ChatGPT Sent Chat Titles to Google Analytics
Published October 20, 2026· By Juan Martinez
Researchers found the free ChatGPT website sent the link and title of each chat to Google Analytics, even after users rejected cookies. What it means for you.

Most people think of a chat with an AI as something between them and the app. A study published in May 2026 shows that this is not always true. A team at a Spanish research institute tested nine popular AI chat services and found that all nine contacted at least one advertising or tracking company. In six of the nine websites, the link to a conversation, its title, the text of a question or even screenshots reached outside companies, often together with an ID that points to the user.[2] The tests were run in Spain, and the findings went public between 4 and 6 May 2026.[2]
ChatGPT was one of the six. The researchers report that when you use the free ChatGPT website while signed in, it sends two things to Google Analytics: the link to each of your chats and the chat’s title.[1] This happened whether you clicked “accept” or “reject” on the cookie banner.[1]
Google Analytics is a tool that a huge number of websites use to count their visitors and study what those visitors do: which pages they open, how long they stay, where they came from. The information goes to Google, which runs the tool.
When the study came out, no fix had been reported, and the researchers’ page records no reply from OpenAI.[1] That is the state of things as of their publication in May 2026.
Why a title says a lot
Every chat in ChatGPT gets a short title in the sidebar. You do not write it yourself: ChatGPT creates it automatically from what you ask. That is handy for finding an old conversation, but it also means the title usually names the topic in a few plain words.
Here are three made-up examples, just to show the idea. Someone who asks about a strange mole might see a chat called “Signs of skin cancer”. Someone sorting out their finances might get “Paying off credit card debt”. Someone worried at work might get “Rights after being fired”.
None of these contain the full conversation. They don’t need to. A title like that, sitting next to an ID that points to you, tells an outside company what was on your mind that day. People tend to guard the long text of their questions and forget about the short label on top.
What rejecting cookies does and doesn’t do
The cookie banner feels like the place where you decide what gets shared. Across the services they tested, the researchers found that rejecting non-essential cookies reduced contact with advertising and analytics companies, but often did not stop it.[3] In ChatGPT’s case, the link and the title went to Google Analytics either way.[1]
OpenAI’s own privacy policy, last updated on 30 July 2026, says the company shares data with web analytics services and marketing partners. It also says Free and Go users see ads, and that there are ad controls in account settings.[4]
What you can do
Start with who this applies to. The researchers describe this behaviour for free users who are signed in on the ChatGPT website.[1]
If that is you, the simplest habit is to use the free website for everyday things: a recipe, a cover letter, a trip plan. Keep questions about your health, your money, your legal troubles or your family out of it. A good test is to picture the chat’s title on a stranger’s screen. If that would bother you, don’t ask it there.
ChatGPT has a Temporary Chat mode and a setting to keep your chats out of training. Nothing the researchers published says that either of these stops the link and title from reaching Google Analytics, so don’t count on them for that.
The same study also looked at the Perplexity website. The researchers found that it sent chat links to Meta until 3 April 2026. Their page also lists two analytics companies, Datadog and Singular, that still received user identifiers whatever you chose on the cookie banner.[1]
What our review says
Our ChatGPT review lists this finding as a minus. The score stays the same, because the review already counts two bigger problems: it keeps your conversations in a form OpenAI can read, and it trains on them unless you opt out. The findings date from May 2026, and the researchers have not recorded any response from OpenAI.