Most PMax accounts structure asset groups around product lines. Running shoes here, accessories there, seasonal bundles somewhere else. The categories make sense to a human manager. They make no useful signal to the machine.
PMax asset groups explained correctly are not product folders. They are signal containers - and the intent you put into each container determines what the algorithm learns, where it spends, and which buyers it finds.
The finding: Asset groups that bundle multiple buyer intents create signal noise PMax cannot resolve. The algorithm defaults to cheapest available inventory rather than the highest-conversion intent path - and the operator concludes PMax does not work.
What we actually looked at
These patterns come from auditing and managing Google Ads accounts across DTC/ecommerce and SaaS/B2B verticals. The structural mistakes repeat regardless of account size, spend level, or product category. The common thread: operators import how they think about product catalogues into how they build campaign structure. PMax does not respond to product catalogues. It responds to intent signals.
Intent-based asset groups outperform product-category asset groups
When PMax wins an auction and drives a conversion, the algorithm builds a model of what that converting visitor looked like - their query type, behavior pattern, device, time, and audience overlap. This model shapes every subsequent bid decision.
When one asset group contains running shoe buyers and yoga mat buyers, a conversion event is ambiguous. The algorithm cannot determine which signal pattern predicted the outcome. So it stops trying to refine the signal and defaults to the inventory surface with the most volume and the lowest cost-per-click.
The result is an account that looks like it is running PMax but is actually running Display-heavy campaigns that happen to be administered through PMax infrastructure.
| Asset Group Structure | What PMax Learns | Where Spend Concentrates |
|---|---|---|
| By product category (mixed intent) | Ambiguous - signal cannot resolve | Low-cost Display and Gmail |
| By buyer intent (one intent per group) | Clear - conversion maps to one signal type | High-intent Shopping and Search |
| All products in one group | Noise - too many competing signals | Cheapest available inventory |
This is the Intent-First Asset Group structure: one asset group equals one buyer intent, one value proposition, one creative theme across all asset types. A DTC/Apparel brand running this correctly might have three groups - gift-occasion buyers, seasonal-need buyers, and general discovery buyers - each with headlines, descriptions, images, and video that speak to one intent only.
Leaving the video slot empty removes YouTube from the auction
PMax spans six Google surfaces simultaneously: Search, Shopping, Display, YouTube, Discover, and Gmail. Asset groups provide the creative and signal input for all six at once.
Video is the only format YouTube requires. When the video slot is empty, YouTube is removed from the campaign's available inventory. The campaign runs across five surfaces, not six - and the performance read never reflects what YouTube demand looked like.
You are not testing PMax. You are testing PMax minus its highest-engagement surface.
This distorts performance conclusions. A brand without video assets measures Shopping-and-Display PMax performance and concludes the campaign type does not work for awareness demand. The ceiling was never reached because one slot was never filled.
The minimum to fill the slot: one video between 15 and 30 seconds in 16:9 format. A low-production creator clip outperforms a missing slot - because a missing slot is a permanent zero on YouTube inventory, and a rough clip at least participates in the auction. The Performance Max setup guide covers asset slot requirements in full before any bid strategy is selected.
Audience signals are seeds, not targeting restrictions
The most misunderstood element in pmax asset groups explained is what audience signals actually do inside each group.
Most operators add their customer email list to audience signals expecting PMax to limit delivery to those people. It does not. Performance max audience signals are starting-point seeds - the algorithm uses them to identify similar behavior patterns and find new buyers who share those characteristics.
When PMax delivers to audiences the operator did not explicitly seed, it is working correctly. It is expanding from the signal, not ignoring it.
For performance max audience signals to work, add two inputs at asset group creation: your highest-converting customer email list, and the landing page URLs of pages that already convert. These provide the clearest behavioral signal of what a buyer looks like before the campaign has generated its own conversion data. Adding them three weeks after launch means the algorithm learned its initial model without them.
Shopping feed quality is the floor PMax works within
For accounts running Shopping surfaces, the product feed determines the ceiling of what PMax can serve before a single bid decision is made. The algorithm draws feed attributes - product titles, descriptions, pricing, images - to generate Shopping units dynamically across placements.
A feed with missing attributes, generic titles, or low-resolution images constrains output before optimization begins. Google shopping feed optimization is not a feed management task separated from campaign performance. It is a prerequisite for PMax accessing Shopping inventory at full quality.
Treating google shopping feed optimization as a monthly hygiene task rather than a campaign prerequisite is where most Shopping-heavy PMax accounts lose ground before bidding begins.
The four feed attributes with the highest impact on PMax Shopping performance:
- Product title quality: include brand name, product type, and key differentiator in the first 70 characters - that is what appears in the Shopping unit.
- Image resolution: minimum 800 x 800 pixels. The algorithm cannot upscale a low-quality thumbnail into a quality placement.
- GTIN and product identifiers: Google uses these to match listings to known product demand signals at the auction level.
- Price accuracy: a mismatch between feed price and landing page price triggers a disapproval that removes the product from inventory entirely.
Where this falls apart
Intent-First Asset Group structure requires clean conversion tracking upstream. If the primary conversion action fires on page views instead of purchases, or double-fires on order confirmation page reloads, the algorithm optimizes toward the pattern that generates the most fires - not the pattern that drives revenue.
Before diagnosing asset group structure as the problem, confirm that conversion events track real purchase outcomes accurately. Server-side conversion setup eliminates the browser-side reliability issues that make PMax learning unreliable before structure is even examined.
The second caveat: this structure needs time and budget. Each asset group needs approximately 50 conversions before bid decisions stabilize. An account at $50 per day with a $40 target CPA cannot build 50 conversions in one asset group in under 40 days.
Spreading that budget across five asset groups at launch means none of them learn. Launch with one group. Add a second only after the first has 50 conversions and a clear performance trend.
What to do with this on Monday
Pull the asset groups in your current PMax campaign. For each one, answer one question: do all the headlines, descriptions, images, and audience signals describe one buying reason - or multiple?
If the answer is multiple, the fix is structural before it is anything else. A bid strategy change on a mixed-intent asset group produces a mixed-intent result. BAVai surfaces asset group creative performance across the Google Ads accounts we manage every morning - and the flag it raises most often is creative themes that span multiple buyer intents competing for the same conversion signal.
The Performance Max vs Search Campaigns post covers how asset groups interact with the broader Google campaign structure once they are built correctly. The sequence is: intent-first asset groups first, then the Search-PMax stack on top.
Structure before optimization. Every time.
If you rebuilt every asset group in your PMax account around one buyer intent instead of one product category, how different would the algorithm's signal quality be - and where would the spend go?