Meta Andromeda Explained: What It Is and How It Changes Your Facebook and Instagram Ads (2026)
Meta Andromeda is the AI system that decides which ads are even considered for a person when they open Facebook or Instagram. Meta announced it in December 2024 and has been rolling it out since. It changed a core rule of Meta advertising: your ad creative now does most of the targeting work, not your audience settings.
I put off making a video on this because most of what’s out there is either hype or panic. So I went through Meta’s own announcements, its latest earnings call, trade press and the few real datasets available. This post is the plain-English version.
The 30-second version
- Andromeda is the first step of ad delivery. It builds a shortlist of ads for each person.
- Meta says it uses far bigger AI models than before, and that they read your ad itself (visuals, message, format), not just your audience settings.
- Because of this, five near-identical ads may count as one. Genuinely different ideas get more chances.
- Meta is also removing manual controls and changed how it counts conversions in 2026. Both make results harder to compare with the past.
- Almost all the success numbers come from Meta itself, so judge results with your own sales data.
What is Meta Andromeda, in simple words?
Meta Andromeda is Meta’s ad retrieval engine. When you open Instagram, Meta has tens of millions of live ads to choose from. Andromeda quickly picks a shortlist of a few thousand that might suit you. A second system then decides which one actually appears.
Think of hiring. Millions of people apply, so someone shortlists a few thousand CVs. Then interviewers pick the best few. Then one person gets the offer.
| Stage | Hiring analogy | What it is at Meta |
|---|---|---|
| 1. Retrieval | Shortlisting CVs | Andromeda picks a few thousand ads |
| 2. Ranking | Interviews | GEM, Lattice and other models predict who will click or buy |
| 3. Auction | Final offer | Bids and predicted quality decide which ad shows |
If your ad doesn’t make the shortlist, your budget and bid never get a chance to matter.
Why did Meta build it?
Meta built Andromeda because AI tools are creating far more ad variations than the old system could sort through. Meta’s engineers said retrieval had to handle a huge increase in ads, quickly, for every person, every time they open the app.
Here are Meta’s own reported numbers from its December 2024 engineering post:
- A 10,000x increase in model complexity at the retrieval stage.
- +6% recall (finding more of the right ads) and +8% ad quality, on selected segments.
- These are Meta’s figures on selected segments. They are not what every advertiser will see.
How is it different from the old way?
The old system leaned on the audience you chose. Andromeda leans on what your ad actually shows and says.
| Old way | New way | |
|---|---|---|
| Who sees my ad? | Decided mostly by my interests and lookalike settings | Decided mostly by my ad content |
| Many similar ads | Good, because more tests | Weak, because similar ads may be grouped as one |
| Campaign setup | Many ad sets, narrow audiences | Fewer ad sets, broad audiences |
| My main lever | Targeting | Creative |
Meta’s own line since 2025 has been that creative variety is now the best way to reach the right people. Jon Loomer, one of the most-followed Meta ads educators, puts it simply. Targeting largely happens inside the ad now, so keep audiences broad and put your effort into the ads.
The 5 big updates, explained simply
1. Andromeda: your creative became your targeting
Update 1 is Andromeda itself. It shortlists ads by reading the ad, so different ideas beat small edits of the same idea.
Many people in the industry say Andromeda groups ads that look or sound alike into one cluster (often called an “Entity ID”). So fifty copies with different headline colors may get roughly one ticket to the shortlist. Meta hasn’t published the exact grouping rules, so treat tips like “keep similarity below 40%” as rules of thumb, not official limits.
In one line: Before, you told Meta who to show your ad to. Now your ad tells Meta who to show it to.
What “different” means in practice: a different person speaking, a different problem highlighted, a different format (video, image, testimonial) or a different opening few seconds.
2. GEM and Lattice: the ranking brain
Update 2 is the ranking layer. After Andromeda builds the shortlist, models like GEM decide which ads are most likely to work for each person.
- GEM (Generative Ads Recommendation Model) is a very large model trained on ad and organic activity. Meta says it is 4x more efficient than its earlier ranking models. It reports +5% ad conversions on Instagram and +3% on Facebook Feed.
- Lattice merges many separate ranking models into one.
In one line: Andromeda picks the shortlist. GEM decides who gets the seat.
A quick myth check. Some articles say “Andromeda is 4x more efficient”, but that 4x claim is about GEM. GEM also gets misnamed in many agency blogs. It stands for Generative Ads Recommendation Model.
3. Advantage+ by default, and fewer manual controls
Update 3 is Meta’s shift toward automation. Advantage+ campaigns are the default for many objectives, and manual controls keep shrinking.
In August 2026, advertisers began seeing a notice that excluding placements, platforms, devices and operating systems will no longer be available at the ad-set level. This is a test for now, limited to Sales and Leads campaigns, and Meta hasn’t formally announced it. The replacement is “value rules”, which can cut a bid by at most 90%. That means a placement you dislike could still get some spend.
Meta says advertisers aren’t penalized for opting out of its AI features. It also says heavy manual settings can switch a campaign to “Advantage+ off” and that the system may nudge you to turn automation back on.
In one line: Meta is moving from “here are the dials” to “tell me the goal and the budget.”
4. Measurement changed in March 2026
Update 4 is a reporting change. Since 3 March 2026, “click-through” conversions count only real link clicks.
Likes, saves, shares and comments now go into a separate bucket called engage-through, with a one-day window. Billing didn’t change, but reported conversions in some accounts dropped noticeably. Agencies estimate anywhere from about 15% to 40%. Those figures come from agency write-ups, not from Meta.
This matters for the Andromeda story. If your results looked worse in 2026, part of that could be how Meta now counts, not how your ads performed. Don’t compare pre-March and post-March numbers directly.
In one line: Meta didn’t just change how ads are delivered. It changed how it counts the wins.
5. AI creative, AI agents and the push toward full automation
Update 5 is AI doing more of the ad work, from making creatives to analyzing your account.
- Meta offers generative tools such as turning product photos into video ads. On its Q2 2026 call, Meta said more than 9 million small businesses use at least one AI creative tool.
- In August 2026, Meta AI gained the ability to review your ad performance and suggest changes. Heavier use is expected to sit behind a paid plan called Meta One, launched on 15 September 2026.
- Meta’s stated goal, per a 2025 Wall Street Journal report, is fully automated ads by the end of 2026. Media buyers interviewed by Marketing Brew in April 2026 think that is much further away.
A note on AI-made ads: an NYU/Emory study found AI-generated ads got up to 19% more clicks than human-made ones, but telling people they were AI-generated cut effectiveness by up to 31.5%. That study was run on Google Ads, not Meta, and disclosure rules like the EU’s are now in force.
In one line: Meta wants you to bring the goal and the budget. The AI brings the ad.
What’s coming next?
The next step is LLM-based ad matching. On its Q2 2026 earnings call, Meta said it introduced the “Meta Generative Recommender”. It uses large language models to reason about ad content and user preferences together. Meta says it put its first generative model into the ads retrieval stage, which is where Andromeda works.
Meta also reported +8.3% ad clicks and +15.7% conversions on Facebook from its latest AI improvements. These are Meta’s own early-deployment numbers. No independent benchmark exists.
Is Andromeda good or bad for advertisers? What the data says
The honest answer is that it depends on who you are. Meta reports gains, but independent data is limited and mixed.
Meta’s side: Meta’s Q2 2026 results show ad revenue up 27% to $59.4 billion, with ad impressions up 14% and the average price per ad up 12%.
The outside data: The largest public study I found comes from Confect, a company that sells catalog-ad tools. It covered 3,014 e-commerce advertisers and about $834 million in 2025 ad spend.
- Average ROAS fell about 7% during the rollout.
- The top third of advertisers by ROAS fell about 31%.
- Mid-sized advertisers were the only size group that improved.
Treat this with care. Confect is a vendor, all the advertisers were its own customers, it only covers e-commerce, and the timing lines up with the rollout without proving Andromeda was the cause.
What practitioners say:
- Good: Simpler account setups, less manual work, and rewards for advertisers who can produce fresh, different creative quickly.
- Bad: Agencies quoted by Marketing Brew say Meta keeps switching on new automation, that ads drift into weak placements, and that Meta’s own creative AI often underperformed their outside tools.
- “It’s overblown”: Some experts argue the algorithm always favored variety and is now simply better at spotting sameness, so the advice hasn’t changed much.
What should you do now? 7 practical steps
- Make different ideas, not different edits. Change the person, the problem, the format and the first few seconds.
- Simplify your account. Fewer campaigns, fewer ad sets, broad audiences.
- Fix your data signal. Set up the Conversions API and check event quality, because ranking models learn from it.
- Reset your baselines. Start comparing from mid-March 2026 onward, and view click-through and engage-through separately.
- Judge with your own numbers. Compare Meta’s reports against your real sales, leads or CRM data.
- Move brand-safety rules to account level. If placement exclusions leave ad sets, use account-level placement controls.
- Budget for higher costs. Plan for a price increase and offset it with stronger creative, not just more spend.
Common Andromeda myths
| Myth | What’s actually true |
|---|---|
| “Andromeda is 4x more efficient” | The 4x figure is Meta’s claim about GEM |
| “Post 100 variations and Meta sees one ad” | Directionally supported by practitioners, but Meta hasn’t published the rules |
| “Opting out of Advantage+ hurts delivery” | Meta says it doesn’t, but heavy manual settings can flip the campaign to “Advantage+ off” |
| “Andromeda is why my conversions dropped” | It may be part of it, but the March 2026 attribution change alone can explain a drop |
| “Ads only last 2 to 4 weeks now” | That comes from a vendor’s write-up, not from Meta |
FAQ
What is Meta Andromeda in simple terms?
It’s Meta’s AI shortlist system for ads. It picks a few thousand ads out of tens of millions for each person, before the final ad is chosen.
When did Andromeda launch?
Meta announced it in December 2024 and rolled it out over 2025. Some vendors report a full global rollout by October 2025, but I couldn’t find that date stated by Meta.
Do I still need interest targeting?
Mostly no. Meta treats interests as hints under Advantage+ Audience, and experts like Jon Loomer recommend broad audiences with varied creative.
How many ads should I run in one ad set?
Meta dropped its old “no more than six” guidance in early 2025. There’s no official number now. Loomer recommends going beyond two or three ads, but says quality and variety matter more than count.
Is Andromeda the same as Advantage+?
No. Andromeda is the retrieval engine that works across campaign types. Advantage+ is a set of automation tools and campaign types.
Is Meta One necessary for advertisers?
Not for running ads. It’s a paid business subscription launched on 15 September 2026, and most of what it gates is insights and organic tools.
Where can I verify all this?
The sources below. Meta’s engineering posts and earnings materials are the primary sources. The rest is trade press and vendor data.
Sources
- Meta Engineering, Andromeda announcement (Dec 2024)
- Meta Engineering, GEM overview (Nov 2025)
- Meta, Q2 2026 results and earnings call transcript
- Jon Loomer, Meta Andromeda: What It Means for Your Ad Strategy
- Confect, Andromeda study (vendor data, March 2026)
- Marketing Brew, How Meta’s AI push is changing ad creation (April 2026)
- Social Media Today, Meta removes option to exclude ad placements (August 2026)
- Dataslayer, Meta attribution change 2026
- Search Engine Land, How Andromeda and GEM work together