ChatGPT now knows what you do on other websites via ad collector
summary
OpenAI uses a specific tracking cookie called __obi to link a user's activity on third-party websites to their ChatGPT account. By installing tracking code on advertiser sites, OpenAI can collect data on products searched and articles read, effectively building a profile of user behavior outside of its own platform. This mechanism operates similarly to traditional ad-tech tracking used by companies like Meta and Google, but is applied here to an AI service. The process remains active even for users who have declined marketing consent but accepted analytics consent.
key points
- The tracking system works by generating a signed identifier that is stored as a first-party cookie on OpenAI's domain, which is then sent back to OpenAI whenever a user visits a site with the OpenAI pixel installed.
- The SDK collects sensitive information from advertiser pages, including hashed emails and phone numbers, and clear-text location data, while scraping data from form fields and tag managers.
- The cookie is specifically configured with SameSite=None and Secure flags to bypass standard browser restrictions that block other OpenAI cookies on third-party sites.
who should read it
Privacy-conscious internet users and security researchers interested in how AI companies handle user data and cross-site tracking.
discussion skeptical
consensus
Commenters largely view the integration of adtech tracking in AI products as an extension of existing, invasive data brokerage practices rather than a new phenomenon. There is a shared skepticism toward the effectiveness of EU regulations like GDPR, with many arguing that fines are merely a cost of doing business for tech monopolies. The overall sentiment is cynical, reflecting a belief that privacy is essentially dead and that both corporate and government entities prioritize surveillance over user rights.
top insight
The discussion highlights a critical distinction between the ability of AI to solve complex mathematical problems and its use as a tool for 'slop' and psychological manipulation. One commenter argues that LLMs are essentially a 'maximally optimized Weizenbaum feed,' suggesting that while the technology is powerful, its primary commercial application is to mimic intelligence to manipulate and mine data from users, similar to early chatbots like ELIZA.
Generated by an LLM from the article and thread. Check the sources before relying on it.