern underneath: a new concept requires a change to what the ad argues, who it argues it to, or what makes the argument believable, and everything else is production.
What to do with your number
Say you counted, and the number is smaller than you expected. There are two honest ways to close that gap.
The first is to make more finished assets: more shoots, more creators, more edit capacity. For some teams that is the right answer.
The second is to brief better, so each asset you can already make carries a new concept instead of another pass at an old one. It's usually the cheaper half: the same shoot day yields three concepts instead of one variation set when the brief arrives with the buyer, the problem, and the proof type decided.
CreatStrat covers both halves. It's a software subscription that delivers finished ads every week, with your creative strategist built in. It researches your brand, customers, market, and connected ad performance, plans the tests, decides which buyer and problem each ad leads with, and writes the brief behind it. Then it produces the finished statics, carousels, mashups, AI video, and AI UGC in your weekly mix, each with its Meta copy. You launch them into your own campaigns. The Meta connection is read-only, and account learning carries into the next batch automatically.
Because every ad arrives with its brief, you can run the census above on each batch and see which concept every ad belongs to.
Two useful next reads: how many new ads per week at your spend level, and the weekly creative operating system that keeps the cadence honest. For how many tests it takes to find one winner, see why published hit rates disagree.
Frequently asked questions
What is Meta Andromeda? Meta Andromeda is Meta's ads retrieval system, introduced in the second half of 2024 (Meta Q4 2024 earnings call, 29 January 2025). It narrows a pool of tens of millions of ads to the few thousand Meta considers showing one person.
When did Meta Andromeda launch? Meta says it introduced Andromeda in the second half of 2024 (Q4 2024 earnings call, 29 January 2025). Meta has published no rollout-completion date, so dates like October 2025 have no Meta source (checked October 2026).
Is the 60% creative similarity threshold real? Meta has not published a similarity threshold. Its article "Demystifying Creative Diversification" (Meta for Business, December 16, 2025) says ads that look or feel alike are treated as variations of one creative, and gives no number. The 60% figure has no Meta source (checked October 2026).
Does Meta treat similar ads as the same ad? Meta says its system groups ads that share key visual and thematic attributes and shares learnings and delivery optimizations at the creative level (Meta for Business, "Demystifying Creative Diversification", December 16, 2025). That is a statement about learning and optimization, not about which ads are eligible to be shown.
What counts as a different creative concept? Two ads are different concepts if the problem or desire they lead with, the buyer they address, or the proof type they use differs. Format, thumbnail, edit length, and hook line alone do not create a new concept.
How many creative concepts should I run at once? Meta hasn't published a number (checked October 2026). Run a concept census on your own account instead: count distinct concepts rather than ad files, and use that count to judge whether your creative supply matches your spend.
Count first, then decide
Ad count was a proxy, and the platform made it cheap enough to stop meaning anything. Concept count is the number Meta's own vocabulary points at, from "distinct creatives" in retrieval to "variations of the same creative" in grouping (Meta for Business, "Demystifying Creative Diversification", December 16, 2025), so count it: twenty minutes, five columns, one rule, and a number that describes your creative supply, not your team's effort.
Want new concepts to count? Share a product page URL and email, and your Gap Scan arrives within 24 hours to check against your census.
Want to see the app first? Book a demo →
Sources
- Meta for Business. 2025. Demystifying Creative Diversification. Meta for Business News, 16 December 2025. https://www.facebook.com/business/news/demystifying-creative-diversification
- Meta for Business. 2024. Three steps to optimize your performance with creative diversification. Meta for Business News, 19 November 2024. https://www.facebook.com/business/news/three-steps-to-optimize-your-performance-with-creative-diversification
- Meta for Business. 2025. AI Innovations in Meta's Ads Ranking Driving Advertiser Performance. Meta for Business News, 27 March 2025. https://www.facebook.com/business/news/ai-innovation-in-metas-ads-ranking-driving-advertiser-performance
- Sun C, Yu N, Lu H, Wang L, Pu Y, Liu G, Bhatia N, Musumeci GP. 2024. Meta Andromeda: Supercharging Advantage+ automation with the next-gen personalized ads retrieval engine. Engineering at Meta, 2 December 2024. https://engineering.fb.com/2024/12/02/production-engineering/meta-andromeda-advantage-automation-next-gen-personalized-ads-retrieval-engine/
- Li H, Liu X, Nie J, Wen E, Yang C, Yang J, Yu N, Beg H, Arditi G, Bhatia N. 2025. Meta's Generative Ads Model (GEM): The Central Brain Accelerating Ads Recommendation AI Innovation. Engineering at Meta, 10 November 2025. https://engineering.fb.com/2025/11/10/ml-applications/metas-generative-ads-model-gem-the-central-brain-accelerating-ads-recommendation-ai-innovation/
- Meta Platforms, Inc. 2025. Fourth Quarter 2024 Results Conference Call transcript, 29 January 2025. Remarks by Susan Li, Chief Financial Officer, page 6. https://s21.q4cdn.com/399680738/files/doc_financials/2024/q4/META-Q4-2024-Earnings-Call-Transcript.pdf
- Meta Platforms, Inc. 2026. Fourth Quarter 2025 Results Conference Call transcript, 28 January 2026. Remarks by Susan Li, Chief Financial Officer, pages 6 to 7. https://s21.q4cdn.com/399680738/files/doc_financials/2025/q4/META-Q4-2025-Earnings-Call-Transcript.pdf
- Meta Platforms, Inc. 2026. First Quarter 2026 Results Conference Call transcript, 29 April 2026. Remarks by Susan Li, Chief Financial Officer, page 7. https://s21.q4cdn.com/399680738/files/doc_financials/2026/q1/META-Q1-2026-Earnings-Call-Transcript.pdf
- Meta Platforms, Inc. 2026. Second Quarter 2026 Results Conference Call transcript, 29 July 2026. Remarks by Susan Li, Chief Financial Officer, page 7. https://s21.q4cdn.com/399680738/files/doc_financials/2026/q2/META-Q2-2026-Earnings-Call-Transcript.pdf
- Schumann DW, Petty RE, Clemons DS. 1990. Predicting the Effectiveness of Different Strategies of Advertising Variation: A Test of the Repetition-Variation Hypotheses. Journal of Consumer Research. 17(2):192-202. doi:10.1086/208549.