Data handling differs by plan: ChatGPT, Claude, and Gemini

A tech on a Friday rewrite provides the whole thread, gate code included, into whichever free app is on their phone. The company "has a policy." Nobody has seen it.

The short answer

The sequence below is the control path: same vendor pages, same prompts, product names kept exact.

Business data is only as safe as what staff provide, which plan they are on, and what that vendor's current page says. No chat box makes a careless provide safe. Write a red list, use the team or Workspace plan for work files, and read the vendor's privacy help before you set the rule. This page is not a compliance certificate.

Why a logo is not a control

Consumer plans and team plans do not promise the same thing about training. A personal login and a work login do not either. The leak that shows up in a small shop is shadow use: the approved seat sits idle while the personal app gets the real email.

ChatGPT

OpenAI's help for Projects and GPTs says that on Business, Enterprise, and Edu, data is not used for training by default. Re-check the current Enterprise privacy page before you set shop policy. Consumer plans (Free, Plus, Pro, and whatever the pricing page lists now) are a different setting. Read openai.com/pricing and the privacy policy for the account you actually use. Do not treat a sentence on this blog as the setting.

Claude

Consumer (Free, Pro, Max). Anthropic's privacy center explains when chats may be used to improve models if you allow model improvement, with separate rules for safety review and feedback. Read Is my data used for model training?.

Team and Enterprise. Anthropic says commercial products are not used to train models by default, unless you opt into a program such as the Development Partner Program. Read processor or controller. Your own admin may still be able to see workplace chats. That is your policy, not a surprise to discover later.

Incognito or similar modes are described in Claude's help. Read the current article. Do not assume a toggle you remember.

Gemini

Personal Gemini Apps, including free and paid Google AI plans. The Gemini Apps Privacy Hub covers gemini.google.com, the mobile app, and the consumer surfaces it lists. It says Google uses collected data to provide and improve services, including the models behind Gemini Apps, and that human reviewers can see a subset. The notice tells people not to submit confidential information they would not want reviewed or used to improve those systems. Activity settings, including whether activity is kept, are in that hub. Read them.

Google Workspace. A work account is a different door. Admin controls and the Workspace agreement apply. Do not connect a company Drive from a personal Gemini login and call it the Workspace setting.

What not to provide, in any of them

Same postcard rule as customer data and AI tools:

• Card numbers, bank logins, passwords
• Full medical charts or health details the draft does not need
• Children's data, government IDs, unrestricted HR files
• Gate codes, alarm codes, and anything you would not put in a shared email

Prefer a role plus the facts the draft needs. Redact names when the job does not need them. Connectors and apps should be least access. See Drive and Microsoft 365.

Law firms collecting this kind of detail should use a secure intake form instead; here's an estate planning questionnaire template built as a Clio Grow intake form.

A one-page rule you can post

1. Red: never provide. The list above.
2. Yellow: only on the team or Workspace account, with an admin who knows.
3. Green: approved templates, public pages, anonymized examples.
4. Name who may connect Drive or Microsoft 365.
5. Pin it where the crew will see it. Reread the vendor pages quarterly. They change.

For the Memory on/off and Temporary Chat checklist specifically, see turn off ChatGPT memory for business.

What this article is not

It is not a HIPAA, SOC 2, or legal opinion. A vendor may list enterprise options, including HIPAA-ready offerings on some Enterprise paths. That is their product surface, not automatic coverage for your clinic. Confirm the contract with the vendor and with counsel.

Example (illustrative): a vet clinic on a Saturday

Problem: the on-call tech provides a client thread into a personal app to "make it shorter." What was built: a red list on the wall and one approved login. Tools it lives in: the team plan you already pay for, not a new logo. Owner still does: the send. Payoff: not measured.

Mistakes and limits

• A personal free account as the easy path.
• Assuming "Business" or "Team" on an invoice means every toggle is off. Read the page.
• Uploading a whole chart to summarize one sentence.
• Treating this article as the policy a regulator will accept.
• Connecting write tools (send mail, create files) before the red list exists.

Doing this inside the tools you already use

If you would rather have this exact job drafted from the software you already use, with you still approving every send, see how a custom build works.

FAQ

Is ChatGPT Business "safe" and Free not?

Business, Enterprise, and Edu are described by OpenAI as not used for training by default. Consumer plans are a different control. "Safe" still depends on what was provided. Read the current help for the plan on the invoice.

Does Claude Team train on our chats?

Anthropic's commercial article says chats are not used to train models by default unless you opt in. Your organization may still access workplace conversations under its own admin policy. Re-read the privacy center before you write the rule.

Is a personal Gemini account the same as Workspace?

No. The Privacy Hub covers consumer Gemini Apps. Workspace is an admin-controlled work account. Connecting company files from a personal login follows the consumer notice.

Can we provide a first name?

If the draft needs it and the rest of the record stays out, that is a yellow decision for your shop. Full charts, IDs, and payment data stay red.

Keep these

• Red, yellow, green. Post it.
• Team or Workspace for work files. Personal apps are not a side door.
• Re-read the vendor privacy page before you set the rule.
• This article is hygiene, not a certificate.

Sources

• OpenAI: Projects and training defaults
• Anthropic: is my data used for training?
• Anthropic: processor or controller
• Gemini Apps Privacy Hub

Related reading

• Customer data and the tools staff already open
• Which plan matches the sensitivity
• Connect Drive or Microsoft 365 without opening everything