Is ChatGPT tracking me across other websites now that it shows ads?

What OpenAI itself documents

Start with the vendor pages, because they are dated, specific, and checkable. OpenAI announced its advertising approach on 16 January 2026 and began testing ads in the United States on 9 February 2026. The same announcement page carries later dated updates: an expansion to Canada, Australia and New Zealand noted on 26 March 2026, a further expansion planned on 7 May 2026, and an update on 11 August 2026 recording launches in the United Kingdom, Mexico, Brazil, Japan and South Korea.

Who sees ads is stated without ambiguity. Ads may appear for users on the Free and Go plans. Plus, Pro, Business, Enterprise and Edu accounts do not get ads. OpenAI says it does not show ads to accounts identified as belonging to people under 18, and that ads are not eligible to appear near sensitive or regulated topics including personal health, mental health and politics. During the test, ads do not appear in the ChatGPT Atlas browser.

On answer independence, OpenAI states that ads run on separate systems from the chat model and that advertisers have no ability to shape, rank or alter responses. It also states that seeing an ad does not mean OpenAI endorses the advertiser. Those are vendor claims about vendor systems. They are not independently verifiable from outside, and this article does not treat them as verified facts, only as the company's documented position.

The part that leaves ChatGPT: the measurement pixel

OpenAI publishes developer documentation for a ChatGPT Ads Measurement Pixel. An advertiser pastes a snippet into their own website that loads https://bzrcdn.openai.com/sdk/oaiq.min.js and initialises it with a pixel ID created in the conversions tab of Ads Manager. Events are then sent to bzr.openai.com using fetch or sendBeacon, with an image request as a fallback.

The documentation is specific enough to be checkable by anyone auditing a site. It includes a Content Security Policy table instructing site owners to allow bzrcdn.openai.com under script-src, bzr.openai.com and bzrcdn.openai.com under connect-src, and bzr.openai.com under img-src. It also documents a consent function: calling oaiq with consent set to false before initialisation stops measurement-event pings, and blocked events are not replayed once consent is later granted. The pixel initialises consent to true by default unless it is set to false or it finds a stored denial.

So the structural answer is yes, OpenAI advertising code now runs on third-party websites, by design and by documentation. This is the same category of cross-site measurement that already exists across the web. It is not evidence that your conversations are being shared, and nothing in OpenAI's documentation says they are.

The disputed part, attributed rather than asserted

On 20 September 2026 an independent security writeup published at buchodi.com became the most-discussed AI item of the day on Hacker News. Its author reports reproducing a mechanism in which ChatGPT requests a short-lived signed token, exchanges it at an OpenAI collector endpoint for a cookie named __obi scoped to .openai.com with SameSite=none, and then observes that cookie being sent to OpenAI from ordinary advertiser sites running the pixel. The author states the identifier is tied to the ChatGPT account and describes cross-checking against observed traffic covering hundreds of advertiser pixels.

RepoRadar has not reproduced this. It is one researcher's claim, published on a personal threat-intelligence blog, and it is reported here as a claim with a named author and a public method description, not as a verified fact. What makes it worth reading rather than dismissing is that the half of the mechanism that can be checked against primary sources, the advertiser-side pixel and its endpoints, matches OpenAI's own published documentation exactly.

The gap between the two is the linkage: whether a cross-site identifier is joined to an individual account. OpenAI's documentation addresses what advertisers receive, and says that is aggregated and non-identifying. It does not, on the pages reviewed here, describe the internal identifier lifecycle in the terms the researcher uses. Readers who care about that specific question should treat it as open rather than settled.

The settings that actually change something

Ad controls live under Settings then Ad Controls, and OpenAI notes they are available only to Free and Go users in regions where ads are available or soon will be. Personalized ads are not initially available in the European Economic Area or Switzerland.

Four controls are documented. View History and Topics shows ads you have seen or interacted with and the topics used to target them. Delete ads data clears ads history and topics without affecting your chats, and OpenAI says full removal from its systems may take up to 30 days. A Personalize ads toggle changes ad selection. A Past chats and memory toggle decides whether prior threads and memory contribute to ad relevance; with it off, selection leans on your current thread.

The critical distinction, stated by OpenAI directly: turning personalization off can make ads less tailored, but it does not remove ads. If personalization is off you still see ads based on the context of the current chat thread. A separate Ads-Free option exists on the Free plan and does remove ads, but OpenAI says it reduces your Free usage limits in exchange. Ads-Free is a Free-plan option, so a Go subscriber wanting no ads either switches back to Free or moves up to a plan that is already ad-free.

Per-ad controls exist too: a three-dot menu on an ad offers Hide ad, Report ad, About this ad, and Ask ChatGPT. OpenAI notes that by default ChatGPT cannot see the ads shown to you unless you share one via Ask ChatGPT, and that hiding, reporting, viewing or clicking an ad does not get written into your ChatGPT memory.

How to decide what to do

If your objection is to ads existing, the only documented removals are the Ads-Free Free-plan option, which costs you usage headroom, or a paid tier that has never carried ads. Toggling personalization is not a substitute and OpenAI does not present it as one.

If your objection is to profiling rather than to ads, the meaningful moves are turning off Past chats and memory so prior threads stop feeding selection, turning off Personalize ads, and periodically using Delete ads data. Using Temporary Chat avoids writing to or reading from memory in the first place, which matters because OpenAI says memory and recent chats may be referenced during ad selection when personalization is on.

If your objection is to cross-site measurement specifically, none of the in-product toggles govern what a third-party website chooses to load. That is browser-level territory: content blockers, cookie restrictions, or a browser that partitions third-party state. Readers who saw the Hacker News thread will have noticed blocklist rules being circulated for the cookie; those are community suggestions, and RepoRadar has not tested any of them.

If you run a website and are considering installing the pixel, the documentation gives you the consent hook and the exact CSP entries to add. Whether you need a consent gate before initialising is a legal question in your jurisdiction, not a technical one, and this article does not answer it.

Limits of this answer

This is a documentation review, not a hands-on test. RepoRadar did not run a packet capture, did not inspect cookies in a live ChatGPT session, and did not audit any advertiser site.

Advertising products change quickly and this one is explicitly described by its vendor as a test that is expanding market by market. Availability, controls, and the plans affected may differ from the descriptions above by the time you read this; the vendor Help Center is the page to re-check, and it carries its own availability notes.

Nothing here evaluates whether the ads are useful or whether the trade is reasonable. That is your call. RepoRadar takes no payment from any party named in this article, and no commercial relationship influenced its selection or its conclusions.

A useful next action

Open ChatGPT, go to Settings then Ad Controls, and read what is already on: look at Topics to see what has been inferred about you, then decide about the Personalize ads and Past chats and memory toggles with that list in front of you rather than in the abstract. If the Topics list is empty or the controls are absent, you are probably on a plan or in a region that does not serve ads yet.

Then check the plan you are on against the documented list, because the simplest determinant of whether you see ads at all is which tier your account sits in.