late-teacher-8634808/25/2022, 4:29 PM
if they're logged-in and falling back to
if they're not. That felt a little hacky though, and we wanted some reassurance!
will cause users to potentially switch variations when they log in. Our recommended approach is to pick just one identifier type for each experiment. Depending on the specific experiment, you might pick
. As a general rule of thumb, if the test is shown primarily to only logged-in users, then pick
. Otherwise, pick
. If a test has a large amount of both logged-in and anonymous users, I would usually just use
to split traffic. Users who switch devices or clear cookies may see multiple variations, but for most tests and products, that should be a relatively small percent of users and unless it's something super critical like pricing, not a huge deal if a few people see both variations.
late-teacher-8634808/25/2022, 7:41 PM