ChatGPT now knows what you do on other websites via ad collector

Original Article ↗ Hacker News Discussion ↗

OpenAI’s rollout of an ad-tracking pixel for ChatGPT, based on standard third‑party cookie and adtech mechanisms, is raising alarm because it links users’ browsing on other sites to their ChatGPT account and conversations. Commenters argue this deepens the “surveillance capitalism” model—especially troubling for a paid AI service—while debating whether such tracking should be illegal, how it might be regulated (e.g., by the EU), and what technical countermeasures like Firefox’s cookie partitioning, DNS blocking, VPNs, and uBlock Origin can realistically mitigate. Many see it as another step toward pervasive behavioral profiling by AI platforms, blurring the line between personal assistant and targeted advertising engine.

Scope of the Tracking / What’s New

  • Mechanism is standard adtech: third‑party pixels and cookies (e.g., __obi on bzr.openai.com) to measure views/clicks and tie them to ChatGPT accounts.
  • Novelty is applying this to an AI chat product that users treat as a private assistant, including paying tiers and potentially logged‑out users via identifiers.

Comparisons to Existing Adtech

  • Many note this is essentially what Google, Meta, etc. have done for years (retargeting, behavioral tracking, cross‑site attribution).
  • Others reject “everyone does it” as whataboutism; past abuse doesn’t legitimize new entrants.

Privacy, Surveillance, and Manipulation Concerns

  • Strong sentiment that this is user‑hostile, creepy, and indistinguishable from surveillance capitalism.
  • Fears that chat logs plus cross‑site data enable unusually fine‑grained psychological profiling, propaganda, and subtle value‑shaping, not just product ads.
  • Some argue the only sustainable AI revenue is ads+tracking; others say that’s an unacceptable trade for a tool used for intimate and professional queries.

Defenses and Technical Workarounds

  • Suggested mitigations:
    • Block bzr.openai.com/related hosts via Pi‑hole, custom DNS, or DNS blocklists (e.g., Hagezi, NextDNS).
    • Use Firefox, Safari, Brave with third‑party cookie blocking / total cookie protection; plus uBlock Origin, Privacy Badger, container tabs.
    • Note that cookies are only part of the story; fingerprinting (IP, fonts, GPU, etc.) still enables tracking, so VPNs and anti‑fingerprinting help but are not perfect.

Regulation, Ethics, and Responsibility

  • Calls for laws limiting ad‑tracking and treating it as mass surveillance; some praise EU privacy efforts, others criticize EU overreach (e.g., encryption backdoor proposals).
  • Debate over whether this is a “tragedy of the commons” or of private enterprise and weak regulation.
  • Ethical debate about engineers who build “spyware”; some argue individual workers share responsibility, others emphasize corporate incentives.

Meta: Article Quality and AI-Generated Text

  • Several commenters suspect the blog post itself is LLM‑generated “slop”; others defend it as useful, especially for readers who can’t easily reverse‑engineer the traffic themselves.
  • Broader discomfort with rising AI‑generated content and calls for clearer disclosure of AI assistance.