AI Chatbots Are Sharing Your Conversations With Advertisers — Even When You Opt Out
A new study finds that ChatGPT, Claude, Grok, and Perplexity all transmit user conversations to third-party ad trackers, sometimes bypassing cookie consent mechanisms entirely.
When users type a prompt into ChatGPT, they expect that conversation to stay between themselves and OpenAI. A study released on 7 May 2026 suggests otherwise. Researchers found that leading AI chatbots — including ChatGPT, Claude, Grok, and Perplexity — routinely share snippets of user conversations with third-party advertising trackers, and in some cases do so even when users have explicitly declined cookies.
The findings, reported by Decrypt, centre on a privacy mechanism known as "IPM tracking" — a method where parts of a user's prompt are embedded directly into URL parameters sent to ad networks. When that URL loads in a browser, it can trigger data collection before any cookie consent banner is even displayed. The result is that advertising companies gain access to conversation fragments, and by extension inferences about user intent, mood, and behaviour, without meaningful consent.
How the Data Flows
The study examined four platforms: OpenAI's ChatGPT, Anthropic's Claude, xAI's Grok, and Perplexity. Each was found to transmit some form of user content to trackers operated by companies including Meta, TikTok, and Google. The mechanism varied by platform. Some injected conversation text into referrer headers; others embedded keywords in URL strings that were then logged by analytics scripts embedded on the page.
The critical finding is timing. In controlled tests, the trackers fired before any consent management platform had an opportunity to block them. Cookie banners displayed after a user declined tracking were, in effect, beside the point — the data had already left the device. Decrypt's reporting notes that this behaviour is technically distinct from traditional cookie-based tracking, which means it may fall outside the scope of consent mechanisms designed for earlier web architecture.
The Industry's Defence
The companies involved have not issued on-record statements responding to the specific findings. The broader defence offered by AI companies in similar circumstances typically rests on terms-of-service disclosures and the argument that using their products constitutes acceptance of data processing. That argument has found limited purchase with privacy regulators in Europe, where consent must be freely given, specific, and unambiguous — conditions that automated data transmission arguably fails to meet.
Advertising networks, for their part, have long argued that tracker-based targeting produces better consumer outcomes by reducing the frequency of irrelevant ads. Whether that trade-off justifies the interception of private AI conversations is a question regulators in multiple jurisdictions are beginning to take seriously.
Why This Matters More Than a Standard Cookie Violation
AI chatbots occupy a different category from search engines or news sites. Users frequently share sensitive material — medical questions, legal queries, workplace frustrations, personal communications — with an assumption of confidentiality that they would not extend to a search bar. The intimacy of the interface makes data leakage not merely a privacy technicality but a substantive betrayal of user trust.
This is not a hypothetical risk. The study's authors note that conversation fragments transmitted to ad networks could be correlated with other data signals — device fingerprints, IP addresses, login states — to build surprisingly detailed profiles. A user who tells Claude they are considering a major purchase, for example, may find that inference surfacing in a downstream advertising auction before they have closed the tab.
Enterprise deployments compound the problem. organisations that have paid for ChatGPT Team or Claude Business tiers may assume they are operating in a closed environment. The study suggests that tracker injection occurs regardless of subscription tier, meaning corporate conversations are potentially as exposed as consumer ones.
Regulatory Reckoning Approaches
The EU's GDPR enforcement regime has levied significant fines against platforms for lesser infractions. Ireland's Data Protection Commission, which oversees Meta and Apple under EU law, has signalled interest in scrutinising AI-adjacent tracking practices. If a formal investigation opens, the industry's terms-of-service defence will face a substantive legal test.
In the United States, the absence of a federal privacy law means enforcement is fragmented. The FTC has broad authority to pursue unfair or deceptive practices, but its docket is crowded. State-level action, particularly in California under the CPRA, remains the most likely American avenue for accountability.
What is clear is that the current model — where AI companies monetise attention through advertising infrastructure while presenting themselves as private, intelligent assistants — contains a structural contradiction that regulators are no longer willing to overlook. The study published on 7 May 2026 provides them with the technical record they needed to begin that work in earnest.
This publication covered the Decrypt investigation as the primary source. Wire services carried the story with brief summaries; Monexus drew on Decrypt's technical reporting for all specifics cited here.