The number everyone quotes that nobody published
When you use Google's own CSS, a fixed share is taken out of every bid before it enters the auction. The industry says 20 %. Google does not publish the figure, not in its documentation and not in the account, and we have not found a primary source that does. Every number you see on this subject, ours included, is therefore measured from the outside or copied from somebody else who measured from the outside.
That is why this page exists. Not because our measurement is large, but because there are almost no public measurements at all.
What we measured
A Danish ecommerce account moved from Google's own CSS to ours in March 2026. The numbers are click weighted, pulled from the Google Ads API, and cover the two months before against the two months after on the single campaign that ran uninterrupted throughout.
| Before | After | Change | |
|---|---|---|---|
| Shopping impression share | 21,4 % | 26,8 % | +25,3 % |
| Impression share lost to rank | 75,5 % | 68,7 % | 6.8 pp better |
| Average cost per click | 10,29 kr | 9,25 kr | -10,1 % |
| Spend | 34.500 kr | 31.600 kr | -8 % |
All three performance metrics improved at the same time while spend fell by 8 %. The gain was not bought with a larger budget.
The 25.3 % on impression share is worth pausing on. The theory says a bid placed through an independent CSS behaves like a bid roughly 25 % higher, because the margin is no longer deducted. We measured 25.3 %. That may be coincidence, and we treat it as one.
Why this is not proof
If March 2026 were a real effect, it should stand out from every other point in the series. So we ran every possible switch point from May 2025 to May 2026 through exactly the same method, two months against two months.
| Metric | March 2026 ranks |
|---|---|
| Impression share | 3rd of 13 |
| Impression share lost to rank | 4th of 13 |
| Cost per click | 3rd of 13 |
March 2026 is among the best, but it is not unique. August 2025 showed a similar pattern across all three metrics with nothing happening on the CSS side. There is also no control group: a CSS switch applies to the entire Merchant Center account at once, so no part of the account carried on unchanged for comparison.
The honest conclusion is an indication, not proof, and we do not sell it as anything else.
Measure it yourself, and measure the right thing
The most common mistake is to look at cost per click. Smart bidding notices it is beating its target and buys more volume, so CPC can sit perfectly still while you get more for the same budget. Measure impression share and impression share lost to rank instead.
In Performance Max the advantage applies to Shopping inventory only. Search text ads, Display, YouTube and Gmail are unaffected, so a combined campaign report dilutes the effect.
The fields we used: search_impression_share, search_rank_lost_impression_share, average_cpc, clicks and cost_micros on the campaign resource, segmented by month. Weight the levels by clicks, otherwise a day with three impressions counts as much as a day with three thousand.
And write down the day, not the month. The exact switch date was never recorded on this account, and that single omission is the most expensive gap in the whole measurement.
What we do from here
One account is an observation. We now measure every new customer before and after with the same method and the same fields, and we publish the numbers here as they come in. Only at around ten measurements does the noise average out into something that can be quoted without caveats.
Measured 31 August 2026. Source: Google Ads API, click weighted, single campaign, Danish ecommerce account. Published 21 September 2026. Write to info@thecsspartner.com for the full method, or to have your own account measured.