Google Already Opened the Performance Max Black Box. Here Is the Audit Nobody Runs.

Most of the advertisers who tell me Performance Max is a black box have never opened the page where Google breaks it open. Campaigns, then Insights and reports, then Channel performance. It has been sitting in every PMax campaign since November 6, 2025, when Google confirmed that channel performance reporting was "now available in all Performance Max campaigns." Google publishes the channel mix. What still goes unexamined in August 2026 is the number those advertisers actually budget against, the conversion count on the campaign card.

That count includes conversions credited off an impression nobody clicked. It also blends brand searches from people who already knew you, remarketing to customers you have, and Shopping and text ads filed under one Search row. That is four things inside one number. The report has segments for most of them, and they sit off by default.

The automation is good at the job it was given. It moves budget across Search, Display, YouTube, Discover, Gmail, Maps and search partners faster than any human team can, and when a channel goes quiet the report tells you why: "this channel isn't showing ads because the budget is being prioritized for conversions on more efficient channels." The catch is the first thing your eye lands on at the top of that page. Mike Ryan of Smarter Ecommerce measured the Channels-to-Goals chart that sits above the table: a band for asset based Search ads was drawn at what looked like a couple of hundred thousand impressions and carried 4,500. The proportions are not drawn to volume, so scroll past it and read the table underneath.

What does the channel performance report actually show you?

It is the Performance Max report that splits one campaign's results by the Google surface that served the ad, and you find it under Campaigns, then Insights and reports, then Channel performance. Google's documentation lists the channel dimension as Google Search, Google Display Network, YouTube, Discover, Maps, Gmail and Search partners, with impressions, clicks, interactions, conversions, conversion value and cost against each. The date picker in the channel performance report goes back as far as June 6, 2025.

Two of the six segments carry nearly all of the value. "Ads using product data" separates feed driven delivery from creative driven delivery, so Shopping stops hiding inside the Search row. "Ad event type" splits conversions by the ad event that preceded them, and the values Google names in its documentation are Engaged-View, Clicks, Interaction and Impression. Read that list again. Impression is a conversion category, so a view through conversion, one Google credits to an ad that was shown but never clicked, sits in the same total as a conversion somebody clicked.

Mike Ryan, writing in Search Engine Land on December 17, 2025, found that segment under Segment then Conversions, where it "lets you easily understand the volume and performance of your click-through versus view-through conversions." Ryan also flagged that Conv. value / Cost and Cost / Conv. sit in the Columns menu switched off by default. Two clicks, and the report ships with both of them unmade.

Is Performance Max eating your brand search?

Yes, on queries your Search campaigns are already buying, and how much of your conversion count that adds up to is a number nobody has published. I went looking for a credible study on what share of PMax conversions are brand queries and came up with nothing, from Google or anyone else, and the share figures circulating in vendor blog posts do not trace back to a source. So there is no outside benchmark to measure yourself against. Whatever you learn here, you learn from your own account or not at all.

What has been measured is the overlap. Optmyzr published a study on July 31, 2025 covering February 1 to February 28, 2025 across 503 accounts managed in their platform, and 91.45% of those accounts had search terms triggering PMax ads whose text exactly matched a keyword in a Search campaign. The overlap held across 56.29% of 5,768 Search campaigns and 27.86% of 40,642 ad groups, at every match type, exact included. Google's own rule is narrower than that sounds. When the query is identical to an exact match keyword in your Search campaign, the Search campaign is prioritized over Performance Max.

The part the cannibalization crowd skips is what happened on those same overlapping terms: 74.92% showed no meaningful difference in conversion rate between the two campaign types, and Optmyzr defined meaningful as a gap over 10%. So overlap is nearly universal and the penalty usually is not. And Optmyzr left CPC, CPA and ROAS out of the study entirely, because Google was not reporting cost at the PMax search term level at the time of the analysis.

Their read on cause matters more than the headline does. Overlap "often happens because your Search campaign becomes ineligible to serve ads in the first place," through budget caps, geography, ad schedules or audience exclusions. If your brand campaign ran out of money at eleven in the morning, PMax taking the query at two is not theft.

Why do brand exclusions not solve this?

Because they only cover part of the inventory, and the part they do cover leaks. Brand exclusions are the Performance Max setting that stops a campaign serving against searches containing brand terms you list, and Google's help page is specific about the limit: for Performance Max they "apply to only Search, Shopping, and YouTube search inventory," with a checkbox that lets Shopping ads keep serving on searches mentioning excluded brands. Display, Discover, Gmail and Maps sit outside the exclusion entirely.

Then there is what gets through the covered part. Mike Ryan, in the SMX Next session written up by Search Engine Land on February 6, 2026, said brand exclusions "can be leaky, with some brand traffic still slipping through," and that negative keywords are more reliable when you need strict control. Optmyzr got to the same place from their own data and recommends branded misspellings as exact match keywords in Search, because "even with brand exclusions enabled, PMax can still trigger ads for fuzzy matches."

Ryan's other finding from that session is the one that actually matters. Performance Max does not separate brand from generic traffic, and it favors brand queries because they are high intent and they convert. You told a bidding system to find cheap conversions and it found the cheapest ones in the building.

How to audit a Performance Max campaign, step by step

This is the sequence we run, and it needs nothing beyond the Google Ads interface and a spreadsheet. Give a first pass on one account half a day, most of which goes on step four: an hour covers most of the spend on a mid sized account, and a large one will eat an afternoon. Steps one and five are the quick ones, at about fifteen minutes and twenty minutes. Every pass after that runs closer to ninety minutes, once your buckets exist.

1. Pull the channel table with two segments on, not zero

Pull the channel table with two segments on by going to Campaigns, then Insights and reports, then Channel performance. Set your window, then add "Ads using product data" and "Ad event type" as the segments and turn on Conv. value / Cost and Cost / Conv. from the Columns menu, which are the only two ratios the report will build for you. Export to Sheets for everything else, because impression credited conversions as a share of total and brand spend as a share of channel spend are the two numbers this whole exercise turns on and Google calculates neither one.

2. Ignore the Sankey diagram

The Channels-to-Goals chart above the table is the Sankey diagram to skip, for the reason set out at the top of this article: its proportions do not track volume. The table is the report and the chart is decoration.

3. Split conversions by ad event type and write down the impression share

Splitting conversions by ad event type separates the conversions somebody clicked from the ones Google credited off an impression alone, and the Impression share is the number to write down. Every conversion sitting under Impression is a view through conversion credited to an ad nobody clicked.

Write that share down as a percentage of total conversions. It is the pull that changes the finance conversation, and no published benchmark exists for a normal share, so treat your first number as month one of your own baseline rather than as a finding. With nothing outside your account to compare it against, the second month is the one that tells you anything.

4. Bucket your search terms into brand, close to brand, non-brand, and the unlabeled remainder

Bucketing your search terms means sorting the campaign level search term view by cost and tagging the top rows by hand as brand, close to brand, non-brand or an unlabeled remainder. That view now carries cost, which the older search term insights report never did. It arrives at search network level without separating search from shopping, so a single term can blend both formats.

Close to brand is where the judgment lives: competitor names, your own product names used generically, misspellings only your customers make. Anything you cannot place goes in the unlabeled remainder, because a bucket you forced is worse than one you left empty. Do this before you touch brand exclusions.

5. Check whether your Search campaigns were even eligible

Eligibility comes down to a short list of settings on the Search campaign: budget caps, ad schedules, geographic targeting and audience exclusions. Optmyzr named these as the usual cause of overlap and did not test them. You can, and it changes what the overlap means.

6. Run the placement report segmented by network

Run the placement report from the "When and where ads showed" tab, where Google said in January 2026 it would sit, and segment it by network. It is impressions only and it is thin, and still enough to turn up placements you would not have chosen.

Scripts that give you a trend line instead of a snapshot

smec publishes a free PMax Brand Traffic Analyzer script that charts brand versus non-brand share by date group, and Mike Rhodes sells one that sorts search categories into those same buckets alongside channel spend. Both beat the native reporting at that job. Neither of them knows which queries matter to your business, and that judgment stays with you. We built query level paid search optimization around it because it kept turning out to be the step that decided whether the rest of the audit was worth anything.

What is still genuinely closed, and it is a shorter list than the black box story needs

Channel budget control does not exist, and Google says so in its own documentation: "you can't directly control budget allocation per channel." Assets, audience signals, search themes and exclusions all influence it. None of them set it. Nothing traces one asset through one placement to one conversion either, which is the separate gap we covered in what PMax asset automation actually reports.

And then there is "interactions" itself, which means different things in different rows. smec documented this in May 2025: clicks on Search and Shopping, views or other video engagements on YouTube, a teaser expansion or a swipe or a hover on Display and Gmail. Compare interaction counts across channels and you are comparing three different definitions. smec notes you can back out engagements by subtracting clicks from interactions. Shopping has no row of its own either, so feed based equals Shopping is a decoding rule you need before the table means anything.

Ryan reported in December 2025 that Google had confirmed API access and manager account reporting were coming for this report, and neither had shipped as far as I can confirm. Report Editor did not support Performance Max as of February 2026. Switching off search partners and Display was still a limited alpha as of PPC Land's report on July 28, 2026, with two checkboxes enabled by default.

But the direction of travel is real. Search partners became a segment, cost arrived in the search term view, placement exclusions moved to account level in January 2026, and Google said in January 2026 that first party audience exclusions would keep existing customers out of a prospecting budget. That is a lot of ground conceded between April 2025 and July 2026 by a product that launched with none of it.

What none of Google's reporting will tell you is which conversions you would have got anyway. Your own query export and your CRM can, and the reason so few advertisers know that number is that the export goes unrun rather than unavailable. We wrote about the wider habit in our read on over reliance on Google Ads automation, where the platform grades itself on activity. Grading it on outcomes is a morning of work you have not scheduled yet.

Frequently asked questions

How do you audit a Performance Max campaign?

Auditing a Performance Max campaign takes six steps and about ninety minutes once your search term buckets exist. Open the channel performance report under Campaigns, then the Insights and reports drop down, then Channel performance, pull the table with the "Ads using product data" and "Ad event type" segments applied, and export it to a spreadsheet. Ignore the Sankey diagram above the table and read the table itself. Split conversions by ad event type and record what share sits under Impression, because those rows were credited to an ad that was shown and never clicked.

Bucket your top spending search terms into brand, close to brand, non-brand and the unlabeled remainder. Check whether your Search campaigns were eligible to serve on the overlapping terms in the first place. Then run the placement report segmented by network. Do all six before you change a brand exclusion setting, and give a first pass half a day rather than ninety minutes. No published benchmark exists for the Impression share, so treat the first pass as the start of your own baseline.

Do brand exclusions stop PMax bidding on my brand?

Brand exclusions stop part of a PMax campaign's brand bidding and not all of it, which is the honest answer. Google limits brand exclusions in Performance Max to Search, Shopping and YouTube search inventory, and Shopping can be allowed through with a checkbox, which leaves Display, Discover, Gmail and Maps outside the exclusion entirely. Mike Ryan reports leakage on the covered inventory too, and Optmyzr recommends branded misspellings as exact match keywords in Search as a backstop.

How much of a PMax campaign's conversions are usually brand?

Nobody has published a defensible figure for the brand share of PMax conversions, and I would treat any number quoted without a named study and a sample size as invented. The brand share is measurable per account, so the answer comes from your own search term export.

Should I split brand into its own campaign?

Split brand into its own campaign only after you have measured your brand share, and only if you know the tradeoff. Removing brand takes your highest converting segment out of the signal PMax learns from, and Mike Ryan of Smarter Ecommerce holds that Performance Max performs best with more data while campaigns with low monthly conversion volume tend to miss targets.