How to audit a Performance Max campaign (the black box)

Performance Max is the campaign type Google most wants you to run and least wants you to see inside. You hand it budget, assets and a goal; it decides where the money goes across Search, Display, YouTube, Gmail, Discover and Maps, and reports back a single ROAS number that, more often than you would like, is flattering. The good news: the black box has seams. This is the audit, ranked by how much money each check tends to recover.

PMax doesn't fail loudly. It fails by spending your budget in the one place you would never have chosen, and taking credit for the sales you would have made anyway.

1. Brand-spend detection: the check that pays for the audit

Start here, always. The most common way Performance Max looks like a hero is that it is serving on your own brand terms (people already searching “[your company]”) and booking those near-certain conversions as its own work. Your reported ROAS goes to the moon; your incremental ROAS barely moves.

The tell is a search term or asset group with a ROAS wildly out of line with everything else: 15× when the account average is 3×. That is almost never a demand-gen miracle. It is branded traffic PMax is intercepting on the way to a click you already owned. Pull the search-terms and asset-group breakdown, find the outlier, and check whether the winning queries contain your brand name.

If they do, the fix is a brand exclusion (an account-level negative keyword list or a brand list applied to PMax) so the campaign competes for demand you don't already have. It is worth its own deep dive, see Performance Max brand cannibalization, but for the audit the rule is simple: any asset group with abnormal ROAS is guilty until the search terms prove otherwise.

2. Conversion-tracking quality: garbage in, garbage out

Performance Max leans on your conversion data more heavily than any other campaign type. It has no manual levers: no keywords, no placements you pick, no bids you set. All it has is the goal you give it and the conversion signal it optimizes toward. If that signal is wrong, every decision downstream is wrong, and no amount of asset tweaking will save it.

So before you touch anything else, verify the tracking is clean:

• One primary conversion, counted once. Double-firing tags and duplicate conversion actions teach PMax that a purchase is worth two, and it overspends accordingly.
• Values that reflect real revenue. If you pass a static value instead of dynamic cart value, PMax optimizes for volume of cheap orders, not margin. Check the values against GA4 and your store's own numbers.
• No soft goals as primary. “Add to cart” or “page view” as a primary conversion turns PMax into a traffic firehose chasing the easy signal, not the sale.
• Consent mode and modelling sanity. If a big share of conversions are modelled rather than observed, the ground under every decision is softer than the ROAS number admits.

Garbage in, garbage out isn't a cliché here; it is the whole mechanism. PMax is a signal-following machine, so the signal is the product.

3. Channel and network breakdown: where it's actually spending

Now find out where the money physically went. The channel-level split (Search vs Display vs Video vs Shopping) is in the Channel performance report under Insights and reports, for data from June 2025 onward; a script or the Google Ads API gives the same split in bulk or for older periods, and the placement report shows the actual apps and sites.

What you are looking for is the classic PMax leak: budget draining into cheap Display impressions and auto-playing Video that generate impressive reach and almost no revenue, while the Search and Shopping slices quietly do the real work. If 60% of spend is Display and 5% of conversions are, you have found a large recoverable chunk.

Two exclusions do most of the cleanup:

• Junk placements. Pull the placement report, find the mobile games and made-for-advertising sites eating budget, and add them to an account-level placement exclusion list. Google confirms Performance Max respects account-level placement exclusions.
• Mobile-app inventory. A mobile-app category exclusion removes a large share of the accidental-tap traffic that flatters your click numbers and converts at nothing.

4. Asset-group strength and audience signals: don't cold-start on your money

Every asset group is a mini-campaign, and PMax launches each one into a cold start while it learns. What shortens the cold start is audience signals: your customer lists, high-intent segments and custom audiences telling Google where to begin looking. Leave them empty and you are paying full price for the machine to guess from scratch.

In the audit, check that every asset group has a genuine audience signal (not blank, not “all users”), the full complement of assets so Google isn't starved for creative to test, and enough conversion volume to have actually exited learning. A thin asset group with no signal is a cold start you are funding indefinitely.

5. Landing-page experience per asset group

PMax uses your final URLs, and unless you turn it off, final URL expansion sends traffic to whatever page on your site Google thinks fits the query. That means an asset group themed around one product can be sending paid clicks to a stale blog post or an out-of-stock page.

Audit the landing experience per asset group: are the URLs the ones you intended, is final URL expansion doing something sensible or something feral, and does each page match the promise of the asset group's creative? A great asset group pointed at a slow, off-message page is spend you are wasting at the last step.

6. Search vs PMax overlap and cannibalization

Finally, zoom out to how PMax coexists with your standard Search campaigns. Google's serving priority means PMax can win queries your Search campaigns would have won anyway, including brand, and the two can end up bidding against your own account. The result is the same conversions attributed twice and a blended picture that hides which campaign actually earned them.

Check the overlap: are PMax and Search competing for the same high-intent queries, is PMax siphoning credit from a well-built Search campaign, and would tighter brand exclusions plus clearer campaign roles stop the two from cannibalizing each other? This closes the loop back to check one: brand spend and Search overlap are the same leak seen from two angles.

Manual audit vs continuous monitoring

Here is the uncomfortable truth about everything above: a manual PMax audit is a snapshot, and PMax is a moving target. Brand-spend creep, a tracking break, placement drift: none of them wait for your monthly review. A once-a-month audit catches the leak weeks after it started spending. Side by side:

Where an AI media buyer comes in

Every check in this playbook rewards running continuously and suffers when run monthly. That is the case for handing it to an AI media buyer: one that pulls the channel and placement breakdown, catches the brand term with abnormal ROAS, raises the tracking break at the next daily read instead of in next month's report, and ranks every finding by recoverable spend, so you fix the biggest leak first instead of the most visible one.

PMax isn't the only black box worth this treatment. The same discipline applies to the whole account, which is why it pairs with a broader Google Ads account audit.

Adgent is a team of AI agents working toward your marketing goals on Google Ads and Meta. They read every connected account each day, put an estimated monthly amount on every decision and rank the list by what it costs. Each leak arrives with the evidence, next move and later result attached.

Frequently asked