PMax Demystified: Asset Groups, Bidding, and What You Can Control

Most advertisers treat Performance Max like the weather — something that happens to them. They upload assets, set a target ROAS, and watch Google spend their budget in ways they can't fully explain. The frustration is real, documented loudly in practitioner communities, and mostly avoidable. PMax does give you control. The controls are just nothing like the knobs you had in standard Shopping or Search.

Here's what actually matters: asset groups are your audience-and-creative hypothesis, not just an organizational folder. Bidding signals are your loudest inputs into the auction. And the things that look like controls — campaign-level audience segments, custom intent lists — behave differently than they do anywhere else in Google Ads. Once you internalize that, PMax stops feeling broken and starts feeling expensive-but-legible.

What an Asset Group Actually Is

Google's documentation calls an asset group "a collection of assets." That's technically accurate and almost completely useless as a mental model.

A more accurate frame: an asset group is a hypothesis you're submitting to the auction. You're saying, "Here is a set of creative signals and audience signals. Figure out which inventory — Search, Shopping, Display, YouTube, Discover, Maps — converts best using this combination." Google then tests that hypothesis across every surface it owns.

This reframe has a practical implication. Asset groups split by product category or audience intent give the model genuinely different hypotheses to evaluate — and different creative and signal sets to learn from. Splitting "running shoes" from "trail shoes" is a real signal difference. Splitting "blue banner" from "red banner" inside the same product category gives the model nothing it can act on at the auction level. Run creative color tests inside Display campaigns where you actually control placement.

How Audience Signals Actually Work

This is where PMax diverges most sharply from everything else in Google Ads, and where the most confusion lives.

In standard campaigns, audiences are targeting or observation layers. In PMax, they are signals — starting points the model uses to seed its audience expansion. Google is explicit about this in its help documentation, but the implication doesn't land until you've watched a campaign drift far outside the segments you added.

Custom intent segments in PMax follow the same logic. When practitioners ask whether PMax supports custom intent or interest segments, the answer is yes — you can add them as audience signals at the asset group level. But you are telling the model where to look first, not where to look exclusively. If Google's auction data suggests there are converters outside your custom segment, it will find them.

This is not necessarily bad. If your conversion tracking is accurate, the model's expansion tends to be directionally correct over time. If your conversion tracking is broken or shallow — optimizing for micro-conversions that don't correlate with revenue — expansion becomes expensive and random.

The Audience Signal Inputs Worth Prioritizing

In rough order of influence on the model:

1. Customer lists (first-party, uploaded or via Customer Match) — highest signal fidelity
2. Website visitors with value differentiation — especially if you pass revenue values
3. Custom intent segments built from your own search terms — more targeted than interest categories
4. Google's in-market and affinity segments — useful as supplementary signal, not primary

The common mistake is adding a broad interest category and treating it as an audience filter. It isn't. Add it, but weight your customer list and custom intent segments more heavily by ensuring they're present and fresh.

The New Customer Bidding Feature — What It Does and Doesn't Do

The "bid higher for new customers" option in PMax has generated real confusion, and the practitioner frustration is understandable. Let's be exact about the mechanics.

When you enable new customer acquisition mode with a value modifier, you are telling Google's bidding model to treat a conversion from a new customer as worth more than a conversion from an existing one — by whatever percentage you set. The model then factors that into auction decisions. It does not suppress serving to existing customers. It does not guarantee new customer conversions. It shifts the value equation at the margin.

The practical effect depends heavily on:

• How accurately you define "new customer" — Google can use your customer list as the baseline, but only if you upload a reasonably fresh, complete list. An old list means the model mis-classifies returning customers as new.
• What modifier value you set — a 10% uplift has nearly no practical effect if your target ROAS is already tight. A 50% uplift changes auction math meaningfully.
• Whether you're using "new customer only" mode — this exists as a harder constraint, but it limits volume significantly and works best when new customer acquisition is the only goal, such as a conquest campaign or a promotional launch.

The Control Surface: What Each Lever Actually Does

The frustration with PMax is often vague. This table makes it concrete. Each lever maps to what it genuinely constrains, where it applies, and what it cannot do.

Lever	What it constrains	Where it applies	What it cannot do
Asset groups	Creative and audience signal inputs	All surfaces	Control which surface gets budget
Audience signals	Starting point for model's audience expansion	All surfaces	Restrict delivery to that audience only
URL exclusions	Specific pages excluded from landing destinations	All surfaces	Exclude query types or placements
Brand exclusions	Prevents brand-term triggering	Search and Shopping only	Apply to Display, YouTube, or Discover
Campaign-level negatives	Blocks specific queries	Search and Shopping only	Fully visible in Search Terms Insight
New customer modifier	Increases bid value weight for new converters	All surfaces	Suppress delivery to existing customers
Conversion value rules	Adjusts downstream value signals by customer type or geography	Bid model inputs	Override the auction in real time
URL expansion opt-out	Forces traffic to your specified URLs only	All surfaces	Restore query-level transparency

Sources: Google Ads Help — Performance Max overview, brand exclusions documentation, new customer acquisition settings.

The asymmetry in this table is real. You have fewer levers than in legacy campaign types, and most of them operate on signals rather than hard rules. But the levers you do have — especially conversion value rules and customer lists — are more powerful than they look when used with clean data.

What You Can Actually Control

Let's be honest about the full list.

You control:

• Campaign-level budget and bid strategy (tCPA or tROAS, or Maximize Conversions/Value)
• Asset group structure and the creative you provide
• Audience signals at the asset group level
• Brand exclusions (Search and Shopping inventory — not Display or YouTube)
• URL exclusions (specific pages you don't want to drive traffic to)
• Conversion actions selected for optimization
• Conversion values and value rules
• Negative keyword lists at the account level (and at the campaign level for accounts where this has rolled out)
• Location and ad scheduling at the campaign level
• New customer acquisition settings

You do not control:

• Which specific search queries trigger your ads (Search Terms Insight shows an aggregated partial view, not the full log)
• Which placements receive budget within the campaign
• Creative assembly at the individual ad level
• Audience expansion radius

Every hour spent cleaning up conversion tracking, importing offline data, and setting accurate values returns more than an hour spent trying to rebuild placement-level control inside an automated system. That's not a consolation prize — it's where the actual leverage is.

Bidding Strategy: Matching the Lever to the Stage

The choice between tROAS, tCPA, and Maximize Conversions/Value is stage-dependent, not preference-dependent.

New campaign with thin conversion history: Use Maximize Conversions or Maximize Conversion Value with no target. Let the model collect data before you constrain it. Setting a tROAS too early causes the campaign to under-deliver while it tries to hit a target it has no data to achieve.

Established campaign with sufficient conversion history: Introduce a tROAS or tCPA, but start loose — set your initial target at roughly your current observed efficiency, not your goal efficiency. Tighten gradually, and give the campaign time to stabilize between adjustments before evaluating whether the change held.

Scaling phase: Raise budgets before raising targets. PMax's auction behavior is volume-sensitive; starving it of budget while demanding efficiency produces neither.

The Comparison to Meta Advantage+ (And Why It's Instructive)

The practitioner frustration with PMax closely mirrors what's happening with Meta's Advantage+ campaigns — a pattern worth naming explicitly. Both products automate placement, audience expansion, and creative selection in ways that reduce direct control. Both claim superior performance when given clean signals. Both generate complaints when advertisers migrate existing campaigns into them without adjusting their measurement and input strategies first.

The difference: PMax has more documented control surfaces, more integration with first-party data via Customer Match, and a clearer conversion signal infrastructure via enhanced conversions. The underlying dynamic is identical — automation products reward advertisers who invest in signal quality, not advertisers who try to rebuild manual control inside an automated system.

If your instinct when PMax underperforms is to add more audience segments or split into more asset groups, stop. Check your conversion data first. That's almost always where the real problem is.

The Asset Group Structure That Actually Works

The campaigns in our labeled corpus that perform most consistently share a specific structural pattern. They use 3 to 5 asset groups, each representing a meaningfully different product-audience hypothesis. Each asset group has complete creative coverage — all headline slots filled, all description slots filled, all image and video formats provided. The asset groups share a campaign but have distinct audience signals and, often, distinct landing pages.

They do not use asset groups as a creative A/B testing vehicle. That's not what PMax optimizes for. They do not add every available audience signal to every asset group. Specificity is the point.

The single most broken pattern we see: one asset group with incomplete creative and a customer list thrown in as an afterthought, then the campaign blamed for poor performance. The model's inputs were poor. The outputs were consistent with that.

---

FAQ

What is the difference between audience signals and audience targeting in PMax? In PMax, audience signals are inputs that tell Google's model where to start its search for converters — they are not hard targeting constraints. The model will expand beyond your signals if auction data supports it. In standard Search or Display campaigns, audiences can function as hard inclusion or exclusion layers. This distinction matters because adding a narrow audience signal to PMax does not limit delivery to that audience.

Can you use custom intent segments in Performance Max campaigns? Yes. You can add custom intent segments as audience signals at the asset group level. They function as directional inputs — the model uses them to seed its initial targeting — but it is not limited to them. Build custom intent segments from your own high-converting search terms for the best signal quality.

Does "bid higher for new customers" in PMax stop ads from showing to existing customers? No. The new customer value modifier makes new customer conversions worth more in the bid model's math, which pushes it to prioritize those conversions — but it does not suppress delivery to existing customers. If your goal is exclusively new customer acquisition, use the "new customers only" mode and accept the associated volume limitation.

Can two asset groups point to the same landing page? Yes, this is a supported and useful pattern. Different asset groups can share a destination URL. The value is in testing different creative themes and audience signals against the same page — you learn which approach drives conversions, not which page does.

How many conversions does PMax need before smart bidding works reliably? The general practitioner guidance is to avoid setting a tROAS or tCPA target until the campaign has meaningful conversion history at the campaign level, not just the account level.

Why can't I see all my search terms in PMax? PMax shows search term data in the Search Terms Insight report, but this is an aggregated view — it does not show every query. Google does not expose the full query log, citing privacy thresholds. Campaign-level negative keywords (where available) and brand exclusions are the main tools for shaping query eligibility.

Is PMax worth running if I already have strong Shopping and Search campaigns? It depends on your conversion volume and your tolerance for opacity. PMax can access inventory — Discover, YouTube, Gmail — that Shopping and Search cannot. But it can also cannibalize existing campaigns' traffic without clear attribution credit. The cleaner approach is to run PMax alongside existing campaigns with clear budget separation, monitor impression share and conversion path data, and make the call on whether incremental volume justifies the loss of control. There is no universal right answer.

---

The one thing worth doing this week: audit your conversion setup before changing anything else in your PMax campaigns. If your conversion data is stale, missing offline events, or optimizing for a proxy metric that doesn't correlate with revenue — fix that first. Every other PMax decision is downstream of signal quality, and no amount of asset group reorganization compensates for a broken foundation.