A new study has found that popular AI chatbots — including ChatGPT, Claude, Grok, and Perplexity — transmit user data to third-party ad trackers. The sharing happens even when people decline cookies, the researchers say, pointing to significant privacy risks in how these tools handle personal information.
Data shared despite cookie opt-outs
The study tested each chatbot by sending prompts and observing what data left the user's device. In all four cases, the researchers detected data packets headed to domains linked to advertising and analytics firms — even after users clicked “decline” on cookie consent banners. The finding suggests that cookie rejection alone doesn't stop the flow of information to trackers.
The four chatbots under scrutiny
The researchers examined OpenAI's ChatGPT, Anthropic's Claude, xAI's Grok, and Perplexity AI's namesake assistant. Each exhibited similar behavior: user inputs, device details, and session identifiers were shared with third-party services not essential to running the chatbot. The study did not name which specific ad trackers received the data, but confirmed the destinations were commercial tracking networks.
Privacy risks for users
When someone types a question into an AI chatbot — whether for work, health advice, or personal topics — that text can end up linked to a tracking profile. The study warns that the combination of conversation content and identifying metadata could allow advertisers to build detailed profiles of users. Privacy advocates have long flagged such practices, and this research provides concrete evidence that the risk extends to the newest generation of AI tools.
The study's authors stress that the issue is not unique to one company. All four chatbots tested engaged in data sharing regardless of cookie preference. That uniformity points to a broader industry pattern rather than a single oversight.
The findings come as regulators in Europe and elsewhere tighten rules around data collection. Under laws like the GDPR, transferring personal data to third parties without clear consent can carry heavy fines. Whether the data shared by these chatbots qualifies as “personal data” under current definitions will likely be a central question in any future enforcement action.
For now, users who want to limit tracking may need to take additional steps beyond declining cookies — such as using ad blockers, reviewing privacy settings, or choosing chatbots with clearer data policies. The study did not test whether those measures are effective.
