Run a holdout, not a correlation. The only honest proof is turning brand off inside Performance Max — or pausing the campaign in a matched geography — and watching what happens to total brand conversions, not to the campaign's own numbers.

On this page

• Why doesn't falling brand spend prove cannibalisation?
• What should I measure to test brand cannibalisation?
• How do I run a brand cannibalisation holdout?
• What measurement change can fake cannibalisation?
• Is Performance Max brand traffic incremental or intercepted?

Why doesn't falling brand spend prove cannibalisation?

The standard argument is that Performance Max spend rose while brand campaign spend fell, therefore cannibalisation. That is a correlation between two campaigns competing for the same queries, which is also exactly what you would see if demand simply shifted. It cannot separate incremental from intercepted.

What should I measure to test brand cannibalisation?

Total brand conversions and total brand cost, across every campaign that touches brand queries. If Performance Max is cannibalising, moving budget between the two changes the split and leaves the total roughly flat. If it is incremental, the total moves.

How do I run a brand cannibalisation holdout?

1. Establish the baseline. Brand conversions and brand cost across all campaigns, over a settled window. Never read days younger than 4 — measured on an apparel account, a window ending 1 day ago backfilled +83.7% on re-read, while windows 4+ days old backfilled 0.0%.

2. Apply brand exclusions to Performance Max, or run a geographic holdout if you have enough volume to split.

3. Wait for a settled read. Do not stack any other change in the window.

4. Compare totals, not campaign lines.

What measurement change can fake cannibalisation?

Before running any of this, check whether the counted conversion actions changed during your comparison window. On one account an extra conversion action was folded into the counted metric for exactly seven months, and every ratio built on it in that period was measuring something different from the months either side.

A cannibalisation test across a ruler change is not a test.

Is Performance Max brand traffic incremental or intercepted?

Some of both. Performance Max does pick up brand queries, and on most accounts a share of them would have converted anyway. The question is not whether cannibalisation exists — it is whether the incremental portion justifies the blended cost. That is an economics question, and it needs the holdout.

Related questions

How do I find out how much of my Performance Max spend is going to my own brand searches?

Performance Max does not report brand spend as a line item, so you build it from the search terms it does expose plus the network split — and accept that the figure is a floor, not a total. Read the full answer.

I added brand exclusions to Performance Max but my brand terms are still showing up — why?

Brand exclusions filter what the campaign can target, not what it can be credited for, and they work on the brand list you gave Google — near-variants, misspellings and brand-plus-modifier queries routinely fall outside it. Read the full answer.

I excluded brand from Performance Max and my results got worse. Should I put it back?

Probably not yet — check whether the excluded traffic actually went somewhere, or whether you strangled the campaign and its queries fell into a worse catcher. That is a structural failure that looks identical to "the exclusion was wrong." Read the full answer.

Written by Usama Khan, who runs Google Ads for e-commerce brands and home-services companies at AdventurAds. Figures are from accounts under active management and are anonymised. Last reviewed 2026-08-26.