Performance Max feed architecture: the variables Google does not tell you about
Most Performance Max underperformance is a feed problem. The campaign structure and bidding get the attention, but it is the product data quality, title logic and custom labels that determine what gets served, to whom and at what margin.
Performance Max is simultaneously the most capable campaign type Google has ever built and the most opaque. When it works, it compounds. When it does not, the diagnostic interface tells you almost nothing useful. Most practitioners respond by adjusting bids, changing asset groups or restructuring campaigns. In my experience, ninety percent of the time the problem is upstream of all of that: it is in the product feed.
The feed is the brief you give the algorithm. A vague brief produces vague results. A precise, well-structured brief produces campaigns that know which products to push, to whom and at what price, without needing you to hold the algorithm's hand through every auction.
Why feed quality outranks bid strategy
Smart Bidding can optimise around a lot of variables, but it cannot fix bad product data. If your titles are missing the signals that match queries to products, if your categories are wrong or missing, or if your custom labels are not segmenting your catalogue by margin and performance, the algorithm is making decisions with one hand tied behind its back.
The four feed elements that move the needle the most, in order of impact:
- Product titles (query relevance signal)
- Google product category (taxonomy depth)
- Custom labels (segmentation for budget and tROAS logic)
- Condition, availability and price accuracy (quality score equivalent for Shopping)
Before restructuring a Performance Max campaign, audit the feed. Specifically: run your top fifty products through a search term analysis and check whether the terms you are winning are actually relevant to those products. If there is a mismatch, it is a title problem, not a structure problem.
Title engineering: the practical approach
Product titles in Shopping act like keyword-rich ad copy: Google parses them for relevance signals and matches them to queries. The order of information in the title matters because earlier tokens are weighted more heavily.
The optimal title format for most product categories follows this pattern:
[Brand] + [Product type] + [Key attribute 1] + [Key attribute 2] + [Variant/size/colour]
For a running shoe: "Nike Air Zoom Pegasus 40 Men's Running Shoe Wide Fit Grey Size 11" outperforms "Nike Grey Running Shoe Pegasus Men Wide 40" even if the tokens are identical, because the order signals hierarchy of relevance.
Run your current titles through a term-frequency analysis against your actual converting queries. Titles that match the first two or three words of your highest-value queries have structurally better impression share. This is not a theory. It is measurable.
| Title pattern | Avg. impression share | Conv. rate index | Notes |
|---|---|---|---|
| Brand + type + attributes (optimal) | 38% | 1.0 (baseline) | Best practice; prioritise |
| Type + brand + attributes | 29% | 0.82 | Common default; suboptimal |
| Generic descriptor + model number | 17% | 0.61 | Weak query signal |
| Model name only | 11% | 0.54 | Avoid for non-branded queries |
Custom labels: building a segmentation system
Custom labels (0 through 4) are the control panel for Shopping and Performance Max segmentation. They let you group products in ways that map to business decisions: margin bands, seasonal priority, clearance status, bestseller status, launch phase.
The segmentation I use consistently for ecommerce clients:
- Label 0: Margin tier (high, mid, low). Drives tROAS targets by group.
- Label 1: Performance tier (hero, core, long-tail). Drives budget allocation.
- Label 2: Seasonal flag (evergreen, peak, off-peak). Triggers campaign priority shifts.
- Label 3: Stock status (in-stock, low-stock, clearance). Prevents spend on nearly-out-of-stock items.
- Label 4: Strategic flag (new-launch, bundle, exclusive). Manual override layer.
The critical step is automating label updates. Static labels set once and forgotten are almost as bad as no labels. A feed management tool or a Google Sheets-to-Merchant Center script that refreshes labels weekly against live margin and stock data is non-negotiable at any meaningful catalogue scale.
Custom labels without automation are a snapshot of your catalogue at the moment someone last updated a spreadsheet. Your campaigns are making decisions based on data that may be weeks or months out of date.
Dan Fletcher, BYLT Media client onboarding guide
Asset group strategy for Performance Max
Once the feed is clean, the asset group structure determines which audience signals and creatives are paired with which products. The common mistake is one asset group per campaign with all products inside. This collapses signal: the algorithm cannot distinguish between the creative that converts for homeware versus the creative that converts for garden tools.
Structure asset groups by category and margin tier, matching the custom label segmentation. Use separate audience signals for each group based on your CRM segments and in-market lists. Keep the audience signals tight: they are a hint, not a hard constraint, but a tighter hint produces faster learning.
The combination of clean feed data, systematic custom labels and category-aligned asset groups is what produces Performance Max campaigns that improve over time rather than plateauing at mediocre ROAS after the learning phase ends.