Tracking5 October 2026
From cookie to server-to-server: how a sale actually gets attributed
Click, cookie, purchase, commission: the diagram is simple and has been wrong for years. Here is what actually happens, and what the French regulator says about it.
By Kamel Rezigui
France Market Lead, Affiliate
The classic diagram has four boxes. The visitor clicks a tracked link; a cookie is dropped in their browser; they buy within thirty days; the publisher is paid. That diagram is right in principle and out of date in its details, because the three things it rests on (the third-party cookie, the lifetime, and a cooperative browser) have all changed.
Consent: the CNIL answers no
First point, and it often comes as a surprise: in France, affiliate tracking requires the user’s consent. The CNIL is explicit. Asked whether trackers used to bill affiliate operations are exempt from consent, its answer is one word: "Non". Such operations do not fall within the exemptions of article 82 of the French Data Protection Act, which must be interpreted strictly. In other words, a visitor who refuses cookies is not tracked: and the sale they make is not attributed.
Safari caps cookies at seven days
Second point, on the browser side: since version 2.1 of its Intelligent Tracking Prevention, Safari caps persistent cookies set by JavaScript via document.cookie at seven days. An attribution window advertised as thirty days is therefore not thirty days for everyone. On a base where Safari dominates (typically a premium mobile audience) the gap is structural, and it always cuts the same way: against the publisher.
Third-party cookies are staying in Chrome
Third point, which runs counter to years of announcements: third-party cookies are not disappearing from Chrome. Google abandoned its planned removal and now says it will maintain its current approach of offering users a third-party cookie choice. In its wake, the Privacy Sandbox programme has been largely dismantled: the Attribution Reporting API, Topics, Protected Audience and seven other technologies are being retired for lack of adoption. Only CHIPS, FedCM and Private State Tokens survive, with work on an interoperable attribution standard continuing at the W3C.
The consequence is counter-intuitive. The deadline that was going to force everyone to modernise their tracking never arrived, but the reasons to modernise remain fully intact: mandatory consent, Safari’s caps, ad blockers, and mobile apps that do not run a browser cookie at all.
Server-to-server tracking, and its limits
That is what drives the move to server-to-server tracking. The principle: instead of reading a cookie in the buyer’s browser, the advertiser’s server sends the conversion straight to the platform’s server, carrying a click identifier captured when the visitor landed. Measurement becomes independent of the browser, of blockers and of lifetime caps.
Two cautions are needed so you know what you are buying. S2S is more reliable, not magic: it requires storing the click identifier on the advertiser side and carrying it through to the order, which is real engineering work. And above all, it does not exempt you from consent: the CNIL reasons by purpose, not by technology. Moving a tracker from the browser to a server does not take it out of the scope of article 82.
Sources
- CNIL, Cookies et autres traceurs, questions-réponsesQuestion 13: "Are trackers used to bill affiliate operations exempt from consent?", Answer: "Non".
- WebKit, Intelligent Tracking Prevention 2.1Apple’s announcement of the seven-day cap on persistent cookies set via document.cookie.
- Google, Update on Plans for Privacy Sandbox TechnologiesGoogle lists the ten retired technologies there and confirms it is maintaining user choice on third-party cookies in Chrome.
