Performance Max is the most misunderstood campaign type in Google Ads. Not because it's complicated in theory — Google's documentation is clear enough. It's misunderstood because the way it behaves in practice diverges sharply from how most operators set it up. This guide breaks down the actual mechanics: how the auction works, what you control, what you don't, and how to stop PMax from quietly undermining the rest of your account.
1. Asset Groups: The Architecture Most People Get Wrong
An asset group is not an ad group. That distinction matters enormously.
In a standard campaign, ad groups define targeting. In PMax, asset groups define creative inputs — the headlines, descriptions, images, videos, and product feeds you hand Google to mix and match. The targeting logic lives elsewhere (more on that below). Conflating the two leads to the single most common PMax setup mistake: creating asset groups by audience intent when they should be organized by creative cohesion.
What an asset group actually controls
Google pulls from your asset group's contents to assemble ads across Search, Shopping, Display, YouTube, Gmail, Maps, and Discover. For a given asset group, every asset can potentially appear alongside every other asset in that group. That means mismatched creative — a luxury lifestyle headline paired with a discount-first description — is on you.
Correct organization principle: Group assets by message, not by product category or audience type.
Example: If you sell running shoes, don't create an asset group for "women's shoes" and another for "men's shoes" unless your headlines, visuals, and value propositions are genuinely distinct between those groups. If your copy is the same ("Free shipping. 30-day returns."), you're splitting assets for no benefit. Instead, split by offer angle: performance gear vs. lifestyle/casual, or full-price vs. clearance. Each group should be able to stand alone as a coherent campaign.
The video gap problem
PMax will auto-generate video if you don't provide one. Auto-generated video is, without exception, worse than anything you'd produce. More critically, PMax often pushes video-eligible inventory (YouTube, Discover) hard when there's budget to spend — and if it has to use auto-generated creative, you're paying for impressions that don't convert.
If you're not providing video assets, you're not opting out of video placement. You're just letting Google decide what your brand looks like on video.
Minimum viable asset group
- 3–5 headlines (including at least one with the primary keyword theme)
- 2–3 descriptions
- 3+ landscape images (1200×628, product in context)
- 3+ square images (1200×1200)
- 1+ portrait images (960×1200) for mobile-first placements
- 1+ video (15–30s, performance or testimonial framing)
- A product feed linked at the campaign level (for Shopping placements)
2. The Auction and Signal Logic
PMax doesn't bid on keywords. It bids on predicted conversion probability for a given user in a given context. That prediction is the output of a model trained on your historical conversion data, your audience signals, your listed search themes, and real-time behavioral signals from Google's broader dataset.
How the signal stack works
Google's model builds a propensity score for each impression opportunity: "What's the probability this user converts if they see this ad?" That score draws from multiple layers:
-
Your conversion history — the anchor. Without sufficient conversion volume (roughly 50 conversions per 30 days at the campaign level), the model doesn't have enough signal to bid accurately. It will spend, but inefficiently.
-
Your audience signals — inputs you provide at setup. Customer lists, website visitors, in-market segments. These are seeds, not constraints. Google expands beyond them.
-
Your search themes — keyword-level intent signals you provide (covered in detail below).
-
Google's own data — browsing patterns, search history, YouTube watch behavior, app usage, Maps activity. This is the part you don't see and can't influence directly.
-
Real-time context — device, time of day, location, query intent at the moment of the impression.
Why bidding "to a ROAS target" behaves differently than you expect
PMax optimizes toward the Target ROAS or Target CPA you set, but it does so across the entire campaign. It cannot optimize one asset group toward ROAS and another toward CPA. That means if one asset group is pulling high-ROAS Shopping traffic and another is pulling low-intent Display traffic, your ROAS target is blending those two and smoothing out the difference. You're not optimizing — you're averaging.
This is why campaign-level segmentation (separate PMax campaigns for your top SKUs vs. broad catalog, for example) produces better control than trying to manage everything through a single campaign with multiple asset groups.
3. Search Themes vs. Audience Signals — What They Actually Do
These two inputs are commonly confused, even by practitioners who've been running PMax for years.
Audience signals
Audience signals tell PMax who to prioritize. They're provided at the asset group level and accept:
- First-party lists (customer match, site visitors, app users)
- Google audiences (in-market, life events, custom intent)
The operative word is prioritize, not restrict. PMax will serve outside your audience signals if its model predicts a conversion. Your signals accelerate the learning phase by telling Google where to start looking, not where to stop.
Practical implication: Uploading a customer match list of your best buyers is highly valuable early in the campaign's life — it anchors the model toward a useful conversion profile. Once the campaign has 200+ conversions, the list's relative influence diminishes as the model's own data dominates.
Search themes
Search themes are keyword-level intent signals added at the asset group level (up to 25 per group). They tell PMax what topics your ads are relevant to on search inventory. They are not keywords in the traditional sense — they don't trigger match types, they don't have bids, and they don't appear in the search terms report labeled as such.
What they do:
- Increase the probability PMax enters auctions for searches related to those themes
- Influence the creative Google selects from your asset group for search-triggered ads
- Provide a paper trail (partial) when diagnosing what search queries PMax is targeting
What they don't do:
- Guarantee your ad appears for those queries
- Exclude queries outside those themes
- Give you the match-type control of a standard Search campaign
A common misconception: Adding branded terms as search themes in PMax does not prevent PMax from stealing branded traffic — it just makes PMax more likely to compete on branded queries. If you want to control branded vs. non-branded traffic, the solution is campaign priority and exclusions (see Section 4).
4. Why PMax Cannibalizes Your Other Campaigns — and How to Stop It
PMax has the highest campaign priority in Google's serving logic, above Smart Shopping and standard Shopping, and effectively competitive with exact-match Search. When a query could trigger both a PMax ad and a standard campaign ad, PMax wins most of the time.
This matters because your branded campaigns, your brand-protection exact match, and your high-performing Shopping campaigns are all at risk of having their traffic absorbed by PMax.
The cannibalization mechanics
Google does offer one rule: if you have a Search campaign with an exact-match keyword and PMax would also match that query, the Search campaign gets preference — in theory. In practice, this holdout is inconsistent, and for phrase/broad match terms, PMax always wins.
For Shopping, there is no priority holdout. PMax cannibalizes Shopping by default.
Prevention tactics
Brand campaign protection:
-
Add your brand terms as negative keywords at the PMax campaign level. This requires contacting Google support — negative keywords cannot be added directly to PMax campaigns through the UI (as of 2025). Submit a request for a negative keyword list to be applied.
-
Run a dedicated brand Search campaign with exact match only and a high ROAS target. The model will deprioritize PMax in brand auctions if the brand campaign is healthy and converting well — but this is not guaranteed.
-
Use brand exclusion lists if your account has access to the brand safety features in the campaign settings.
Shopping campaign protection:
The cleanest approach is to use listing group subdivisions in PMax to exclude your highest-margin or highest-performing SKUs, then run those SKUs in a standard Shopping campaign with explicit bidding. You control the bidding strategy; PMax doesn't touch those products.
Alternatively, if you can't exclude specific products from PMax: set your Standard Shopping campaign to Campaign Priority: High and your PMax budget allocation such that Standard Shopping has room to win on its best products. This is a rough workaround, not a clean solution.
5. The Budget Allocation Black Box
PMax doesn't show you how it allocates budget across channels. You will not see a line item saying "we spent €340 on Shopping, €120 on YouTube, and €80 on Display." That data is intentionally aggregated.
What you can infer
From the Insights tab:
- Asset group performance labels (Best, Good, Low) indicate relative conversion contribution per group — not absolute spend
- Search categories showing up in Insights reveal the query themes PMax is actually targeting
- Audience insights show which segments converted, which is useful for refining your audience signals
From the Dimensions tab (campaign level):
- Network breakdowns are available: Search, Shopping, Display, YouTube, Discovery/Demand Gen. This gives you spend-by-network if you pull the right columns.
- Segment by Advertising channel to get a directional read on where money is going
From the Search terms report (partial):
- Only shows search-triggered queries above Google's privacy threshold
- Does not show Display, YouTube, or Shopping query paths
- Still valuable: if you see irrelevant queries here, those are candidates for negative keywords or search theme adjustments
Reading the signal
If your CPC is high and conversion rate is low, PMax is likely spending heavily on non-Search inventory (Display, YouTube) where intent is weaker and the funnel is longer. This isn't wrong in all cases — upper-funnel exposure has value — but if your objective is direct ROAS, it's a problem.
Counter-measure: tighten your ROAS target incrementally (5–10% above current) to force the model toward higher-confidence conversions, which skews spend back toward Shopping and Search.
6. Diagnostic Checks: What a Consultant Looks at First
When I audit a PMax campaign, these are the first five places I look.
Check 1: Conversion action configuration
Open the campaign settings → conversions. Is the campaign using the account-level default conversion set, or has someone added specific conversion actions? PMax will optimize toward all checked conversion actions — including micro-conversions like "Add to Cart" if those are enabled account-wide.
If PMax is optimizing toward add-to-cart instead of purchase, you're training the model on the wrong signal. Strip out micro-conversions from the campaign-level conversion set and leave only purchase.
Check 2: Asset group message coherence
Pull each asset group and read every headline + description combination as if you're a stranger seeing it cold. Does every combination make sense? Are there headline/description combinations that would confuse someone, undersell the product, or contradict each other?
Then look at the images: do the lifestyle images match the brand tone of the copy? Is there a visible product in each image (critical for Shopping placements)?
Incoherent asset groups are the rule, not the exception. Most teams upload assets in bulk without stress-testing the combinations.
Check 3: Search terms report for query bleed
Filter the search terms report for low-intent queries: navigational searches for competitor brands, informational queries with no purchase intent ("how to clean running shoes"), and irrelevant category terms. Any query matching those patterns represents wasted spend and a signal contaminating the model.
Add negatives via a campaign-level negative list. If the list is long (50+ terms), that's a sign the campaign has been operating with poor signal for a while and the model needs a learning reset.
Check 4: Network spend distribution
Pull the Dimensions report and segment by network. Calculate what percentage of spend goes to Search + Shopping vs. Display + YouTube. For a direct-response e-commerce brand, if Display + YouTube together exceed 25–30% of total PMax spend, dig into why. Low ROAS target, low conversion volume, or broad audience signals can all push spend toward upper-funnel placements.
This isn't always bad — but it should be a deliberate choice, not a default.
Check 5: Asset performance labels over time
Check which asset groups are labeled "Low" for performance. If the same asset group has been "Low" for 4+ weeks, it's either (a) pulling in bad traffic, (b) serving a weak offer, or (c) competing against a better asset group in the same campaign.
"Low" assets don't just underperform — they drag on the campaign's overall signal quality. Pause them, rebuild the creative, or consolidate into a stronger group.
The Bottom Line on PMax
PMax works well when the model has what it needs: sufficient conversion volume, coherent creative inputs, clean signal, and a ROAS target that reflects reality. It breaks down when operators treat it like a Smart Shopping upgrade, ignore the cannibalization risk, or expect keyword-level control that the campaign type doesn't offer.
The ceiling on PMax performance is mostly determined by what you put in — feed quality, creative quality, first-party data quality, and conversion tracking accuracy. Get those right, and the model has something real to work with.
Run a Full Audit on Your PMax Setup
The five checks above are a starting point. The full diagnostic goes deeper: campaign structure, conversion setup, feed health, audience list quality, search theme coverage, and budget allocation analysis.
We've compiled all of it into a single PMax Audit Checklist — the same framework used to review accounts before any client engagement. It's available as a free download.
[Lead magnet CTA — checklist link to be added once published]
Part of the Track A Series — Google Ads Technical Guides for E-Commerce
Next: Guide #2 — Conversion Tracking Fundamentals