Did My New Product Image Win, or Was It Shown in Easier Ad Placements?

List the placements each creative could enter

Start with the actual advertising product. Record every eligible placement, the creative format required there and whether each asset met that requirement.

A square product image may be eligible for a different set of presentations than a vertical composition. If one version cannot enter a placement, the comparison is not simply two images shown under identical conditions.

Separate three questions: what placements the campaign allowed, which placements the creative was technically eligible for, and where impressions actually occurred. An allowed placement with zero delivered impressions is not the same as a placement that was excluded.

Google Ads documents experiment structures that divide eligible traffic and budgets, but the exact placement reporting and available controls depend on the campaign type. Do not assume that a platform-wide experiment feature exposes every placement-level detail for every advertising product.

Inspect the actual platform report and retain the creative's aspect ratio, version and approved product content.

Look for aggregate-versus-within-placement disagreement

Record the same outcome definition and denominator for A and B within each comparable placement. Then separately calculate the aggregate result using the actual exposure counts.

If A receives most of its impressions in a placement with generally higher engagement and B receives most of its impressions elsewhere, A can lead overall without leading within either placement.

This is a composition problem: the aggregate combines different exposure distributions. It does not by itself prove that the platform favored a particular creative, that a placement caused the difference or that shoppers prefer one layout.

Do not repair the report by averaging placement percentages equally when the actual traffic volumes differ. A simple unweighted average answers a different question from the observed aggregate.

When a placement cannot display one creative properly, treat that as an eligibility or delivery issue rather than manufacturing a like-for-like performance comparison.

Hypothetical two-placement reversal

Hypothetical arithmetic exercise—not a campaign result: a fictional merchant compares two accurate backpack images across Placement P and Placement Q.

In P, A receives 900 impressions and 90 clicks, while B receives 100 impressions and 12 clicks. In Q, A receives 100 impressions and 2 clicks, while B receives 900 impressions and 27 clicks.

Within P, B has the higher descriptive click rate: 12% versus A's 10%. Within Q, B again leads: 3% versus A's 2%. Yet A has 92 clicks across 1,000 impressions, or 9.2%, while B has 39 clicks across 1,000 impressions, or 3.9%.

The apparent aggregate winner is A even though B leads within both specified placements. The explanation is the very different distribution of impressions across P and Q.

These invented values illustrate why placement composition matters. They are not evidence of a statistically reliable creative effect or a real platform's behavior.

Decide whether to rerun or qualify the comparison

If the intended question is which image performs better within a specific placement, compare eligible versions there using a suitable controlled method. If the business question concerns the entire planned campaign mix, define that mix and the evaluation method before testing.

Do not silently remove inconvenient placements after seeing the results. Preserve the original aggregate report and label any placement-specific follow-up analysis.

If one creative was cropped incorrectly or lacked a supported format, repair that delivery defect and assign a new version. Historical observations belong to the old version.

For the hypothetical backpack case, the justified conclusion is that aggregate click rates are confounded by placement mix. The next step is to verify the actual placement breakdown and decide whether a controlled placement-specific comparison is feasible.

CartVisual can provide visual candidates, but placement assignment and measured outcomes are controlled by the publishing platform.

Record your own observations

The filled rows are illustrative examples. Use the blank row for your own files and observations.

Download the editable CSV worksheet

What the references establish

• Google Ads Help — Set Up a Custom Experiment: Documents campaign experiment allocation and the need to distinguish controlled comparison settings from ordinary delivery.
• Google Ads Help — About the Experiments Page: Describes comparison of campaign settings over a specified period; does not guarantee identical placement composition.