At DuckDuckGo, we don't track you, ever. That's our Privacy Policy in a nutshell. To improve Duck.ai, we develop completely anonymous ways to test changes with a subset of users before rolling them out to everyone. For example, we might test an adjustment to how chat models respond to see whether users rate the answers as more helpful. We've designed these experiments to work without building any profile of you or your activity. We do not use or store your IP address when running experiments.
How experiments work
When an experiment is running, your browser may be randomly assigned to a test group — for example, the group that sees a change, or the group that doesn't. Duck.ai saves this assignment in a cookie on your device so that your experience is consistent. Each experiment uses its own separate cookie, containing only:
- The version of the experiment and which test group your browser was assigned to.
- Whether your browser is participating in the experiment.
- Technical details about the experiment itself, like the percentage of users it includes and the cookie's expiration date.
- A digital signature that prevents tampering, computed only from the values above.
Nothing in the cookie identifies you. Every value stored in the cookie describes the experiment, not you or your device, and each field can only take a handful of possible values shared by large groups of users. Each cookie has a fixed expiration date, which is the same for every participant, and once an experiment expires, the cookie and any related data stored on your device are deleted.
Some experiments for Duck.ai paid subscribers work without a cookie, instead deriving test groups from the anonymous subscriber ID associated with each subscription. For these experiments, nothing is saved on your device. Experiments that include both free users and subscribers use the cookie described above for everyone.
How we measure results
To learn whether a change worked, we count how often each test group takes certain actions, like starting a chat. We do this by adding an experiment label to anonymous usage statistics we already collect. The label contains the name of the experiment, the experiment version, and which test group your browser was assigned to. It may also indicate whether your browser has reported that statistic before, using a marker stored on your device, so that we only count each browser once. None of the information in this label is unique to individual users.
If you subscribe to DuckDuckGo then, for some experiments, no marker is stored on your device. In these cases, statistics within an experiment are grouped using a code created from your anonymous subscriber ID with a one-way computation, meaning they can't be linked back to you or your subscription. Each experiment uses a completely different code, and we recompute it whenever it's needed instead of storing it.
Deleting experiment data
Experiment cookies and markers are stored like ordinary website data, so clearing your browsing data for Duck.ai removes them. If you clear this data while an experiment is running, your browser may be randomly included in the experiment again the next time you visit Duck.ai.
Experiments are subject to our internal privacy review process, and will never access the contents of your chats. See How does Duck.ai protect my privacy? for more.