Is ChatGPT confidential? What OpenAI publishes, plan by plan

Whether ChatGPT is confidential enough for your work depends on your plan, your settings and your firm's policy. Here is what OpenAI publishes about what it keeps from a chat, what it may train on and who else can see it, quoted from its own pages and dated, so you can decide what to put in.

You have a client's draft agreement open in one window and ChatGPT in the other, and the question is whether the two should meet. This guide doesn't answer that for you. Nobody outside your firm can: it turns on your plan, your settings, your firm's policy and, for some professions, rules this page does not interpret. What it does is set out what OpenAI itself publishes, in OpenAI's words, so the decision rests on the terms rather than on a forum thread.

Equerry, which publishes this guide, makes a personal assistant for client work, and it comes up only near the end. Every OpenAI statement below was read on its own help center or website on 6 November 2026. Nothing here is legal or compliance advice.

Is ChatGPT safe for confidential information? What OpenAI's pages say

OpenAI's pages do not use the word "safe" about what you paste, and this guide will not either. What they do is state terms. For a personal account (Free, Go, Plus, Pro), the consumer data article, read 6 November 2026: "A limited number of authorized OpenAI personnel, as well as trusted service providers that are subject to confidentiality and security obligations, may access user content only as needed for these reasons" (the four reasons are quoted in full further down) and, in the same section: "Please do not enter sensitive information that you would not want reviewed or used." The memory article, read the same day, adds: "Sensitive information can appear in Memory if you share it with ChatGPT." Training on a personal account is on by default, with one switch to turn it off; the switch and its exceptions are in the training section below.

A business workspace starts from a different sentence. OpenAI's Enterprise privacy page (marked "Updated: January 8, 2026", read 6 November 2026), on ChatGPT Business: "Workspace admins have control over workspaces and can view, access, export, and delete end user conversations in the workspace." Training is off by default there; the table below has OpenAI's wording. So "who can see it" has a different answer on each plan, and on neither plan is the answer nobody.

One documented exception most readers will never meet shows why this page quotes terms rather than summarising them: the data controls article (read 6 November 2026) says of Trusted Access programs "like Daybreak Blue" that "During Daybreak sessions on your personal account, we may use content from your interactions to improve our models even when “Improve the model for everyone” is off."

So the usable form of "is it safe for confidential information" is a short checklist: which plan the account is on, whether Improve the model for everyone is off, whether the chat is temporary, who administers the workspace, and what your firm's policy and the client's engagement terms allow. The sections below take each in turn.

What not to share with ChatGPT, as a starting point: OpenAI's own instruction is the sentence above about sensitive information, and its sharing article draws the same line for shared links (quoted lower down). Beyond that the list is your firm's; where its policy or a client's terms keep a category out, that settles it whatever the plan says. The section on what people at work choose not to paste, lower on this page, lists where professionals who handle other people's information commonly draw the line.

What decides the answer: your plan, your settings, your firm

OpenAI's data controls article puts it plainly: "Available controls depend on whether you are signed in, your plan, and your workspace settings." Three things to check before anything client-related goes in:

• Your plan. Personal plans (Free, Go, Plus, Pro) and business workspaces (Business, Enterprise, Edu, ChatGPT for Healthcare) start from different defaults on training, retention and who else can see a chat.
• Your settings. On a personal plan, one switch decides whether new conversations can train OpenAI's models, and the kind of chat you open decides what is kept.
• Your firm's policy. The managed-workspaces section of OpenAI's data controls article says: "Your organization's privacy, retention, and access policies still apply." A policy that says no settles it, whatever the plan.

What OpenAI keeps from a chat

From OpenAI's "Chat and file retention in ChatGPT" and "Memory in ChatGPT" articles, read 6 November 2026:

• Regular and archived chats stay until you delete them. They "remain saved to your account until you delete them or an applicable workspace retention policy removes them." Archiving "does not change its retention period."
• A deleted chat is scheduled for deletion, with exceptions. OpenAI "schedules it for permanent deletion from its systems within 30 days, unless the chat was already de-identified and disassociated from your account or OpenAI must retain it longer for security or legal obligations."
• A temporary chat is still kept for a while. "OpenAI may keep a copy for up to 30 days for safety purposes." If you save it, it becomes a regular chat. Our guide to ChatGPT Temporary Chat covers how to start one and what it can still read.
• Files can outlive the chat. "Deleting a chat does not delete a separately saved file. Delete the file from Library or Space if you also want to remove that saved copy."
• Memory is separate again. "Deleting a chat alone does not necessarily delete a separate saved memory created from that chat," and "OpenAI may retain logs of deleted saved memories for up to 30 days for safety and debugging purposes."

So "I deleted it" can mean three different things: the chat, the file and what ChatGPT remembered from it. Our guide to whether ChatGPT remembers deleted chats takes each one in turn, and how to delete ChatGPT memory has the steps for the last.

What OpenAI may train on, and the switch that changes it

OpenAI's article "How your data is used to improve model performance" (read 6 November 2026) is direct about personal plans: "When you use our services for individuals (Free, Plus, Pro), we may use your content to train our models." The same article says: "Before using eligible user content to improve our models, we take steps to reduce the amount of personal information in training datasets." OpenAI's consumer data article gives examples of that content: "the user's prompts, the model's responses, and other content such as images and files." Its article on ChatGPT Go says the same in shorter form: "OpenAI may use conversations to improve model performance and safety. You can opt out of having your data used for training at any time."

The opt-out is one switch, Improve the model for everyone, or Do not train on my content in OpenAI's Privacy Portal; OpenAI says "Either option is sufficient." Three details from the same pages decide how far it reaches:

• It covers new conversations. "After you opt out, we won't use your new conversations to improve our models." And it "does not delete or hide saved chats."
• A rating reopens one conversation. Even after opting out, "If you choose to provide feedback, the entire conversation associated with that feedback may be used to train our models."
• Temporary chats are outside it. They "are not used to improve OpenAI models while they remain temporary, even if you choose personalization."

OpenAI's steps, from "Data controls in ChatGPT". On the web, signed in: open your account menu, select Settings, select Data controls, then select Improve the model for everyone, turn it off, and select Done. On iOS and Android: open the sidebar, select your profile icon to open Settings, select Data controls and turn off Improve the model for everyone. OpenAI adds: "When you're signed in, Improve the model for everyone applies to your account across web and mobile devices."

How it differs by plan, as of 6 November 2026

Each cell quotes or closely follows OpenAI's help center and its "Enterprise privacy at OpenAI" page (marked "Updated: January 8, 2026"), both read 6 November 2026. It is not a ranking.

The trade runs both ways. A business workspace changes the training default and puts retention in your firm's hands, and it also adds people who can read your chats. OpenAI also says it can sign a Data Processing Addendum for ChatGPT Business, ChatGPT Enterprise and the API. Whether any of this meets your obligations is for your firm to judge.

Who else can see a ChatGPT conversation

Five groups, each as OpenAI describes it (read 6 November 2026):

• OpenAI personnel and service providers, for listed reasons, on personal accounts. In OpenAI's consumer data article: "A limited number of authorized OpenAI personnel, as well as trusted service providers that are subject to confidentiality and security obligations, may access user content only as needed for these reasons: (1) investigating abuse or a security incident; (2) to provide support to you if you reach out to us with questions about your account, or to troubleshoot issues affecting the product experience; (3) to handle legal matters; or (4) to improve model performance (unless you have opted out)." In the same section: "Please do not enter sensitive information that you would not want reviewed or used." For business plans, OpenAI's Enterprise privacy page sets different terms. Business: "(1) authorized employees that require access for engineering support, investigating potential platform abuse, and legal compliance and (2) specialized third-party contractors who are bound by confidentiality and security obligations, solely to review for abuse and misuse." Enterprise, Edu and Healthcare: "Authorized OpenAI employees will only ever access your conversations for the purposes of resolving incidents, recovering end user conversations with your explicit permission, or where required by applicable law."
• Your workspace admins, on a business plan, as in the table above.
• Anyone with a shared link. "A link does not provide access to the creator's account, but anyone permitted to open it can view its shared content." On a personal account the link is a snapshot; on a workspace it "can include messages added after sharing."
• Not advertisers, by OpenAI's statement. "OpenAI does not sell your data or share your ChatGPT conversations with advertisers." The same article: "We do not share your chat content for marketing or advertising purposes." Ads themselves are a separate matter: OpenAI's ChatGPT Go article (read 6 November 2026) says "Ads are being tested with eligible ChatGPT Go users in select regions."
• The other direction: connected apps. Memory can draw on "content from connected apps, such as Gmail." That is not someone seeing ChatGPT; it is ChatGPT reading an account of yours, and a connected mailbox carries every client at once.

OpenAI's path to find and delete them: go to Settings, select Data controls, then for Shared links select Manage. From the list, Delete shared link removes one; More actions then Delete all shared links removes every one. Deleting a link "does not remove a conversation or task another person already saved."

What people at work choose not to paste

There is no universal list, and your firm's policy outranks any of this. But the professionals who handle other people's information (lawyers, advisors, accountants, consultants) tend to draw the line in similar places when a tool and plan have not been approved:

• Anything that identifies a client together with their matter: names, addresses, the deal or dispute they are in.
• Numbers that unlock something: account, policy and ID numbers, dates of birth, passwords.
• Privileged or regulated material: legal advice and the facts behind it, tax return information, health records.
• The unannounced: board papers, a deal before signing, figures before they are public.
• Whatever a client sent in confidence, including the attachment that came with the email.

What often goes in instead: public material you pulled yourself, drafts with placeholders where the names were, and work with no client in it at all. OpenAI's sharing article draws a similar line for shared links: "Do not include health information, financial details, passwords, account numbers, or other sensitive personal information in shared content." For profession-specific lists, see ChatGPT for lawyers, ChatGPT for legal work and ChatGPT for accountants.

What this page does not cover

It does not say whether using ChatGPT is consistent with privilege, a professional conduct rule, a client contract or a data-protection law; those are judgments for your firm and its advisers. It does not cover OpenAI's API, Codex, or OpenAI's contracts beyond what its public pages say, and features vary by region and account, as OpenAI notes throughout. It is not a security assessment of OpenAI, and it is not one of Equerry either.

A narrower question: one thread instead of a whole account

ChatGPT does remember. Depending on plan and region, its Memory draws on past chats, saved memories, saved files in Library or Space and connected apps, and that is where the decision gets wide: to have it know your client correspondence without pasting it in, you connect a mailbox, and with it every other client in that mailbox. What you can review afterwards is, in OpenAI's words, "a high-level view of information that ChatGPT may use to personalize responses."

Equerry is a personal assistant who helps you get things done, built around the other end of that trade. It reads only what you send it, in the ways listed below, plus your calendar, read-only, if you allow it. It never asks for your inbox. That makes the question for your firm a narrower one, about the specific things you chose to send, not an account. It is not a smaller question, and it is still yours to answer under your policy.

• Email, one thread at a time. CC Equerry's address on a live thread, or forward one after the fact. Replies that keep the address on the thread keep arriving. The original stays in your own mailbox. Everyone on a thread can see the address on the CC line.
• Files and notes from your iPhone. Share a PDF, an image, a link or text from the share sheet, type a note, or record a voice note inside Equerry.
• What it remembered, line by line. Settings → Your memory lists the facts it drew from what you sent, each marked with where it came from, and you can delete any line.

Equerry makes no claim about privilege, about any professional rule, or about whether what you send it is confidential. What you send is processed so it can be searched and answered from, including by the outside AI providers named on our AI disclosure page, which says what each one receives and states that we never train AI models on what you save. There are no ads in Equerry, and it sends nothing, books nothing and buys nothing on your behalf. Read that page the way this guide reads OpenAI's: as published terms, against your firm's policy.

Comparing assistants? Our guide to what Anthropic publishes about Claude asks the same questions of Anthropic's terms, set out plan by plan.

Sources, all read and verified 2026-11-06 as rendered pages. OpenAI Help Center: Data controls in ChatGPT (article 7730893); How your data is used to improve model performance (5722486); How OpenAI handles data in consumer services (7039943); What is ChatGPT Go? (11989085); Chat and file retention in ChatGPT (8983778); Temporary chat in ChatGPT (8914046); Memory in ChatGPT (8590148); Sharing conversations and scheduled tasks in ChatGPT (7925741). OpenAI, Enterprise privacy at OpenAI (marked "Updated: January 8, 2026"). All quoted wording is OpenAI's; where its pages are silent, this guide says so rather than filling the gap. OpenAI changes its products and pages often; where its current page disagrees with this one, OpenAI's page is right. This guide is general information, not legal or compliance advice.

Last updated 2026-11-06.

All product names, logos, and brands are property of their respective owners. Equerry is not affiliated with, endorsed by, or sponsored by any other product mentioned; comparisons are provided for informational purposes.