Your 30 ads are really 3 (Meta counts ideas, not files)
Chinmay Raibagkar · has watched 30 “new” ads spend like 3🧩 with a concept-cluster audit, not a lecturePublished October 4, 2026
The 60-second version
- Meta now uses the ad itself to find people, so near-identical ads compete for the same people instead of reaching new ones.
- A new concept changes the angle, the person or the format — colours, crops and headline tweaks are variants, not concepts.
- The fix: tag every ad, keep the best of each concept, brief new angles, and track effective ads next to ads live.
The creative team is proud of the numbers: thirty new ads this month. Fourteen versions of the "flat 30% off" static, nine cuts of the founder talking to camera, seven before-and-after carousels. New colours, new headlines, new crops.
Two weeks later the report is strange. Three ads have taken 81% of the spend. Nineteen ads have spent less than ₹500 each. CPMs are creeping up, and CPA is drifting with them.
The ads are not bad. Meta just does not see thirty ads. It sees three ideas, each wearing ten outfits. same idea, new shirt Since Meta rebuilt how it picks which ads to show, small variations of one idea mostly compete with each other instead of reaching new people. The job now is to count concepts, not files.
A 30-ad account, audited
Creative is the targeting now
Meta's ad system works in two steps. First it retrieves a shortlist of ads that might suit a person, then it ranks that shortlist in the auction. Andromeda is the name Meta gave its newer retrieval engine, announced at the end of 2024. It is much better at matching ads to people — using what the ad itself says and shows.
That changes what your targeting settings do, and what your ads do.
🎯 Targeting finds people
You chose the audience; the ad just had to convert it. Many ad sets, narrow interests, and small creative tests inside each.
- Audience settings did most of the work
- Five ad sets × five similar ads was normal
- Variants were cheap tests
🧩 The ad finds people
Broad audience, and the ad's message decides who sees it. A discount ad finds bargain hunters; a founder story finds people who care about the brand.
- Each distinct concept opens a new pocket of people
- Look-alike ads share one pocket and split it
- Variety of ideas beats volume of files
If every ad says the same thing, the system can only find one kind of person — no matter how broad your targeting is.
How strict is "similar"? Meta does not publish a threshold. Agency playbooks from 2026 commonly suggest 8–12 clearly different concepts per campaign and a refresh every two to three weeks. Treat those as starting points, not rules — your spend concentration (below) is the real test.
Variants vs concepts
A concept is a different reason to buy, told to a different person, in a different way. A variant is the same thing with a new coat of paint. Tap the box below to see your uploads the way the system groups them.
- Discount hook · static — 14 ads, colour, headline and crop changes
- Founder story · UGC video — 9 ads, same script, different cuts
- Before/after · carousel — 7 ads, same images, new order
Three questions decide whether an ad is a new concept or just a variant:
Different angle?
A new reason to buy: price, quality, speed, status, a problem solved. "30% off" and "25% off" are the same angle.
Different person?
Who is it talking to — a first-time buyer, a gifter, a professional, a parent? Same person, same pocket.
Different format?
Static, talking-head video, demo, carousel, meme. A new crop of the same static is not a new format.
If an ad changes none of the three, it is a variant. Change at least one — ideally two — and you have a new concept that can reach a new group of people.
Count your real concepts
The fastest health check is how concentrated your spend is. If a handful of ads take nearly everything, the system has decided the rest are duplicates. Plug in your numbers:
Count groups that differ in angle, person or format
The system sees only a handful of ads, so most uploads split the same people. Delete near-copies and brief genuinely new angles before adding more files.
To do this with data instead of a gut feel, tag each ad once in a small sheet (ad ID, angle, person, format) and let SQL count concepts and spend concentration. "Effective ads" turns spend shares into one number: how many equally funded ads your spend actually behaves like.
If effective_ads sits near concepts — or below it — the system is treating your variants as copies. Thirty uploads with an effective count of four is the classic picture.
Building a mix the system can tell apart
The concept rebuild, in order
Process FlowTag every live ad
Angle, person, format — one row per ad in a sheet. It takes thirty minutes and makes every later step obvious.
Keep the best variant of each concept
Within each concept, keep the one or two ads with the best CPA. Pause the rest; they were splitting the same people.
Fill the empty boxes
List angles and people you are not speaking to yet. Brief new concepts for those gaps — not new versions of the winner.
Launch concepts together
Add new concepts to one broad ad set so the system can compare them in the same auction, instead of scattering them across many small ad sets.
Refresh on a rhythm
Every two to three weeks, retire the most tired concept and add a new one. Watch frequency and CTR decay to time it.
Don't confuse this with fatigue. Look-alike ads cause spend concentration from day one. Creative fatigue is a good ad wearing out over weeks as frequency climbs. Both need new creative, but the first needs different creative. See creative fatigue vs audience saturation for the second.
Briefing habits for the Andromeda era share this with the creative team →
- Brief concepts, not sizes — every brief names a new angle, person or format. "Five more versions of the winner" is not a brief.
- Count concepts in the weekly report — show concepts live and effective ads next to ads live, so variety is tracked like spend.
- Retire before you add — when a new concept launches, pause the weakest near-copy so concepts don't drown in variants.
Quick gut-check
One question. If you get it, the whole post clicks. no maths, promise
Your campaign has 24 ads. The top 2 take 78% of spend, and 16 ads have barely spent. What is the best next move?
Frequently asked questions
What is Meta Andromeda?
Andromeda is Meta's ad retrieval system — the step that shortlists which ads could be shown to a person before the auction ranks them. Meta introduced it at the end of 2024. Because it matches people using the content of each ad, a varied set of ads reaches a wider range of people than many near-identical ones.
How many ads should I run per campaign?
There is no official number. A practical target is enough clearly different concepts to reach several groups of people — many teams aim for 8–12 — with one or two strong executions of each. The number of files matters far less than the number of distinct ideas.
Do small changes like new colours or headlines still help?
They help a little inside a concept, mostly by fighting fatigue. They do not open new audiences. Use them to keep a good concept fresh, not to fill a campaign.
The summary
- Meta now uses what an ad says and shows to decide who sees it, so creative does much of the targeting.
- Near-identical ads compete for the same people; the system quickly funnels spend to one or two of them.
- A concept differs in angle, person or format. Anything else is a variant.
- Spend concentration and effective ad count reveal how many ads the system really sees.
- Rebuild around fewer, more different ads, and refresh concepts every few weeks.
Takeaways for your next report
- Meta sees ideas, not files: 30 near-identical ads behave like three.
- A new concept changes the angle, the person it speaks to, or the format — preferably two of the three.
- If the top three ads take most of the spend, your variants are being treated as copies.
- Effective ads = 1 ÷ sum of squared spend shares. Track it next to ads live.
- Brief new angles for people you don't reach yet, and retire a near-copy for every concept you add.
Frequency & Fatigue Visualizer
Drag impressions, audience size and creative decay to see when frequency turns from reach into waste — and what cap and refresh cadence that implies.
Chinmay Raibagkar
About author →Founder of DataLens AI. He helps non-technical teams read their ad and database numbers with confidence — which number to trust, what to do next, and what to ignore.