A product can be approved in Merchant Center, included in a Performance Max campaign and still receive little useful traffic. Those are three different states. Large catalogues hide the distinction because campaign ROAS can look healthy while delivery concentrates on a relatively small set of proven items.

That concentration is not automatically a problem. Bestsellers often deserve more spend. The problem starts when commercially important products never collect enough evidence to judge, and the team cannot tell whether the cause is demand, feed quality, bidding or simple lack of exposure.

Build the product exposure funnel

Start with a complete product list for one market and one consistent reporting window. Count:

  1. Active, in-stock products in the commerce platform
  2. Products eligible in Merchant Center and Google Ads
  3. Eligible products included in the campaign’s listing groups
  4. Included products with at least one product impression
  5. Included products with at least one direct product click

Keep the zero rows. Exporting only products with clicks removes the very items the exercise is meant to find. Google’s product reporting now covers products across Shopping and all Performance Max networks, while the Shopping Product report can return item IDs, status and performance metrics together.

Check eligibility before interpreting traffic. A disapproved item has a feed or policy problem, which belongs in the feed-quality workflow. An eligible item with sparse traffic has a different question attached to it.

Measure concentration from your own catalogue

There is no credible universal percentage of products that should receive traffic. Calculate the pattern in the account instead.

Rank eligible items by product impressions, direct product clicks and cost. For each metric, calculate the share captured by the top 10% of items. Then report the median and quartiles across the whole eligible set, including zeros. For new products, track time to first impression and first product click by category, price band and margin band.

Use product impressions carefully. Google counts an impression for each product included in an ad event, even when the item was outside the shopper’s viewport. A direct product click is a cleaner signal that the item itself attracted attention. Google documents the distinction in its Shopping reporting guide.

One reporting break also matters: from 15 June 2026, the Shopping Performance View includes all Performance Max networks. Comparisons that cross that date can show a mechanical increase in product metrics.

Treat cold start as an evidence gap

It is tempting to say that Performance Max knows nothing about a new SKU. Google’s own explanation is more nuanced. Smart Bidding can use query-level information when conversion history is sparse, then refine its models as more granular data arrives.

The useful diagnosis is an evidence gap. A product with no direct clicks cannot produce direct click-to-sale evidence during that period. That still does not prove the algorithm suppressed it. Weak demand, restrictive ROAS targets, low Ad Rank, competition, seasonality and poor product data remain possible causes.

This distinction changes the action. Some unseen SKUs have no commercial case for further spend. A new, high-margin or strategically important item may deserve a controlled chance to collect evidence.

Fund a cohort that is worth testing

Create a discovery cohort with a stable Merchant Center custom label. Include only products that are eligible, in stock, commercially viable and backed by complete product data. Group them by a real business reason such as new range, high margin or strategic category.

Exclude that cohort from the evergreen campaign, then place it in a separate Performance Max campaign with its own budget or ROAS target. This removes product overlap and creates a clear funding boundary. Google’s retail guidance allows separate PMax campaigns when new products or other ranges need different budgets or targets.

Set the spending ceiling before launch from margin and acceptable acquisition cost. A separate campaign will not spread traffic evenly across every item. It gives the cohort room to compete. Avoid one campaign per SKU and repeated product reshuffling. Both fragment data and make the result harder to interpret. The broader guide to Performance Max controls covers the surrounding campaign decisions.

If the test needs explicit product-group bids or an advertiser-set max CPC, use Standard Shopping with Manual CPC. Check bid adjustments, Search Partner settings and any feature that can raise the effective bid before treating that amount as a hard click-cost ceiling. Keep treatment products out of PMax during the test. When the same products and targeting overlap, Ad Rank decides whether PMax or Standard Shopping serves, so campaign reports no longer provide a clean comparison.

Test coverage and commercial value together

For a stronger comparison, match products by category, price band, margin band, product age and feed completeness. Randomly assign items within those groups. Keep the control products in the evergreen setup and move the treatment products into the funded discovery cohort.

Treat this product-level comparison as directional unless the assignment, sample size and auction design support causal inference. If the question is PMax versus Standard Shopping for the same products, use Google’s native traffic-split experiment.

Freeze membership and primary settings during the test, except when stock or policy status requires removal. Google recommends at least six weeks before evaluation and says to exclude the first one to two weeks of ramp-up from test analysis. Set the analysis end date from the account’s conversion-delay pattern so recent clicks have had time to mature.

Read the result on two levels. First, did the treatment increase the share of eligible items with product impressions and direct product clicks? Second, was the added coverage associated with acceptable attributed sold-product revenue or gross profit, and did backend contribution support the same decision?

That second question prevents a coverage project from becoming a traffic-buying exercise. Conversions with Cart Data can separate the product advertised from the product actually sold and expose gross-profit metrics where the implementation is reliable. Graduate products that earn a commercial case, stop those that exhaust the agreed test budget, and send the next coherent cohort through the same process.

Use the process to distinguish products that lacked evidence from products that received a fair test and failed it. Equal spend across the catalogue would only replace one allocation problem with another.