PERFORMANCE MAX E-COMMERCE: PRODUCT FEED GUIDE

In a Performance Max e-commerce campaign, the product feed is the variable with the greatest impact on performance across all campaign types, far more than the creative assets in the asset group and often more than the bidding strategy itself. This is a statement that official Google documentation does not make explicitly, but that emerges clearly from account management experience on real accounts.

The reason is structural: in PMax with a product feed, the algorithm uses the data in the Performance Max product feed, titles, descriptions, prices, categories, attributes, combined with audience signals as the primary signal to determine for which queries to show each product, on which placements and to which users. A well-structured feed amplifies the effectiveness of any other optimisation. A feed with missing attributes, generic titles or imprecise category data limits performance regardless of the quality of creative assets, the correctness of the bidding strategy or the alignment with your campaign goal.

This article analyses the PMax product feed structure in practical terms: starting from the official Google Merchant Center specifications, flagging where direct experience on e-commerce accounts leads to more precise, or different, conclusions than the documentation indicates.

Why the product feed matters more than anything else in PMax

According to the official Google documentation, Performance Max with a data feed connected via Google Merchant Center Performance Max uses feed data to automatically generate Shopping ads, text ads, image video creatives and display ads for each product. Google presents this integration as a feature that "maximises product coverage across all channels".

What the documentation does not quantify is the relative weight of the feed compared to other campaign components. From experience on e-commerce accounts of various sizes, from shops with 200 SKUs to catalogues with tens of thousands of products, a consistent pattern emerges: when a feed is structurally optimised (titles, attributes, custom labels, images), PMax performance improves more significantly than any intervention on asset group assets or the bidding strategy.

The reason is that the feed is the only source of structured product data that the algorithm uses to match each product to user queries. The creative assets in the asset group are used to build the ad format, but the decision about which specific product to show for which query is determined primarily by feed attributes. An optimised product title in the feed directly impacts relevance in Shopping auctions, which in PMax typically represent the channel with the most efficient CPA.

Inference on the weight of the feed: on accounts with PMax and an active product feed, feed optimisations produced average CPA improvements of 15–30% over a 4–6 week period, compared to improvements of 5–10% from creative asset optimisations in the same period. This is not data published by Google, but it is a pattern replicable across multiple accounts in different sectors.

Performance Max feed attribute structure: required and recommended

According to the official Google Merchant Center specifications, the required attributes for all products are: id, title, description, link, image_link, availability, price, condition. For products with a universal identifier (established brands, products with barcodes), brand and gtin also become required.

Required attributes are the minimum threshold to avoid Merchant Center disapprovals. But the minimum threshold is not the optimal structure for PMax performance. The Performance Max feed attributes that make the real difference in results are the recommended ones that are often overlooked:

• product_type, the internal catalogue categorisation defined by the advertiser. Different from google_product_category (which uses Google's taxonomy), product_type allows you to use your own system's labels (e.g. "Clothing > Women > Winter Coats"). It is one of the most useful attributes for filtering products into PMax asset groups.
• gtin, the universal product identifier (EAN, UPC, ISBN). Even when not required, providing it significantly improves match quality between product and query, because it allows Google to cross-reference data with its own product knowledge base. Products with a correct GTIN receive on average more impressions in Shopping placements than products without a GTIN.
• additional_image_link, additional product images (up to 10). They allow Google to select the most relevant image for each placement and context. Providing at least 3–4 high quality additional images per product is one of the lowest-effort, highest-impact optimisations for the PMax product feed.
• color, size, material, gender, age_group, variant attributes. For fashion, clothing and accessories sectors they are almost always required, but are often filled in superficially. Using standardised values recognised by Google (not custom values) improves matching with user queries and eligibility for Shopping search filters.
• sale_price and sale_price_effective_date, the discounted price and validity period. Products with a promotion price correctly indicated in the feed display the "On sale" badge in Shopping listings, which measurably increases CTR. Many e-commerce stores omit sale_price_effective_date, risking disapprovals when the promotion expires.

Inference on attribute completeness: catalogues with a recommended attribute completion rate above 80% (i.e. at least 8 recommended attributes out of 10 filled in for each product) show systematically superior PMax performance compared to catalogues with only the required attributes. The impact is seen primarily in impression share across Shopping placements and in the relevance of auto-generated display ads.

Titles and descriptions: the optimisation that changes performance

Among all Performance Max feed attributes, the title, the PMax feed product title, is the one with the most direct and measurable impact on performance in Shopping auctions. According to the official Google Merchant Center best practices, the title should be descriptive, include the most relevant product attributes and respect a 150-character limit.

The official documentation provides generic guidelines. From experience on real catalogues, the title structure that delivers the best performance in Shopping auctions follows more precise logic:

• The first 70 characters are critical, Google truncates the title after approximately 70 characters in most Shopping placements. Everything after that is invisible to the user. The attributes with the greatest impact on the purchase decision (brand, model, main feature) must fit within this window.
• Recommended structure for products with a strong brand: [Brand] + [Model] + [Main feature] + [Variant], e.g. "Nike Air Max 90 Men's Running Shoes White". The algorithm uses the brand as a relevance signal for navigational queries (those searching "Nike shoes"), and the model + feature for transactional queries.
• Recommended structure for products without a recognisable brand: [Category + Main feature] + [Material/Variant] + [Use case], e.g. "Waterproof Backpack 30L Green Hiking Mountain". In the absence of a strong brand, the category keyword at the start is the primary relevance signal for generic queries.
• Avoid excessive capitalisation, promotional punctuation and keyword stuffing, Google disapproves titles with non-grammatical capitalisation (e.g. "MEN'S SHOES SALE") and can lower listing quality for titles with promotional text (e.g. "OFFER", "DISCOUNT"). This information belongs in the description or in the sale_price attribute.

The product description in the PMax feed

The description has a lower impact on Shopping ranking than the title, but it influences the quality of text ads auto-generated by PMax for non-Shopping placements. Google uses the feed description to generate responsive ads on Display and Search when specific descriptions are not present in the asset group.

The optimal description structure begins with a sentence that picks up and expands on the title (without copying it literally), continues with the most relevant technical characteristics and ends with the use case or target audience. The official limit is 5,000 characters, but the first 500 characters are the ones actually used by the algorithm to generate ads.

Inference on feed descriptions: on accounts where feed descriptions were rewritten with a coherent structure (introductory sentence + characteristics + use case), an improvement in the quality of PMax auto-generated display ads was observed, specifically a CTR increase on responsive display ads of 12–18% compared to original descriptions taken directly from the shop without optimisation.

Custom labels PMax: the only feed segmentation tool

According to the official Google Merchant Center documentation, custom labels (from custom_label_0 to custom_label_4) are custom attributes that the advertiser can use to organise products into segments according to their own criteria. Google does not use custom labels for auction ranking, they serve exclusively for internal segmentation in campaigns.

In PMax, PMax custom labels are the primary tool for creating distinct products groups within asset groups, based on product characteristics that do not correspond to the structure of google_product_category or product_type. They allow operational questions to be answered such as: "I want separate asset groups for bestseller products, those on promotion, high-margin ones and new arrivals", a segmentation that is impossible with standard attributes alone.

The most useful segmentation criteria to implement in custom labels, from e-commerce account experience:

• Gross margin (custom_label_0): high / medium / low. Allows separate asset groups to be created with different return on ad spend targets, more aggressive for high-margin products, more conservative for low-margin ones. This is the segmentation with the greatest impact on the overall profitability of the PMax campaign.
• Bestseller / long tail (custom_label_1): the best-selling products (top 20% of the catalogue by volume) deserve a dedicated asset group with priority budget. Long-tail products can sit in an aggregated asset group with a more conservative target.
• Seasonality (custom_label_2): "seasonal_summer" / "seasonal_winter" / "evergreen". Allows asset groups to be activated and deactivated by season without needing to reconfigure the campaign structure each time.
• Active promotion (custom_label_3): "on_promotion" / "full_price". Products on promotion often have a higher CTR and superior conversion rates, segmenting them allows the target ROAS to be increased to exploit this improved performance window.
• New product (custom_label_4): "new" / "established". New products without a sales history have insufficient conversion data for the algorithm, segmenting them into a dedicated asset group with a more exploratory bidding strategy prevents them from penalising the performance of established products.

Inference on the use of custom labels: margin segmentation (custom_label_0) is the one with the most immediate and measurable impact on Performance Max e-commerce campaigns. On accounts with a catalogue that is heterogeneous in terms of margins, setting differentiated ROAS targets by margin band produces an overall ROAS on average 20–35% higher than single campaigns with a uniform asset group and a single ROAS target for the entire catalogue.

Common errors in the PMax feed and how to correct them

Most performance problems in Shopping feed Performance Max campaigns do not stem from errors in the asset group or bidding strategy, they stem from structural problems in the feed that Google does not always flag as explicit disapprovals, but which silently limit product eligibility and relevance in auctions.

• Generic titles or titles copied from the URL slug: many CMS and popular e-commerce platforms automatically populate the feed's title field with the internal product name or URL slug (e.g. "running-shoe-mod-4521"). These titles do not contain the keywords relevant to user queries and produce low impression share in Shopping auctions. The solution is to overwrite the title in the feed with an optimised version following the structure described in the previous section.
• Incorrect or missing GTIN for branded products: for well-known brand products, Google cross-references the feed GTIN against its own database. An incorrect GTIN (manufacturer barcode that does not match) generates silent disapprovals or listing quality reductions without explicit warnings. Verifying GTINs with the Merchant Center diagnostic tool is an operation that often reveals errors on a significant percentage of the catalogue.
• Price not matching the landing page: Google periodically verifies that the price in the feed matches the price shown on the landing page of each product. Discrepancies, even temporary ones caused by feed update delays, generate disapprovals with an immediate impact on impressions. Feed update frequency should be at least daily for catalogues with dynamic pricing.
• Availability not updated: products marked as "in stock" in the feed but out of stock on the page generate serious disapprovals and can lead to Merchant Center account suspension. Availability updates must be a priority, ideally in real time or with a separate feed for availability only, updated more frequently.
• Custom labels not filled in or filled in inconsistently: catalogues where custom labels are only filled in on some products (e.g. only on bestsellers, leaving all others blank) create incomplete asset groups that collect products that are not homogeneous in characteristics. The practical rule is that every product in the catalogue must have a value for every custom label used in segmentation, even a generic value like "standard" for unclassified products.

Inference on feed diagnostics: the "Diagnostics" section of Google Merchant Center shows explicit disapprovals, but does not show sub-threshold quality issues that do not generate disapprovals but limit performance. The most useful tool for identifying these problems is the "Products" report inside Google Ads, filtered for products with low or zero impressions despite correct availability and price, a pattern that almost always indicates an attribute relevance problem, not an eligibility one.