If Performance Max is beating your Search campaigns on reported cost per lead, work out who generated the click before you move a single dollar of budget. In a Performance Max vs Search comparison the two campaigns draw from the same query pool, and a published tie-break rule decides which one gets counted. Most advertisers have never read that rule.
Here it is. A Search campaign with a keyword that exactly matches the user’s query, or is spell-corrected to it, is preferred over Performance Max. Where no exact match keyword covers the query, the campaign with the higher Ad Rank serves. Google states this in its documentation on how Performance Max interacts with other campaigns. So the split you see in your campaign table is part bidding rule and part genuine incremental reach, and reported CPA cannot tell you which is which.
Performance Max vs Search: how Google decides which one serves
The documented example uses plumbers. If someone types “plumbrs” and your account holds both the exact match keyword [plumbers] and a broad match keyword, the exact match is prioritised first and the Search campaign takes the impression. Strip that exact match keyword out and the same query becomes a coin toss decided by Ad Rank, which PMax often wins because it bids across a wider signal set.
(Yes, this rule lives in a help article almost nobody opens. No, nothing in the Google Ads interface tells you it is happening while you sit there comparing two campaigns with wildly different CPAs.)
| Query situation | Which campaign serves |
|---|---|
| Exact match keyword [plumbers] exists, someone searches plumbers | Search campaign |
| Query spell-corrects to your exact match keyword | Search campaign |
| Only broad or phrase match keywords cover the query | Higher Ad Rank wins |
| No keyword in the account covers the query at all | Performance Max |
That last row is the only one where PMax is unambiguously adding reach.
One thing changed recently, and it widens the overlap rather than narrowing it. Google’s developer blog, published 13 August 2026, confirmed that campaign level language targeting is being removed from Search campaigns in late September 2026, and that on Google Search, Performance Max will match the same way Search does. The CampaignCriterion.language field now throws ContextError.OPERATION_NOT_PERMITTED_FOR_CONTEXT if you try to set it on a Search campaign. Language settings still guide delivery on YouTube, Display, Discover and Gmail. If you were using language targeting to keep the two campaign types apart in a multilingual market, that lever is going away, and you can read the detail in Google’s own announcement.
Reported credit and actual cause are two different numbers
Attribution inside Google Ads answers a narrow question: which campaign touched the converting user, under whichever model you selected. It doesn’t answer whether that conversion needed the ad. Brand queries are the clearest case, because somebody typing your company name was already on the way. PMax will take credit for that click and post a flattering CPA next to it.
Your Search campaign then looks expensive by comparison, and the gap is partly an artefact of who was allowed to claim the brand click.
What the 2026 channel reporting actually answers
Channel level reporting landed in Google Ads API v23 on 28 January 2026. The ad_network_type segment now returns real channels instead of the old MIXED bucket, so a PMax campaign splits into Google Search, Search Partners, YouTube, Display, Discover, Gmail and Maps. Placement data sits under the “When and where ads showed” tab in the interface. Asset group level breakdowns stayed API only and never made it into the web UI at all.
This is a real improvement and I use it every week. It also settles nothing about credit, and I think the industry has quietly conflated the two. Knowing that 61 percent of a PMax campaign’s spend landed on Google Search tells you where the money went. It says nothing about whether those conversions would have happened anyway. Brand exclusions get recommended as the fix and they’re worth applying, but a brand exclusion is a term list approximating intent, and it will not catch every phrasing somebody uses when they already know your name. Treating a brand exclusion as proof of incrementality is the same error as treating a click as a lead.
Small tangent that cost me an afternoon: those channel reports only return data for dates on or after 1 June 2025. Pull a two year comparison and the earlier window comes back empty with nothing in the response explaining why.
What the geo holdout tests found
Haus ran three cell geo experiments splitting markets into a control with no PMax, a cell running PMax including brand terms, and a cell excluding them. Excluding brand terms drove more incremental revenue in half the experiments, and when it won it produced 24 percent more incremental revenue on average. On new customer acquisition it won every time, with acquisition cost falling between 19 and 60 percent, averaging a 40 percent reduction. One published example, the kitchenware brand Caraway, found the platform overstated performance by 33 percent. The full write-up is on the Haus experiment blog.
Read the conditions before you borrow the conclusion. Those are ecommerce advertisers with order values and repeat purchase behaviour, the study splits its results at an average order value of $238, and brands above that line actually did better including brand terms. Lead generation accounts have no AOV to split on. The finding is genuinely contested because the right answer moves with the business model, and I wouldn’t hand these numbers to a plumbing client as a forecast of what their account will do.
The account that made this concrete for me
An Australian roofing account I audited reported 318 conversions at roughly $10 cost per acquisition. It had looked that good for 16 months. Tracing the conversion action back, 96 percent of those 318 were taps on the mobile phone number button, counted the instant the tap fired, with no connection required. Real traceable leads, meaning form fills plus calls recorded through the call extension, came to 12 across the entire period. True cost per acquisition was closer to $263.
That matters here because the account ran PMax alongside Search, and the PMax campaign carried the better reported CPA. It carried the better CPA because it was collecting more of the junk taps. Smart bidding had spent 16 months learning that Display placements producing phone taps were the good stuff, so it went and bought more of them. The full teardown is in The 318-Lead Mirage.
Before you argue about which campaign deserves the credit, confirm the thing being credited is an outcome.
How to check the overlap in your own account
- Open the PMax campaign, go to the “When and where ads showed” tab, and segment the placement report by network. You want the share of spend and conversions sitting on Google Search.
- Pull the PMax search terms report. Google rolled this out from 30 April 2025 at the same granularity you get for Search and standard Shopping. Sort by conversions and read the top 50 terms.
- Mark every term containing your brand, a misspelling of your company name, a product name only your customers would know, or a competitor comparison. Sum the conversions attributed to them.
- Compare that list against the queries your Search campaigns already cover with exact match keywords. Overlap there is volume PMax could only have taken on Ad Rank.
- Check the conversion action itself. Goals, then Conversions, then Summary, select the action, open Settings, and read the Count setting. If a phone tap or a page view is set to count Every conversion instead of One, stop the analysis and fix that first.
Step five moves real money and almost nobody does it. If the underlying conversion action is measuring intent, every campaign comparison built on top of it is arithmetic on a broken number, which is the same pattern behind where wasted spend actually hides in a Google Ads account.
Where I still do not have a clean answer
The honest way to settle this is a geo holdout: pause PMax across a matched set of regions, run it long enough, compare total conversions rather than attributed ones. On accounts spending $20,000 a month I can do that and trust the result. At $4,000 a month across one metro service area the conversion volume is too thin to reach significance before the season shifts underneath the test, and I have not found a rule for that spend level that I’m happy with. Right now I fall back on tracing search terms and reading brand versus non brand splits, which is weaker evidence, and I know it.
What could go wrong with this
The obvious failure mode is over-correcting. Find PMax taking brand traffic, pause PMax outright, and you also lose whatever genuine prospecting it was doing on YouTube, Discover and Display. That loss tends to show up four to eight weeks later as a quiet decline in non brand Search volume, by which point nobody connects it to the change. Taper the budget across three or four weeks instead and watch non brand Search impression volume as your control.
The second failure mode is slower to notice. Search terms reporting in PMax still aggregates low volume queries, so what you’re reading is a sample rather than the full picture, and any brand exclusion list built from it will have gaps. If you automate this check on a schedule, and you should, build the alert to compare shares over several weeks instead of firing on a single week’s number. I described that reporting pipeline in automating Google Ads reporting with n8n and AI.
Pull both reports before your next budget decision. If you want a second pair of eyes on where PMax and Search are overlapping in your account, and on whether the conversion actions underneath them are counting outcomes, book a free 30-minute audit. Senior-led, month to month, no lock-in, no reporting theatre. There’s more on the same failure pattern in why your Google Ads cost per lead keeps climbing.
Researched and drafted with AI assistance, then reviewed, fact-checked and edited by Rajat Verma before publishing. All account examples are from real campaigns managed by Turquoise Digital and are anonymised.