OpenAI has implemented a cross-site cookie in its ad measurement system that ties a person’s activity on third-party advertiser sites back to their ChatGPT account. The cookie, called __obi, is created when a user accesses ChatGPT and can be activated again when that same user later visits websites using OpenAI’s ad pixel.
How the Tracking Works
Upon each session, ChatGPT generates a random identifier and requests a short-lived, signed token. This token includes three key pieces: a subject value linked to the user’s account, the __obi identifier, and a flag indicating whether analytics tracking is allowed. The token is sent to a backend service known as Bazaar (operating under bzr.openai.com), which then sets the __obi cookie on the .openai.com domain.
The cookie is configured with SameSite=None and Secure flags, meaning it can be sent with cross-site requests and only over HTTPS. It remains valid for up to one year. Its behavior contrasts with other OpenAI cookies, which have stricter limitations (like SameSite=Lax) or are blocked during third-party requests due to domain restrictions.
What Data Is Collected and When
Advertisers using ChatGPT ads can embed OpenAI’s measurement pixel into their sites. When visitors land on those sites and carry the __obi cookie, their browser may attach the identifier to requests, sending OpenAI data about pages visited, purchase paths, and conversion events. Although query parameters are stripped from URLs, page paths remain, and have been observed to include sensitive topics like medical conditions, debt solutions, and legal services.
OpenAI’s pixel also collects additional user data. In practice this includes advertisers passing data explicitly, but often includes scraping data from page content or form fields, including names, hashed emails, phone numbers, location data (from country down to postal code). Identity info automatically pulled from tag management systems appeared more frequently than that submitted directly by the advertiser.
This tracking persists even if a user is not logged into ChatGPT. In testing (primarily on Chrome for Android), an anonymous version of the subject identifier remained stable for at least 27 days. Because Safari’s Intelligent Tracking Prevention blocks third-party cookies, the mechanism appears inactive on browsers using WebKit, such as Safari and iOS browsers.
Consent and Classification Issues
Despite its use in behavior-tracking across sites, OpenAI labels __obi as an analytics cookie, not marketing. The signed tokens that enable its issuance are provided when users grant analytics consent—even in cases where they have declined marketing consent. OpenAI has acknowledged being asked for clarity about this, but has yet to supply full details about how consent classes determine when and how the cookie is used.
This blend of cross-site tracking, data collection, and consent classification raises multiple privacy red flags. Paths exposed—though stripped of query data—still reveal user interests and potential vulnerabilities, especially in sectors like health, debt, or legal services. The fact that this works even for users who are anonymous further complicates typical notions of privacy protection.
OpenAI has been made aware of these practices and is being pressed on whether the cookie classification aligns with user expectations and regulatory norms. As of now, there has been no published detailed response from the company clarifying how consent is honored across the analytics vs. marketing divide.
Why This Matters and What Might Happen Next
This discovery underscores just how thin the line has become between analytics and marketing when it comes to data collection—and how companies can leverage advanced cookies and pixels to track user behavior across the web. With growing scrutiny from privacy regulators worldwide, actions like this are likely to draw investigations into whether informed consent is truly being obtained, especially under EU law, California’s privacy rules, or similar regimes.
For users, this means paying attention not just to privacy settings, but to how cookies are described and whether opt-outs are truly effective. For developers and advertisers, the risks of compliance failures are rising sharply.
The rise of this kind of tracking also feeds into broader conversations about privacy architecture: browsers restricting third-party cookies, laws redefining consent, and possibly more transparency required of companies collecting cross-site data. What to watch next: regulatory responses, potential browser changes, and whether OpenAI or other AI platforms adapt policies to more clearly separate analytics from marketing.
