Skip to content

OpenAI (ChatGPT)

The maker of ChatGPT — the place a lot of EA/AIS/nonprofit work gets drafted, not a host for anything you publicly share on this wiki. Nothing here ships a file to a recipient; the questions below are all about the private chats you type, not a thing you hand out. For almost everything you'd do here, the honest answer is fine.

The one knob worth knowing: on a personal plan (Free, Plus, Pro), your chats train future OpenAI models unless you switch that off — a single toggle, named below. On a business plan (Business, Enterprise, Edu, API) that's already the default, no toggle needed.

Last verified: 2026-06-07 · Confidence: high on the training toggle, the business no-training default, retention, and the residency split (all from OpenAI's own docs); the EEA/UK consumer-default nuance is the soft spot.

Does it train AI on what you type?

The answer splits cleanly on plan type.

  • Personal plans (Free, Plus, Pro): yes, by default — but you can opt out. OpenAI's privacy policy says "we may use Content you provide us to improve our Services, for example to train the models that power ChatGPT." [confirmed] (privacy policy)
  • Business plans (Business, Enterprise, Edu, API): no, by default. "We do not train our models on your data by default" — and there's no toggle to police, it just doesn't happen. [confirmed] (enterprise privacy)

Turn the personal-account training off

One setting, ~30 sec — personal plans only (business plans already don't train):

  1. Your Profile → Settings → Data Controls → Improve the model for everyone.
  2. Switch the toggle off.

"Once you opt out, new conversations will not be used to train our models," and your chat history stays put. The setting syncs across web and mobile. [confirmed] (keep history, disable training)

Don't want to flip a global toggle? A Temporary Chat is never used for training and is deleted after 30 days — handy for a one-off sensitive question. [confirmed] (data controls FAQ)

Retention and deletion

  • Delete a conversation and it's gone within ~30 days. Deleted chats are "removed from our systems within 30 days, unless we are legally required to retain them." [confirmed] (enterprise privacy)
  • Opting out isn't retroactive. Switching training off only stops future conversations from being used — anything already folded into a finished training run stays in. [estimate] (standard for this kind of pipeline; OpenAI's docs don't spell out the past-runs case)
  • Temporary Chats auto-delete after 30 days, aren't trained on, and may be reviewed only for abuse. [confirmed] (data controls FAQ)

What a business/Enterprise tier changes

If your work touches regulated data (health, personal, anything under a grant's data terms), a paid business plan is the tier that earns what a compliance review wants:

  • No training, by default — across ChatGPT Business, Enterprise, Edu, ChatGPT for Healthcare, and the API. [confirmed]
  • You own your inputs and outputs, and on Enterprise/Edu/Healthcare your admins control how long data is retained. [confirmed] (enterprise privacy)
  • A real data agreement (DPA). OpenAI signs a Data Processing Addendum for Business, Enterprise, and API use "in support of their compliance with GDPR and other privacy laws," plus a completed SOC 2 audit and AES-256 encryption at rest. [confirmed]

For an individual drafting a doc or a workshop outline, a free account with the toggle off is genuinely fine — the business extras are for orgs handling regulated data.

Where your data lives (matters under GDPR)

  • Personal plans: no region choice, US-hosted. There's no setting on a Free/Plus/Pro account to pin your data to the EU or UK. Under GDPR that's a US transfer — usually fine for non-sensitive drafting, but name it if a grant or DPA restricts where data may sit. [confirmed] (no consumer residency option in OpenAI's docs)
  • Choosing Europe/UK is a business feature. Data residency lets eligible Enterprise, Edu, and API customers keep customer content stored at rest in a chosen region — Europe, the UK, the US, and others. It's not self-serve on a personal account, and it covers storage, not inference (model execution still defaults to the US unless inference residency is enabled). [confirmed] (data residency)

Bottom line for sharing

Nothing you do through ChatGPT lands on a recipient via this wiki, so the data hygiene above is about your own private workspace. On a personal plan, one toggle (Improve the model for everyone → off) settles the training question; a Temporary Chat handles the occasional sensitive one-off.

Sources