Skip to main content
    Back to Blog
    14 min readLast reviewed:
    ChatGPT
    OpenAI
    Privacy
    Data Retention
    AI Privacy

    Does ChatGPT Save Your Prompts and Chats? Retention by Plan (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: Yes, ChatGPT saves your prompts — though which prompts, for how long, and for what purpose depends on the plan you're on, the settings you've toggled, and whether the conversation is a Temporary Chat. Even on the most privacy-friendly configuration (Temporary Chat with the training opt-out), OpenAI keeps a copy for up to 30 days for abuse monitoring before deletion. On the API with Zero Data Retention, retention can be effectively zero. This guide walks through exactly what OpenAI keeps, on which plan, for how long, and what you can do about it.

    The two questions hiding inside "save"

    When people ask "does ChatGPT save my prompts?" they usually mean one of two things:

    1. Does the conversation appear in my history later? This is a UI question. It's controlled by whether you use a Temporary Chat and whether you delete the conversation; there is no longer a separate chat-history switch.
    2. Does OpenAI store the prompt on its servers, and for how long? This is the privacy question that actually matters. It's controlled by your plan, your Data Controls, and OpenAI's standard abuse-monitoring policy.

    The two settings are related but not identical. You can have a chat that doesn't appear in your history but is still retained on OpenAI's servers for up to 30 days. You can have a chat that is in your history but is excluded from training. They are separate switches.

    What OpenAI actually retains, by plan (reviewed November 2026)

    PlanConversation historyTraining on your promptsStandard server-side retentionZero retention available?
    ChatGPT FreeVisible to you, retained per Data ControlsYes, unless you opt outUp to 30 days for abuse monitoring after deletionNo
    ChatGPT PlusVisible to you, retained per Data ControlsYes, unless you opt outUp to 30 days for abuse monitoring after deletionNo
    ChatGPT ProVisible to you, retained per Data ControlsYes, unless you opt out (same as Free and Plus)Up to 30 days for abuse monitoring after deletionNo
    ChatGPT Business (formerly Team)Visible to user; admins can view, export and delete conversationsNoKept until deleted; deleted chats removed within 30 days (custom retention is an Enterprise and Edu control)No
    ChatGPT Enterprise / Edu / HealthcareVisible to user; admins have workspace accessNoOwner-set retention (minimum 90 days on Enterprise); BAA on healthcare and regulated workspacesNo — zero data retention is an API arrangement
    OpenAI API (default)API customer controls retention on their sideNo (API excluded from training by default)Up to 30 days for abuse monitoringYes (subject to OpenAI approval)
    OpenAI API + Zero Data Retentionn/aNoEffectively zero (no logging beyond the request itself)Active
    Temporary Chat (any tier)Not stored in chat historyNoUp to 30 days for safety purposes; reviewed only for abusen/a

    A few things follow:

    • "Off" doesn't mean "never recorded." Turning off "Improve the model for everyone" stops your prompts from being used for training while keeping them in your history; a Temporary Chat keeps them out of your history too. Neither stops OpenAI from holding a copy for abuse monitoring (currently up to 30 days).
    • Temporary Chat is the closest thing to ephemeral, but the ~30-day abuse-monitoring window still applies on free and consumer-paid tiers.
    • Zero Data Retention is contractual, not a checkbox. It is an API arrangement subject to OpenAI's prior approval; ChatGPT Enterprise offers admin-controlled retention and a BAA rather than ZDR.

    Why OpenAI keeps prompts at all

    OpenAI's documentation gives three kinds of reason for retention:

    1. Abuse monitoring. Detecting use of the API for prohibited purposes (CSAM, weapons proliferation, mass-disinformation campaigns, illegal financial conduct, etc.). The 30-day window exists for this purpose even on accounts that have disabled training or use Temporary Chats.
    2. Trust & safety operations. Investigating incidents, responding to law enforcement requests, complying with legal hold and discovery orders. This is not hypothetical: from May 13 to September 26, 2025, a court order in the New York Times litigation required OpenAI to keep deleted and Temporary chats from Free, Plus, Pro, and Team accounts and from non-ZDR API customers; OpenAI says it still holds that April–September 2025 data under legal hold.
    3. Product improvement. On consumer Free / Plus / Pro accounts, prompts can be sampled (subject to your training opt-in/out and OpenAI's policy) for model and product improvement.

    Reason 1 is non-negotiable on most tiers; reasons 2 and 3 are governed by your settings and contract.

    What "deleted" actually means

    When you click delete in ChatGPT:

    • The conversation is removed from your visible history immediately.
    • Internally, OpenAI's documented policy is to remove deleted conversations from their systems within 30 days, except where retention is required for abuse monitoring or legal compliance.
    • In practice, "deletion" on a SaaS service is typically a marker for purge from primary storage, with backups and logs aging out on their own schedule. OpenAI's approach is consistent with industry norms; if you need cryptographic guarantees of unrecoverability, no major consumer AI service offers them today.

    If you have a specific compliance reason to require provable deletion, the right product is the API with Zero Data Retention, subject to OpenAI's approval — not a consumer plan. ChatGPT Enterprise offers admin-controlled retention and a BAA, not ZDR.

    What about the prompts your employer can read?

    Even on a personal account, your prompts may be visible to:

    • Your employer, if you used a managed device, a corporate VPN, or a company-managed account. (See Can My Employer See My ChatGPT Prompts? for the full breakdown.)
    • Other workspace members and admins, on a Business or Enterprise plan, depending on configuration and admin permissions.
    • Plugin or connector vendors, if you used a third-party connector — those vendors see the slice of the conversation routed to their tool.
    • Anyone with whom you shared a conversation link. OpenAI's shared links FAQ says shared-link pages are not intended for search-engine indexing, but that this does not make a link private; deleting the conversation removes its link.

    "Saved by OpenAI" is just one of several places your prompts might persist.

    Settings you can change today

    Five minutes of settings changes, ranked by impact:

    1. ChatGPT Data Controls

    Settings → Data Controls. Two switches matter:

    • Improve the model for everyone — Off opts your prompts out of training. Recommended for any workload involving anything you wouldn't want to read on a job interview transcript. Where every other major assistant sits on that default is in Which AI Tools Train On Your Data by Default?.
    • Memory (Personalization → Memory) — Review what ChatGPT has stored about you. Delete anything sensitive. Decide whether to keep memory on. Memory is a separate retention surface from chat history.

    2. Temporary Chat

    The chat picker has a Temporary Chat toggle. Use it for one-off prompts you don't want to keep in your history. Remember the abuse-monitoring window still applies; Temporary Chat is "do not store in history," not "never transmitted." ChatGPT Temporary Chat: What It Actually Saves covers the mode end to end.

    3. Sharing

    If you've ever clicked "Share" on a conversation, anyone with the link can view it; OpenAI says the pages are not intended for indexing, but that does not make the link private. Periodically review and revoke share links you no longer need; deleting the conversation also removes its link.

    4. Custom GPTs and Connectors

    If you've connected ChatGPT to Gmail, Google Drive, GitHub, Notion, or a custom Action, the connector sees the parts of the conversation it's invoked for. Audit your connectors regularly.

    5. Plan choice

    If you handle confidential or regulated data, the privacy gap between Plus / Pro and Enterprise / API+ZDR is much larger than the price gap. For regulated workflows, the consumer tiers are usually the wrong tool.

    When settings aren't enough: keep regulated data out of the prompt

    The strongest privacy guarantee is the data you never send. For regulated data — PHI, PII, financial NPI, source code with embedded secrets, attorney work product — the right control is:

    • Don't paste it. Use placeholders. AI is usually as good at "rewrite this email for [Patient]" as it is at the original.
    • Mask on the device. Locke 1.0 runs on your device as a desktop app; its companion browser extension screens AI websites such as ChatGPT, and a local proxy screens command-line, script, and SDK traffic. It detects sensitive entities — names, account numbers, health terms, secrets — before the prompt leaves your device, and replaces them with one-way placeholders. The model sees a usable prompt; the detected values do not reach OpenAI. Because the masking happens locally, the unmasked content is not sent to Sonomos either; only settings and licence data leave the device. Coverage depends on the app, platform, and settings. Reversible tokenization is planned shortly after 1.0.

    Whatever OpenAI saves on its servers can only contain what was sent. If the regulated data never went, retention becomes a non-issue for that data.

    Frequently asked questions

    Does ChatGPT save my prompts if I delete them?

    OpenAI removes deleted conversations from systems generally within 30 days, except where retention is required for abuse monitoring or legal compliance. The conversation is removed from your visible history immediately. Backups and logs age out on their own schedule. For provable deletion, the documented route is the API with Zero Data Retention, subject to OpenAI's approval.

    Are Temporary Chats saved?

    Temporary Chats are not stored in your chat history and are not used for training. A copy may still be kept for up to 30 days for safety purposes and reviewed only to monitor for abuse. They are best understood as "do not store in history," not "do not transmit."

    Does ChatGPT use my prompts for training?

    It depends on the plan. Free, Plus, and Pro may use your prompts for training unless you opt out in Data Controls. Business (formerly Team), Enterprise, Edu, and the API are excluded from training by default. Always verify in Data Controls — the defaults have changed several times.

    Can OpenAI see my prompts even if I have history off?

    OpenAI's systems handle the prompt to generate the response. There is no longer a separate chat-history switch. With "Improve the model for everyone" off, a saved chat stays in your history until you delete it and is not used for training; a Temporary Chat stays out of your history, and a copy is kept for up to 30 days and reviewed only for abuse. "Not visible to OpenAI staff" is closer to true on the API with ZDR; on consumer plans it is a continuum, not a binary.

    Can my prompts be subpoenaed?

    Yes. OpenAI complies with valid legal process. Stored prompts — including conversation history, files, and account data — can be produced in response to subpoenas, court orders, and law-enforcement requests. The retention window determines what is producible. For what that means for a company once litigation is foreseeable, see why employees' AI chats belong on a litigation hold.

    Are my prompts encrypted?

    Prompts are encrypted in transit (TLS) and at rest (AES). OpenAI staff with appropriate authorization can access the unencrypted content for the documented purposes. End-to-end encryption (where only you can read the content) is not the model — by definition, the model server has to read your prompt to respond to it.

    What about ChatGPT memory?

    Memory is a separate retention surface from chat history. Items in Memory persist across chats until you delete them. Treat Memory like a journal — anything you don't want stored should not be stored.

    Does this apply to Claude and Gemini too?

    The shape is similar — short retention windows on enterprise tiers, longer on consumer, training-on-prompts varies by tier — but the specifics differ. Anthropic now asks Claude Free, Pro, and Max users whether their chats may be used for training, and keeps them for up to five years if they say yes; Claude for Work and the API are excluded. Google's free Gemini may be reviewed by humans. Read each provider's privacy page; don't assume the policies are the same.

    How does Locke help?

    Locke doesn't change what OpenAI stores; it changes what reaches OpenAI in the first place. By detecting sensitive entities in your prompt and replacing them with one-way placeholders on your device before submission, Locke 1.0 keeps the detected values out of what reaches OpenAI. Coverage depends on the app, platform, and settings. Whatever OpenAI's retention policy is, it can't retain what you didn't send.

    The bottom line

    ChatGPT saves your prompts — but how much, for how long, and to what end depends entirely on the plan and the settings. On the consumer tiers in 2026, assume: visible in your history unless you use a Temporary Chat, held for up to 30 days for abuse monitoring even then, and used for training unless you opt out. On the API with ZDR, assume the contractual minimum. For anything you genuinely cannot afford to have stored, don't put the original value in the prompt — mask it on your device first.

    Sources and primary references

    Reviewed November 2, 2026; the plan retention rows were rechecked on November 18, 2026 against OpenAI's plan comparison and admin guide below. Regulations are cited to their official text. Vendor statements are cited to the vendor's own documentation as it read on that date; vendors revise these pages, so confirm before relying on them. The court filing is cited to a hosted copy of the stipulation and order.

    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