Full Answer

The mechanism is link decoration detection. Safari maintains a view of parameters commonly used to carry an identity from a referring site, and when it sees one on a cross-site navigation it treats the destination's storage as suspect. Historically this applied mainly in Private Browsing and to domains already classified as trackers, but the trajectory has been steadily toward on-by-default.

Click identifiers are the clearest casualties: fbclid from Meta, gclid and its variants from Google, and equivalent IDs from other ad platforms. These are exactly the parameters attribution depends on, because they are how a platform recognises that the visitor in front of you is the one who clicked a specific ad. Standard utm_* values are campaign labels rather than per-user identifiers, so they survive more often, but they can't reconstruct user-level attribution on their own.

Safari 26 escalated this by enabling link-tracking parameter stripping and Advanced Fingerprinting Protection for all users. The takeaway for store owners is that you cannot assume the URL you receive is the URL the visitor clicked. The defence is to read and persist any identifier server-side at the first request, before stripping and cookie caps apply, rather than trusting the browser to carry it through to checkout.