For years the advice was the same everywhere you looked. Volume wins. Flood the account. Fifty creatives, a hundred, three hundred. Let the algorithm sort it out.
Plenty of advertisers took that literally. They built accounts with hundreds of live ads, then watched performance flatten and could not work out why more was suddenly doing less.
Two things changed underneath them. In December 2024 Meta announced Andromeda, a new ads retrieval engine that picks a few thousand candidates out of tens of millions of ads before the ranking models ever see them. And Meta's ads system got very good at recognising when two ads are the same ad wearing a different hat.
Meta groups look-alike creative together. In Meta's own words, when multiple ads look or feel alike they are seen as variations of the same creative, and what the system learns is shared across the group. So 300 ads built on one hook and one visual style do not get 300 separate chances. You paid for 300 exports and bought a handful of shots. Fifteen different ideas can beat that.
The number that gets misread
You have probably seen the benchmark. Confect studied 3,014 ecommerce advertisers across 2025 and found the top-performing third run 395 live ads on average, against 296 for the bottom third. Agency decks screenshotted that stat and turned it into "make more ads."
We think they read the wrong half of it. A count of live ads says nothing about how many ideas sit inside them. An account can hold 395 ads built on forty ideas, or 296 built on four, and Meta has told you which one its system treats as more.
Andromeda does not reward how many files you uploaded. It rewards how many different ways you gave it to reach a human.
None of this means you stop producing. Common Thread Collective reports that brands testing 20 or more new ads per month see 65% higher return on ad spend than brands testing fewer than 10. Volume matters. Volume of sameness does not.
What counts as a different ad
This is where most accounts fall over, because "different" gets defined as "different colour grade." Meta's own example of what does not count is two ads with identical visuals and different call to action text. It asks for ads that are distinct in visuals, messaging and format. We break that into four dimensions.
Hook. The first three seconds and the first line of copy. A price hook, a problem hook, a myth busting hook and a founder story hook are four ads. Four crops of the same product shot are one.
Format. Static, carousel, talking head, screen recording, UGC selfie video, motion graphic, before and after. Each one looks and feels different in the feed, which is the test Meta describes.
Presenter and setting. A customer in their kitchen and a studio product shot are different ads even with an identical script, because the face and the room change everything the viewer sees.
Angle. The argument itself. Speed versus price versus risk reversal versus social proof versus fear of getting it wrong. Asset production can't fake this one. It takes strategy.
Just as useful is knowing what does not count: a new colour grade, a different font, a swapped music bed, a tighter crop, new wording on the call to action button. These are finishing decisions. Test them inside a concept if you want to, but never count them as new concepts. Our rule of thumb: if a designer could make the variant in ten minutes, treat it as the same ad.
The structure that goes with it
Creative diversity only pays if the account is built to let the system move budget freely. This is the shape to build toward.
- Fifteen to twenty distinct live ads. That range is our working number. Meta publishes no such rule. Fifteen real concepts, refreshed constantly, will tell you more than eighty variants.
- Consolidated campaign structure. Put the bulk of spend through an Advantage+ sales campaign (Meta renamed Advantage+ shopping campaigns in February 2025) instead of a museum of legacy ad sets that each hold a fraction of the data.
- Clean signal. Pixel and Conversions API running together, with a healthy Event Match Quality, which Meta defines as a score out of 10 for how well your customer information matches events to Meta accounts. Feed a prediction engine noisy conversion data and it predicts badly, however good the creative is.
Some of the lift people credit to "Andromeda creative" is really the second and third points: the same ads, in an account the system can finally read.
Clean signal gets skipped most, because nothing on the ads manager dashboard shows it. If you have never checked your Event Match Quality, start there. Our guide to the boring number quietly deciding your ad results walks through what it measures and how to raise it.
The fatigue trap underneath all of this
A second change makes the diversity problem urgent. ScaledOn's analysis puts effective ad lifespan at two to four weeks now, down from six to eight before Andromeda. That is one agency's estimate. Meta has published no such number.
Do the arithmetic anyway. If your winners burn out inside a month and you produce creative in quarterly bursts, you spend most of every quarter running ads the audience has already tuned out. Then you conclude that Meta stopped working, when really your production cadence fell behind the platform's metabolism.
The two changes compound. Fatigue means you need fresh creative more often, and grouping means fresh creative only counts if it is new in substance. Refresh every fortnight with recoloured versions of September's winner and you get the worst of both: the audience is tired of the idea, and the system sees nothing new to test. We go deeper on the fatigue side in your winning ad now wears out in weeks.
How to build the pipeline without a studio
You do not need a production company. You need a repeatable day.
One shoot day, structured properly, can feed a quarter. Book a half day with a real customer or two. Bring a phone and a clip mic, plus a shot list built around angles. Capture four hooks, three formats and two settings and you walk out with 24 different combinations before a single edit.
Then edit against the four dimensions above. If two exports differ only by font, delete one.
The full playbook is in one video shoot, 90 days of content.
Ship in weekly batches instead of monthly dumps. Weekly gives the system a rolling supply and gives you a clean read on which angle is carrying the account.
How to tell if your account has a sameness problem
No special tools needed. This takes an honest hour with your live ads.
- Export every live ad. Put the first frame and the first line of copy for each one side by side.
- Tag each ad on the four dimensions. Hook, format, presenter and setting, angle. Be strict. A new headline on the same argument is the same angle.
- Count the unique combinations. Ignore the ad count. The number of different concepts is closer to what the system is working with.
- Look at where spend is going. One or two concepts soaking up almost all the budget while dozens of near duplicates get scraps is a strong hint the grouping has already happened.
- Check the dates. How old is your top spending concept? If it has been running well past a month, plan its replacement now.
Run this audit and you will usually find a long list of ads and a short list of ideas. That gap is the work.
A worked example: the skincare brand with 120 ads
Say you run a hypothetical Melbourne skincare brand spending $15,000 a month on Meta. You have 120 live ads. Tag them and you find the truth: nearly all of them are the same product shot on a marble background with a "clean ingredients" message, in different crops and fonts. Three concepts, dressed up as 120.
The rebuild starts with the list of reasons people buy. Sensitive skin that reacts to everything. A routine that takes too long. Scepticism about "natural" claims. Wanting to look less tired. Distrust of influencer hype. Each becomes an angle. Then you book a half day with two real customers. You capture a founder story hook, a myth busting hook about ingredients, a problem hook about reactions and a routine-in-sixty-seconds hook, as talking heads, selfie UGC and screen-recorded demos. Half is shot in a bathroom, half in a kitchen.
You cut 120 ads down to about fifteen distinct concepts, consolidate campaigns, fix the Conversions API setup, and ship three or four new concepts every week. No guarantees, but the account now gives the system fifteen different ways to find a buyer instead of three, and gives you a clean read on which angles are selling.
The uncomfortable part
Creative strategy is now the constraint, and you cannot buy it off a shelf.
Meta will bid for you, find the audience for you and choose the placement for you. What is left is knowing your customer well enough to have fifteen different true things to say to them.
Most businesses do not have fifteen. They have three, and eleven ways of restating the second one.
So do the work. Open a blank doc and write out every distinct reason a person buys from you, every objection that stops them, and every proof you can put on camera. Get that list past 20 and you will never be short of a new concept to test.
That is also where outside help earns its fee. Pushing buttons in ads manager is the part the platform increasingly does for you. Pulling those 20 true things out of your business and turning them into distinct creative every week is the part it can't. Our advertising team builds that pipeline: the angle list, the shoot days, the edit against the four dimensions, and the account structure that lets Meta's system use it.
Frequently asked questions
What is Meta Andromeda?
Andromeda is Meta's ads retrieval engine, announced in December 2024. It scans tens of millions of ads and shortlists a few thousand for the ranking models that choose what each person sees. Alongside it, Meta's ads system groups creative that looks or feels alike and treats it as variations of one creative, so creative diversity matters more than ad count.
How many ads should I run on Meta after Andromeda?
Meta gives no official number. Our working range is fifteen to twenty distinct live ads instead of dozens of variants. Distinct means a different hook or format, a different presenter or setting, or a new angle. A new colour or font does not count. Keep refreshing with new concepts, since Common Thread Collective reports brands testing 20 or more new ads a month see 65% higher return on ad spend than those testing fewer than 10.
Does changing the colour or headline make a new ad for Meta?
Usually not. Meta says ads that look or feel alike are seen as variations of the same creative, and its own example is two ads with identical visuals and different call to action text. To register as different, change something structural, such as the opening hook, the format, who is on screen and where, or the argument you are making to the buyer.
Should I still use Advantage+ campaigns?
For most accounts, consolidating spend is the right move, with the bulk of it running through an Advantage+ sales campaign instead of many small legacy ad sets. Consolidation gives the system more data per campaign and more freedom to move budget toward what works. It only pays off when paired with diverse creative and clean conversion signal from the Pixel and Conversions API.