A CSS test compares the same products placed through Google Shopping (Google's CSS) and through verteco.shop. Because a Merchant Center account has one CSS at a time, every test needs two accounts: your current one, and a sub-account under the verteco.shop multi-client account that the team creates for you (the second switch mode in the Switchover tool), both with the same product feed and both linked to your Google Ads account.
Method 1: Google Ads experiment
- Create a Shopping campaign for each account with the same products, bids and budget.
- In Google Ads open Experiments and set the two campaigns up as an experiment with a 50/50 split. Choose the user-based (cookie-based) split so one person only ever sees one arm.
- Run the experiment for at least two weeks, longer when daily clicks are low.
Method 2: two accounts, parallel campaigns
- Run one campaign per account at the same time, with the same bids, budget and targeting.
- Both campaigns compete in the same auctions and Google serves one of them per auction. Compare per-click metrics, not volumes: how many impressions each arm got reflects the allocation, not the CSS.
- To reduce the effect of the allocation, split the catalogue in two halves by item id, give each account one half, and swap the halves for the second half of the test period.
How long
The CPC comparison needs at least 30 clicks per arm; 100 or more per arm gives a clearer result. Avoid weeks with sales, holidays or feed problems in only one arm, and keep bids and budgets unchanged during the test.
What to compare
- Average CPC and CTR per arm (the main metrics).
- Conversion rate and cost per conversion, when conversions are tracked in both arms.
- Not impressions or clicks as totals: they depend on how traffic was split.
Reading the result
Enter impressions, clicks, cost and optionally conversions of both arms in the Verdict tool in Komplet. It computes the CPC and CTR differences, tests them for statistical significance (two-proportion z-test for CTR and conversion rate, an approximation for CPC, threshold p < 0.05) and states the assumptions it makes. A result that is not significant means the test needs more clicks, not that there is no difference.