Pre-tick the boxes and 83.55% accept. Leave them empty and 0.16% do.
Same cookie banner, same website, same visitors. The only change was the default state of the checkboxes. This is a field experiment on a real German online shop (1,239 and 1,380 visitors who made a decision), plus a check against about 3 million Cookiebot logs. Guess first, then flip the default.
Guess what visitors do with empty boxes
30 of 100
This website collects your data
Choose which purposes you allow.
10,000 dots, scaled from the measured share. Dot positions are shuffled; the share is the paper's Table 2.
Now be the visitor: refuse everything
Three real banner styles. Your task is the same each time. Count your clicks and seconds, then see what the field data says the design does to everyone else.
It is not one quirky website
Cookiebot supplied its own logs for a similar category banner, and they show the same split. The paper compares them in its Table 2. Share of visitors who accepted, by default state.
Study site (German online shop)
Empty n = 1,239. Pre-ticked n = 1,380.
Cookiebot logs, Feb 2019
Empty n = 1,135,090. Pre-ticked n = 1,988,681.
A louder accept button is a smaller nudge, but it still works
Confirmation-only banner. Share of visitors who clicked Accept.
| Accept shown as | Mobile | Desktop |
|---|---|---|
| Plain text link | 39.2% | 21.1% |
| Highlighted button | 50.8% | 26.9% |
| Change | +11.6 points | +5.8 points |
The paper reports the median time to submit a choice as 4 to 5 seconds for confirmation and binary banners and 7 to 8 seconds for category banners.
What a fair banner does
- Nothing is pre-ticked. Defaults decide most outcomes, as the 83.55% against 0.16% shows.
- Reject sits next to Accept with the same size, weight and colour. Design A takes one click, design B needs five.
- The second layer is optional. Detail belongs behind a link, but refusing must not live there.
- Test it with a stopwatch. Count the clicks to say no and the clicks to say yes. If they differ, that is the design.
Caveat: one site, one country, 2019, and the study counts visitors who made a decision. The direction matches the Cookiebot logs, but the exact percentages will differ on other sites.