Skip to main content
    Back to Blog
    10 min readLast reviewed:
    ChatGPT
    OpenAI
    Privacy
    AI Privacy
    Pricing

    Is Free ChatGPT Private? Free vs Paid ChatGPT Privacy (2026)

    T

    Team Sonomos · Sonomos

    The Sonomos research team writes about AI privacy, data protection, and how to use generative AI safely at work.

    Short answer: Paying for ChatGPT changes some privacy defaults but not all of them. The biggest privacy gap in 2026 is not Free vs. Plus — it's consumer tiers (Free, Plus, Pro, personal Team) vs. contracted tiers (Enterprise, Edu, Team-with-admin-controls, API with Zero Data Retention). Free, Plus and Pro are all trained on by default unless you opt out in Data Controls; Business (formerly Team), Enterprise, Edu and the API are excluded by default and come with a DPA, and a BAA is available for eligible healthcare workloads on the enterprise products and the API. This guide breaks down exactly what changes — and what doesn't — across every ChatGPT tier in 2026.

    The four privacy levers, across all ChatGPT tiers

    There are really only four levers that matter for ChatGPT privacy. Every other policy boils down to a configuration of these:

    1. Training. Does OpenAI use your prompts to improve future models?
    2. Retention. How long does OpenAI keep your prompts on its servers, and for what purpose?
    3. Visibility. Who can see your conversations — you, OpenAI staff under abuse review, your workspace admin, the public?
    4. Compulsion. Under what legal process can your prompts be produced?

    The right way to compare tiers is to ask how each one positions all four levers.

    Tier-by-tier breakdown (checked November 2026)

    TierTraining on your promptsRetentionAdmin / workspace visibilityBAA / DPA availableZDR available
    ChatGPT FreeYes (opt-out in Data Controls)History per user setting; up to 30 days for abuse monitoring even when offNoneNoNo
    ChatGPT PlusYes (opt-out in Data Controls)Same as FreeNoneNoNo
    ChatGPT Pro (consumer)Yes (opt-out in Data Controls)Same as FreeNoneNoNo
    ChatGPT Business (formerly Team)No (excluded by default)Kept until deleted; deleted chats removed within 30 daysWorkspace adminsDPA; no BAA documentedNo
    ChatGPT EnterpriseNo (excluded by default)Owner-set retention (minimum 90 days)Workspace admins, Compliance API, audit logsDPA; BAA on healthcare and regulated workspacesNo (ZDR is an API arrangement)
    ChatGPT EduNo (excluded by default)Owner-set retentionInstitution admins, audit logsYes (per institution)No
    OpenAI API (default)No (excluded by default)30 days for abuse monitoringn/aDPA; BAA on requestYes, with OpenAI's approval
    OpenAI API + ZDRNoEffectively zeron/aYesActive
    Temporary Chat (any tier)NoUp to 30 days for abuse monitoringn/an/an/a

    A few things to notice:

    • The big gap isn't between free and paid personal plans. Free, Plus and Pro share the same training default and the same lack of a contract. The real gap is personal plan → Business, Enterprise or API, where training is off by default and you get real contracts, BAA / DPA coverage, and admin tooling.
    • Plus does not buy you a stronger contract. Plus is a consumer subscription. There is no DPA, BAA, or admin tenancy. Faster access and more features, yes — fundamentally different privacy posture, no.
    • Pro does not change the training default. OpenAI's Data Controls help page describes "Improve the model for everyone" as on by default for personal Free, Plus and Pro accounts, so on every personal plan the fix is the same switch. Verify in Data Controls; defaults have moved more than once.

    What actually improves when you upgrade — by jump

    Free → Plus

    • Better access to GPT-class models, longer context, more features (browsing, image generation, voice).
    • Same privacy defaults: trained on by default unless opted out, same retention, same lack of contract.
    • What you should change either way: Settings → Data Controls → turn off "Improve the model for everyone." Settings → Personalization → review or disable Memory.

    Plus → Pro (consumer)

    • Training default stays the same: Pro is trained on by default unless you opt out, like Free and Plus. (For the same line drawn across every other vendor, see Which AI Tools Train On Your Data by Default?.)
    • Larger usage limits, deeper-reasoning models, agent features.
    • No change in retention or contract. Still no DPA, BAA, or admin tenancy.

    Plus / Pro → Business (formerly Team)

    • Workspace tenancy: your account is part of an organization, not a personal subscription.
    • Excluded from training by default.
    • Admin visibility: workspace admins can see usage, manage members, and access workspace data — different privacy model than a personal account.
    • DPA available for Team customers; BAA generally requires Enterprise.
    • No custom retention setting: chats are kept until they are deleted. OpenAI's plan comparison lists retention controls only for Enterprise and Edu.

    Team → Enterprise

    • All the Team controls, plus:
    • BAA for HIPAA-eligible workloads.
    • Custom data retention: workspace owners can set a retention period, with a 90-day minimum according to OpenAI's admin guide. Zero Data Retention is an API arrangement, not a ChatGPT Enterprise setting.
    • Mature admin / audit-log surface, including SSO, SCIM, audit APIs, advanced compliance configuration.
    • Connector and GPT governance at workspace scope.
    • Procurement-friendly contract with your enterprise legal terms.

    Any tier → API with ZDR

    • Strongest baseline: no training, effectively zero retention, contractually documented.
    • Trades the polished UI for code-level control. Used directly by application backends; humans don't typically chat with the API.

    The rest of the privacy picture, by tier

    Even with the above table memorized, four practical issues drive most of the leak risk in 2026 — independent of which tier you're on.

    1. Memory

    Memory persists items about you across chats until you delete them. On any consumer tier, only you (and OpenAI) can see Memory. On Team / Enterprise, workspace admins may have access to a member's workspace data. If anything in your Memory is sensitive, consider clearing it or disabling Memory entirely; it's easy to forget what's in there.

    2. Connectors, GPTs, and Plugins

    When ChatGPT calls a connector — Gmail, Drive, GitHub, a custom action — the connector vendor sees the slice of conversation routed to it. This is independent of your ChatGPT tier and is governed by the connector's own terms. Audit your connectors regularly.

    3. Sharing

    Shareable conversation URLs were briefly indexed by search engines in 2025 before defaults changed. Verify the current sharing model and revoke old share links you don't need.

    4. Account-level vs. device-level monitoring

    Your tier governs OpenAI's view of your prompts. It does not govern your employer's view, if you're on a managed device or network. See Can My Employer See My ChatGPT Prompts? for the device / network angle.

    What "upgrade for privacy" actually buys you

    If your concern is "I don't want my prompts used for training":

    • Cheapest fix: opt out in Data Controls (Free / Plus). Free.
    • Upgrading does not fix it: Plus and Pro have the same default as Free. Business, Enterprise, Edu and the API are excluded by default.

    If your concern is "I want a contract":

    • Minimum: Team (DPA available).
    • Standard for regulated work: Enterprise (custom retention, audit logs, and a BAA on OpenAI's healthcare and regulated workspaces).

    If your concern is "I want as little retained as possible":

    • Best practical option: the API with ZDR, which needs OpenAI's approval, or Enterprise with the shortest retention period your record-keeping duties allow.
    • On consumer plans: Temporary Chat reduces history retention but the abuse-monitoring window still applies — see ChatGPT Temporary Chat: What It Actually Saves.

    If your concern is "I work with regulated data and an employer":

    • Either sign up for ChatGPT Enterprise + BAA via the company,
    • or stop using ChatGPT for that data category and route the workload through a sanctioned tool.
    • In both cases: layer a local-first redaction tool on your device so the regulated entities never leave it, regardless of what tier you're on. Locke sits in this layer.

    Frequently asked questions

    Is ChatGPT Plus more private than Free?

    Marginally. The defaults are the same on both tiers (training-by-default with opt-out, same retention, no contract). Plus buys you more capability, not more privacy.

    Is ChatGPT Pro private enough for confidential work?

    Not on its own. Pro is trained on by default unless you turn off "Improve the model for everyone", and even with that off it is a personal subscription with the same ~30-day abuse-monitoring window, no DPA, and no BAA. For confidential work involving regulated data, you want a contracted tier (Enterprise or Edu under a BAA, or the API), not Pro.

    Is ChatGPT Team enough for HIPAA?

    In general, no. Business (formerly Team) has DPA support and excluded-from-training defaults, but OpenAI's HIPAA guide names ChatGPT for Healthcare, ChatGPT for Clinicians, Regulated workspaces and API services with modified retention as what its BAA covers. For PHI, use one of those under a signed BAA.

    Does ChatGPT Enterprise see my prompts?

    OpenAI's systems handle the prompt to generate the response (this is true on every tier). Enterprise excludes training by default and lets workspace owners shorten retention, though not to zero. Workspace admins — your employer — can have access to workspace conversations, particularly for legal hold or investigation. Enterprise is private from OpenAI in a contractual sense; it is not private from your admin.

    Are my prompts encrypted on Plus?

    Encrypted in transit (TLS) and at rest (AES) on every tier. End-to-end encryption (where only you can read your prompts) is not the model on any consumer AI service — the server has to read the prompt to generate the response.

    Is the API more private than ChatGPT Plus?

    Yes, by default. The API excludes prompts from training by default, and ZDR is available for eligible endpoints with OpenAI's prior approval. The API is the right path when you need the strongest contractual posture and don't need the polished chat UI.

    How do I know which tier I'm signed into?

    Open Settings → Data Controls. The visible options reflect your tier. Enterprise / Edu / Team users will see workspace controls; consumer users will see personal Data Controls. If you're not sure whether you're on a personal or work account, check the email address on your ChatGPT profile against your work directory.

    How does Locke relate to tier choice?

    Locke doesn't change your ChatGPT tier — it changes what reaches ChatGPT in the first place. By detecting sensitive entities on your device and replacing them with one-way placeholders before submission, Locke keeps the regulated parts of your prompt from leaving it. That's useful on every tier: it makes Free / Plus less risky for the non-regulated parts of your work, and it makes Enterprise or API + ZDR a defense-in-depth setup rather than a single point of trust.

    The bottom line

    Upgrading from Free to Plus does very little for privacy. Upgrading to Pro does not change the training default either. Upgrading to Enterprise (or going API + ZDR + BAA) is the inflection point that gets you a real contract, real retention controls, real audit logs, and real BAA / DPA coverage. For the moments when even that is too much surface area to trust — privileged communication, PHI, NPI, source-code secrets — keep the data out of the prompt entirely with a local-first redaction layer. The cheapest privacy upgrade is the data you never send.

    Sonomos publishes this blog and makes Locke, which detects and masks sensitive information on your device before it reaches an AI model. How Locke works

    This article is general information about privacy and AI, not legal advice, and reading it does not create a solicitor–client or attorney–client relationship. Regulatory requirements turn on your specific facts and jurisdiction and change over time. Consult qualified counsel before relying on anything here, and check the publication and last-reviewed dates above. We may revise posts, but we do not undertake to keep any post current as the law changes.

    Reference pages for this post