Andromeda 2.0 Doesn’t Care That You Launched 30 Ads

How Different Is Different Enough? Creative Diversity and Entity ID Under Andromeda 2.0


Why It Matters

“I launched 30 ads and Meta spent on one” usually means Meta counted them as the same ad. Under Andromeda 2.0, ads are classified at the concept level through an entity ID that reads setting, format, and audio, so 30 ads sharing a style, location, and persona amount to one auction ticket. An ad that never makes the shortlist never competes, no matter your bid, budget, or structure. Swapping voiceovers, music, captions, or headlines won’t un-cluster them. This post gives you a way to tell creative fatigue (a winner that stopped working) from similarity (ads that never got spend), since the fixes are opposites. It also offers a 20-minute grid audit with concentration thresholds and a cheaper path to diversity: re-express winning concepts in new formats, tones, and angles instead of producing everything from scratch. That concentration is often what’s inflating CPMs, a problem inside your own account that no amount of bid or budget tweaking will fix.

Who This Is For

Built for Founders members running Meta ads or managing creative strategy, especially media buyers and creative strategists whose new ads aren’t getting spend or whose CPMs are climbing. Most useful if you’re launching batches of ads from a single shoot, testing messaging with headline swaps, or over-segmenting ad sets to compensate for a thin creative library.


The most common version of this question in our community goes something like: I launched 30 ads and Meta spent on one of them. What did I do wrong?

It’s nothing you’d recognize as wrong; you made 30 ads that Meta decided were the same ad.

The problem is that the threshold is invisible because you can't see entity ID in Ads Manager. Members have been asking for months for Meta to simply tell us when assets have been bundled together, and so far we’ve gotten nothing. 

The practical question becomes how different two ads have to be before the system counts them separately, and that's answerable even if the exact mechanism isn't visible.

What's Happening Under the Hood

Understanding Meta’s ad architecture tells you where your leverage is.

There are three layers. Lattice is the foundation, one unified infrastructure holding every ad, campaign, and signal. Andromeda 2.0 is the retrieval layer, scanning millions of ads in milliseconds and narrowing them down to a shortlist of roughly a thousand candidates. Then GEM, Meta's generative ads recommendation model, makes the final pick from that shortlist based on everything the system knows about the person.

If your ad doesn't make the Andromeda 2.0 shortlist, your bid, budget, and campaign structure don't matter. You were never in the auction.

The way Andromeda 2.0 decides what makes the shortlist is by classifying your ads at the concept level, not the file level. Think of entity ID as a semantic fingerprint. The signals it reads include background and setting, so the same product in a gym versus a living room reads as different context and potentially a different persona. They also include visuals and format, so an iPhone UGC clip reads differently from a studio production. And they include audio, so a voiceover ad reads differently from a fast cut edit.

If your ads are too close across those signals, they get clustered as one entity. Thirty ads sharing a style, a visual language, and a target persona is one auction ticket. 

My favorite example of how easy this is to do accidentally came from somebody who tells this story on himself. When Andromeda 2.0 went live, his team worked with a creator and had them film three deliberately different styles of ads for three different personas and use cases. His instinct and intent were both good, but they filmed all three in the same location. Voila: all the ads were grouped into one entity ID.

Two Different Problems That Look Identical 

Before you try fix anything, work out which of two problems you have, because the fixes are opposites.

Problem one: the ad worked, then stopped working. That's ordinary creative fatigue. Your audience saw it too many times. The fix is to refresh it with a new hook and/or new talent and keep iterating on the winning concept. This is the comfortable problem.

Problem two: the ads never got spend at all. That's similarity. Meta bundled them before they ever reached the auction. Refreshing won’t help here, because the concepts are too alike to begin with and a refresh produces another member of the same cluster. The only fix is making genuinely different ads. 


The email marketers actually read.
Actionable ideas you can put to work the same day you get them.


What Doesn’t Get Counted as Different

None of the following will un-cluster your ads in Meta’s eyes:

The same visual frames with a different voiceover. A different video from the same shoot. Multiple hooks that say the same thing with the words in a different order. Changing the music. Changing the captions. A 60-second video trimmed to 50. A different creator reading the same script in the same setting.

Headline swaps belong on this list now too. Most of our community has moved off the same image with five different headlines, because Andromeda 2.0 appears to treat a headline only change as basically the same ad. Nobody has shared a clean head to head on it yet, so take that as the current consensus rather than proof. But the practical adjustment people have made is good: if you want to test messaging angles, give each headline its own visual – meaning different background, texture, and composition. One member on a recent call said he's switched entirely from headline variants to visual variants to add a variable to the test. This isn't ideal in a purist sense, but it gets the ad into the auction as its own entity.

The Grid Audit, with Thresholds

You can find out whether you have a concentration problem in about 20 minutes, like so:

  • Pull every active ad into one view. Tag each one across three dimensions.

  • Format: static, video, carousel, flex.

  • Tone: UGC, founder led, polished brand, meme style.

  • Angle: problem aware, product demo, social proof, lifestyle, comparison.

Now look for concentration. The rough thresholds worth flagging: any single format sitting at 50 percent or more of your active ads is a format gap. Any single tone at 40 percent or more is a tone gap. Any single angle at 30 percent or more is an angle gap.

That concentration is very often what's inflating your CPMs, and it's a much better explanation than the ones people reach for first, like more competition in the auction and post-holiday budget shifts. Those are real, but a lot of the time the problem is inside your own account, and because you've given the system fewer options to work with, it's bidding you up to reach the same people.

On volume, the number that holds up across accounts is roughly 10 to 15 genuinely distinct creative concepts per campaign as a working target, with the caveat that it varies a lot by spend, niche, and account history. If your library has fewer than six truly distinct concepts, you're running light. 

Remember: they don't all have to launch at once. Rotate in new ones every week or two and replace the underperformers while keeping a spread of styles.

Get Diversity without Tripling Your Production Budget

The grid is also the production brief. Instead of more ideas, you need more wrappers for the ideas that already work.

Take a concept that's winning and deliberately move it across the grid. The UGC script becomes a founder-led script. The static becomes a carousel. The comparison ad becomes a lifestyle ad. From Andromeda's point of view those are different entities because the setting, format, and audio all changed.

This is also where the iteration data backs you up. At least half of the ads in our members' accounts that have crossed $20,000 in spend are second- or third-level iterations of an earlier concept rather than brand new swings. So the move is to express what works again in a way the system reads as new.

Where you do need net new ideas, the highest-return work I've seen is going narrower on the person rather than broader. Instead of demographics, look at sub avatars, meaning a specific behavior or pain point. Not women 25 to 44 who like wellness, but the person who gets bloated after pizza. One of the best diagnostics on this I've heard: go search your problem on Instagram, look at the organic content that performs, then go look at your own ad library. If your content doesn't look anything like the content that does well on that topic, that's your gap.

And keep a slice of the account deliberately weird. I tell brands to hold 20 to 30 percent of their creative on absurdly out-there concepts at all times rather than refine yourself into a corner where every ad is a cousin of every other ad and you've got nothing ready when the winner dies.

Stop Doing These Two Things

Stop killing new ads fast. Andromeda 2.0 takes longer to work out which creative fits which user segment than the old system did. Give a new ad real time and spend before you make a call on it, or you'll keep chopping ads that were about to find their audience.

Stop fragmenting the structure to compensate. Seven ad sets all testing the same creative concept is over-segmenting in a world that no longer rewards that control. Flatten the structure, put the diverse library in one place, and let the system cross learn between ad types. The control you used to exert through targeting now gets exerted through what you put in the library.

The short version

Different enough means different setting, format, audio, and ideally the person being spoken to. If you’ve changed nothing but the words, it's the same ad. If you changed the room, the camera, the voice, and who it's for, it's a new one.

Go pull your active ads into a grid this week and count. If most of them land in one square, you've found your CPM problem without needing a single new idea.

Trusted by agency owners, brand founders, and marketers everywhere, our advertising membership is where the top 1% of digital marketers go when they need help. Meta ads, creative strategy, Google ads, CRO, email and SMS, we've got experts on it all. Come hang out with us.


In your first hour as a member, you’ll unlock 12+ hours of Meta ad training, $6,500 worth of premium courses, and a Slack of 600+ marketers who answer questions around the clock.


Andrew Foxwell | Co-Founder of Foxwell Digital

Co-Founder of Foxwell Digital, a social media advisory firm focused on honesty and transparency across paid social. Through its membership offerings, online courses, account management, and consulting services, Foxwell Digital helps brands and agencies make better decisions and scale sustainably.

https://foxwellfounders.com/
Next
Next

Meta Account Disabled or Ads Rejected in Q4? A Response Plan for Agencies