Free tool
Cannibalization detector
Are two ad sets fighting each other in the auction? Drop in their names + CPLs — get a verdict.
Want this running 24/7 on your account?
Free tool
Are two ad sets fighting each other in the auction? Drop in their names + CPLs — get a verdict.
Want this running 24/7 on your account?
It takes two ad set names, each one's cost per lead, and your account median CPL, then returns three measurements and a verdict. Name overlap is a Jaccard similarity on the tokens in the two names: both are lowercased, everything that is not a letter or digit becomes a space, tokens of two characters or fewer are dropped, and the score is the shared tokens divided by all distinct tokens across both names.
Peer lift is the average of the two CPLs measured against the account median — a +40% reading means the pair costs 1.4× what a typical ad set in your account costs. CPL spread is the higher CPL divided by the lower one, which says whether the two are performing alike or whether one is clearly better.
The pair is called cannibalizing only when all three conditions hold at once: name overlap at 34% or higher, peer lift at 1.4× the median or higher, and CPL spread at 1.3× or lower. The logic is that a cannibal pair looks like two versions of the same ad set, both paying an auction premium, both landing in the same place.
The near misses are diagnoses in their own right. Names overlap but the CPL spread is wide? The audiences are genuinely different and one is simply better — scale that one. Both above median but the names look distinct? These are independent under-performers, so pause or refresh rather than consolidate. Neither condition met? Nothing to fix here.
Use one currency consistently across all three CPL fields. The comparison is a ratio, so the unit cancels out — the euro label on the inputs does not constrain what you enter.
Consolidate budget into the larger ad set rather than splitting it. Two ad sets each gathering half the conversions both stay stuck in learning; one ad set gathering all of them exits faster and gets better delivery — check the timing with the learning-phase estimator. If you have a reason to keep both live, add each audience as an exclusion on the other so a person can only be served by one.
This tool tests one pair at a time from numbers you type in. To see where the budget should land across every campaign at once, run the budget reallocator. AutoAdy runs the same cannibalization check continuously against live delivery data, where it can see the actual audience definitions rather than inferring them from names.
Last updated August 31, 2026. All 17 calculators and AI helpers are listed on the free tools hub.
FAQ
Cannibalization is when two of your own ad sets compete for the same impressions. Because Meta runs a single auction per impression, your second ad set is not adding reach so much as raising the price of reach you were already winning. The tell is a pair of near-identical ad sets that are both more expensive than the rest of the account while performing about the same as each other.
Three conditions must all hold. Name similarity of 34% or more, measured as Jaccard token overlap after lowercasing and dropping words of two characters or fewer. Average CPL at least 40% above the account median. And a CPL spread of 1.3× or less between the two, meaning they perform similarly. Any one of those on its own is a different diagnosis, which is why the tool reports all three numbers.
Names are the only structural signal you can supply without connecting an account, and in practice media buyers encode the targeting in the name — Lookalike Broad USA versus Lookalike Broad USA Test. It is a proxy, not a measurement. If your naming is inconsistent, the similarity score will understate the real overlap and you should lean on the CPL figures instead.