How ChatGPT Business protects your personal information

Most people running a business from a personal ChatGPT account are pasting more identifying information into it than they realize. Their own address on a contract draft. A client's full name and phone number in an email they are rewriting. Bank details inside a transaction export. An EIN on a form they asked for help filling in.

ChatGPT Business changes the default underneath all of that. Business, Enterprise, Edu, and the API do not use your data to train OpenAI's models. On free and Plus, conversations may be used for training unless you individually opt out. That is the single most important difference, and it is worth knowing precisely what it protects and what it leaves entirely up to you.

What is different about ChatGPT Business?

Business, formerly called Team, runs $25 per seat per month billed annually or $30 month to month. Four things change compared with a personal account.

Your inputs are excluded from model training by default, meaning you do not have to find and flip a setting. The account carries SOC 2 Type 2 certification covering security, availability, confidentiality, and privacy. It holds ISO/IEC 27001 certification, received in January 2026. And it supports GDPR and CCPA compliance with a Data Processing Addendum available, which matters if anyone ever asks you in writing how you handle their data.

Data retention is also administrable rather than fixed, so a workspace owner sets policy instead of accepting a default.

Why does the training difference matter for personal information?

Because training is the one thing you cannot take back.

Deleting a conversation removes it from your history. It does not remove whatever influence that text already had on a model that trained on it. Nobody is going to type your name into ChatGPT and get your bank details back, and that is not the realistic risk. The realistic risk is that information you had no right to hand over, meaning your client's private facts, went somewhere permanent and outside your control, and you agreed to it by using a consumer tier for business work.

On Business that path is closed by default. Your inputs stay out of training without you having to remember a setting.

What does it still not protect?

This is where people over-read the upgrade, so it is worth being blunt.

Your data still travels to and gets processed on OpenAI's servers. Business is safer, and it is not private in the sense of never leaving your machine.

Authorized third-party contractors may review data to investigate potential abuse. Human review remains possible.

And no plan protects you from your own paste. Certain categories stay out regardless of tier: passwords and API keys, customer payment and financial account numbers, government identifiers like Social Security numbers, health information, and anything under an NDA you signed.

The habit that handles all of it is redact-then-prompt, which AI privacy basics covers in full. Swap real names and numbers for stand-ins, let the model work on the structure, then put the real details back in your own document. It costs about ten seconds.

Which categories should never go in, on any plan?

Five, and the reason each one is different.

Credentials, meaning passwords, API keys, and access tokens. These grant access, and access is the thing you cannot un-share.

Payment and financial account numbers. Card numbers, bank account and routing numbers. Strip them from any export before pasting.

Government identifiers. Social Security numbers, tax IDs tied to an individual, passport and license numbers. These are the raw material of identity theft.

Other people's personal information. Client names paired with private facts, contact details, and anything they told you in confidence. This is theirs, not yours to hand over, and it is the category owners violate most casually.

Anything under an NDA. If you signed something saying you would not share it, a chatbot is sharing it.

Does the business tier make you compliant?

No, and this is where a Data Processing Addendum gets misread as a compliance certificate.

The DPA and the certifications mean OpenAI has documented how it handles data as your processor. Your obligations as the controller stay yours. If a client asks whether their data goes into AI tools, you still need an answer, a policy, and consistency with what you told them.

A short written policy handles this, and it does not need to be long. A responsible AI policy your team can copy has a one-page version covering disclosure, data, verification, approvals, and tools. If anyone besides you touches the account, that document is what keeps the protections from depending on memory.

When is the upgrade worth $25 a seat?

Three triggers, and any one of them is enough.

You handle other people's data. Client names, contact details, financial records, or health information anywhere in your work. The obligation is not really to yourself, it is to them.

Somebody else uses the account. Shared credentials on a personal plan means no admin controls, no retention policy, and no audit trail. Business gives you SSO and administrable retention.

A client has asked, or will ask, how you handle data. Being able to point at a business tier with a DPA and SOC 2 is a straightforward answer. "I use the regular version" is not.

If none of those apply, meaning you work alone on your own material with no client data involved, Plus at $20 with training opted out gets you most of the way. The cost comparison across tiers is in 8 ways ChatGPT saves money.

What should you set up on day one?

Turn on SSO if you have it available. Set the retention policy deliberately rather than leaving the default. Write the five no-go categories into a note everyone can see. Then check whether anyone on your team has been using a personal account for client work, because that history exists on a consumer tier and moving forward does not retroactively clean it.

For the wider question of which tasks should involve AI at all versus which need a person deciding, human-centered AI covers where to put the human in the loop on purpose.

The AI policy templates and the data-handling checklists inside the WorkSmart OS include the one-page version of all of this, written for a business small enough that you are the compliance department.