When did you last touch your ChatGPT privacy settings without hunting through three menus first?

If the answer is never, the ChatGPT Privacy Center was built for you. Probably.

The ChatGPT Privacy Center is a new page inside ChatGPT that gathers every privacy control into one place and explains what each one does. It changes nothing on its own. It deletes nothing. It only shows you where the switches live. That sounds modest, and it is. It also arrived during the noisiest stretch ChatGPT privacy has seen in a while: an investigation revealed hundreds of contractors reading real user conversations; a proposed class action followed two days later accusing OpenAI of never properly disclosing it; then, the next week, the hub started rolling out.

The ChatGPT Privacy Center Is a Map, Not a Switchboard

OpenAI announced the hub in ChatGPT’s release notes, careful about what it is not. This is no new dashboard. Your ChatGPT privacy settings remain in Settings, exactly where they were. The ChatGPT Privacy Center explains what each control does and walks you to it. That is the entire job description.

Inside, everything is sorted into three groups. Personalized chats and ads covers Memory, personalized ads, and location. Chat privacy holds a single topic, temporary chat, with a button to start one on the spot. Data use and access covers connected apps and plugins, multi-factor authentication, model improvement, and exporting or deleting your data. Seven topics total. Each one opens a short panel with a few questions and a Manage button. The button carries a small arrow, because it takes you out of the center and into Settings.

The ChatGPT Privacy Center reaches signed-in users on the Free, Go, Plus, Pro, and Business plans, on the web and in the iOS and Android apps. Enterprise, Edu, and Healthcare accounts are out. The rollout is gradual, and OpenAI says so itself: if you do not see it yet, it may simply not have reached your account or app version. What you see can also shift with your plan, region, and workspace settings.

Four short principles sit at the bottom of the page: OpenAI does not sell your data or share conversations with advertisers, it secures information with encryption and monitoring, you control your data, and you control whether your chats improve its models. None of that is new. A page built to explain existing settings was never going to contain surprises.

Opening the ChatGPT Privacy Center Takes About Ten Seconds

Reading Notebookcheck’s walkthrough of the launch, what stuck with me was a small, concrete detail. Their reviewer opened the ChatGPT Privacy Center the morning after it appeared, in a European account on the free plan, and found the whole thing already localized. The observation they led with was the Manage button’s little arrow. It tells you the truth about the page before you read a word: every change happens somewhere else. They also named the two stops that repay the minute they cost, multi-factor authentication and the connected-apps list. I would add a third. We will get to it.

If the entry is missing on your account, give it time. The rollout is gradual, and a missing menu item is a waiting game, not a malfunction.

The Week Before It, Nobody Was Talking About a Menu

To understand why a settings index landed like news, rewind about a week. 404 Media’s Joseph Cox published an investigation into an internal OpenAI program called Project Lily, built from leaked documents and conversations with people who do the work. The shape of it: hundreds of contractors, hired through a staffing firm and paid through an AI-training vendor at rates above fifty dollars an hour, reading real ChatGPT prompts, sometimes whole conversations, then summarizing what the user wanted and scoring the model’s replies on a one-to-seven scale. The goal was steering ChatGPT away from sycophancy and from pretending to be human.

OpenAI’s defenses were on the record. Reviewers do not see usernames, and conversations pass through an automated Privacy Filter first. But the company acknowledged to 404 Media that sensitive details can still slip through the filter. Reviewers also see something else above each prompt: a “user memories” summary that can reveal what someone has used ChatGPT for before and roughly where they live. Strip the username all you want. The little bio pinned above the chat does its own talking.

Reading Cox’s investigation, the line that stayed with me was not a number. It was a contractor’s answer when asked whether users know humans read their chats: “No… I don’t think they would imagine some contractor somewhere is analyzing the conversations.” The whole story in one sentence. Most of ChatGPT’s users picture a private exchange with a machine. The investigation showed a workforce in the loop.

Two days later came the lawsuit. Two California consumers filed a proposed class action, Vredenburgh v. OpenAI, alleging the company never adequately disclosed that outside contractors could read, summarize, and score real conversations. The complaint points at the company’s own privacy policy, which names categories of outside data recipients but never specifically names data-annotation or human-evaluation vendors. It asks for damages and for injunctions that would flip the defaults: opt-in consent before human review, and the training toggle off unless you turn it on. All of it is allegation, not verdict. No class has been certified. There is nothing to claim and nothing settled.

The complaint is, at bottom, a disclosure fight: what OpenAI told consumers versus what its reviewers could see.

The record is silent on one point. Nothing in the reporting or the filing claims OpenAI timed the hub to answer them. The sequence is just the sequence: the investigation, the lawsuit two days later, the ChatGPT Privacy Center the next week. I am not connecting those dots beyond that, because there is nothing verified to connect them with.

One caveat before the framing runs away: human review of chatbot data is not new. Canadian privacy regulators confirmed earlier that OpenAI had used a subset of user interactions for model improvement with employee and contractor review, and found the filtering incomplete. What the Lily reporting added was the scale and the mechanics: a codename, a staffing firm, a pay rate, and the memory summary above the prompt.

Three Things the ChatGPT Privacy Center Says Out Loud

Here is where the hub earns its keep: buried in those seven panels are three sentences most users have never seen spelled out, each correcting something people commonly get wrong.

First: deleting a chat does not delete the memory it created. Memory and chat history are separate stores. You can wipe every conversation and still have ChatGPT remember your job title, your preferences, the project you mentioned two months ago. A real cleanup means deleting the saved memory and the chat where you shared it, then checking the memory summary, archived chats, files, and connected apps. The ChatGPT Privacy Center says this outright. Most settings pages never do.

Second, the training toggle only faces forward. Turning off “Improve the model for everyone” keeps future conversations out of model training. It does not pull back chats already used. It does not clear your history. It does not touch your memory settings. People flip that switch expecting a cleanup. Nothing visibly changes, because nothing should. It was never a deletion tool.

Third, and this one surprised me: turning off ad personalization does not turn off ads. On the Free and Go plans, the only plans that show ads at all, switching personalized ads off just changes how ads get picked. You still see them, chosen from the context of your current conversation instead of your profile. The company’s line is blunt about it. The one hard boundary the company does draw: advertisers get none of the underlying material. No chats, no history, no memories, no personal details.

Getting Something Out of the ChatGPT Privacy Center

The hub is most useful as an inventory. Walk the seven topics from top to bottom and hit Manage on each one. In a few minutes you will have touched every one of your ChatGPT privacy settings: memory, ads, location, temporary chat, connected apps, multi-factor authentication, model improvement, export, deletion. Nobody does this for fun. Do it once.

A few stops matter more than others. Turn on multi-factor authentication if it is off. Open the connected-apps list and revoke the apps you authorized long ago and forgot about. Decide on the training toggle with open eyes: off means future chats stay out of training, and everything already there stays put.

For anything genuinely sensitive, use a temporary chat and do not save it afterward. Temporary chats stay out of history, do not build memory, and are not used for training. And remember the warning above: the thumbs-up button feeds training too. On sensitive threads, leave it alone.

If what you want is deletion rather than orientation, the full walkthrough lives in our guide to deleting your ChatGPT data, which covers the three separate systems you have to clear. And if you are wondering whether the grass is more private elsewhere: Anthropic has confirmed it also uses human review to improve Claude. Our Claude vs ChatGPT vs Gemini comparison goes deeper on how the three differ.

The ChatGPT Privacy Center will not make any of those decisions for you. It just finally puts them all in one place where you can see them. A smaller promise than a privacy revolution, and a more useful one than another buried settings menu.

FAQ