Does Claude Train on Your Data? What Your Team Needs to Know

Not on Claude Team, Enterprise or the API: Anthropic's commercial terms say it may not train on your content. On Free, Pro, Max and Ultra, it's a setting each person controls. Not training isn't the same as not keeping.

10 minute read

Published 12 November 2026

By Manie Eksteen

Does Claude train on your data: an open laptop seen from the side, a closed leather guest register with a pen resting on it, a ring of brass room keys and a mug of coffee on a wooden reception desk in a guest house, with protea in a vase, in warm morning light.

Does Claude train on your data? Not on Claude Team, Claude Enterprise or Anthropic's API: Anthropic's commercial terms say it "may not train models on Customer Content". On the personal plans, Free, Pro, Max and Ultra, it's a setting each person controls.

It's the first question everybody asks me. Can my team link their Gmail? Can Cowork work on a computer with client files on it? Does the business plan actually protect our guests' details? Honestly, the answers out there are contradictory and confusing. Even the companies' own pages don't always agree. So everything here comes from Anthropic's and OpenAI's own pages, and where they disagree, I say so.

Checked against Anthropic's and OpenAI's own pages on 12 November 2026. The sources are listed at the end of this guide.

Does Claude train on your data? The answer by plan

The plan decides the answer, not the app. The same Claude, on the same laptop, follows different rules depending on whose account it's signed into.

Personal plans: Free, Pro, Max, Ultra Business plans: Team, Enterprise, the API
Trains on your chats? Yes, unless the person opts out in their privacy settings; flagged chats and feedback still can be No, by default, and the contract says Anthropic may not
Who decides Each person, for their own account Your business, under Anthropic's commercial terms
How long Anthropic keeps them Up to five years if they allow training; if not, a deleted chat is gone within 30 days Until someone deletes them, then gone within 30 days; Enterprise can set its own period
Gmail, Drive and Calendar connectors Not trained on, but anything copied into a chat follows the person's setting Not trained on
A data processing agreement No Yes, part of the commercial terms

On the personal plans, Anthropic's privacy policy says chats may be used to train its models "unless you opt out" in your account settings. Even after opting out, conversations flagged for safety review, and anything you report to Anthropic through feedback, can still be used. Incognito chats aren't used to improve Claude, even with training allowed.

The business plans sit under Anthropic's commercial terms. Those terms call everything your team submits, and every reply, your content, and say Anthropic may not train on it. Its data processing addendum is part of the same terms. That's the line you want in writing, and on a business plan you have it.

What's kept, even when it doesn't train

Not training isn't the same as not keeping. Even on a business plan, three things sit outside the "no".

Feedback your team sends

A thumbs up or down on a reply sends the whole conversation to Anthropic. It may be used for training and kept for up to five years, separated from the person's and the company's IDs first. On Team and Enterprise, an owner can switch feedback off for everyone, in the organisation's data and privacy settings.

Chats flagged for safety

If Anthropic's automated checks flag a conversation, its inputs and outputs are kept for up to two years, and the safety scores for up to seven. Business plans included.

The chats themselves

On Team, chats stay in Claude until someone deletes them. Setting your own retention period is an Enterprise feature, 30 days at the least. A deleted chat leaves your history straight away and Anthropic's systems within 30 days.

Can your team use the Claude Gmail connector?

All my clients ask this one: can the team link their Gmail and send emails on the business's behalf? Yes, with two things to know.

First, training. Anthropic says it doesn't train its models on your Gmail, Drive or Calendar connector data. On a personal plan with training allowed, though, anything copied from an email into a chat, or a reply that quotes it, may be used.

Second, what it can do, and here Anthropic's own pages disagree. Its help centre says Claude can search, read, draft, send, reply to and forward emails, asking your approval before each of those by default. Its connector guide calls the Gmail connector read-only: it "can't create, send, or modify messages". Check what yours does before your team relies on it.

On Team and Enterprise, nothing connects until an owner switches the connector on for the organisation, and owners decide whether people may let actions run without asking each time. Keep the approval on. An email that goes out in your name should be read by a person first.

Is Claude Cowork safe with client files?

Cowork is Claude doing the work on your computer: in the folders you connect, and in a browser. It's on every paid plan, and on Team and Enterprise it's switched on unless an owner turns it off.

Anthropic's own advice is the clearest answer I've found: "Avoid granting access to local files with sensitive information, like financial documents." It also strongly advises against letting Claude act in a browser on anything sensitive, because a web page can hide instructions meant for Claude. That trick is called prompt injection.

Three things to know before you connect a folder:

  • Claude can only read and write in the folders you connect, so connect a working folder, never your whole drive.
  • In cloud sessions, the work runs in an isolated, temporary space on Anthropic's servers, removed when the session ends. These are on by default on Team and off by default on Enterprise.
  • In local sessions, the chat history is stored on that person's computer, where an owner can't manage or delete it centrally.

So full access to a computer full of client information: no. A folder with only what the task needs: yes.

Can it read your guest spreadsheets?

Yes, if someone uploads one or connects the folder it lives in. On a business plan, Anthropic's commercial terms count everything your team submits as your content, and Anthropic may not train on it. A spreadsheet uploaded to a chat is a submission like any other. On a personal plan with training allowed, it may be used.

But training isn't the real risk here. The list leaves your systems and sits in the chat until someone deletes it. So send only what the task needs: the columns it uses, without ID or passport numbers, card details or health notes. To write a welcome email for each arrival this week, Claude needs names and dates, not every guest's history.

Who's responsible if guest data leaks?

You are. Whatever plan your team uses, your business stays responsible for the guest data it holds, under its own data-protection law: GDPR in Europe and POPIA in South Africa, among others. This isn't legal advice, so check your setup with your attorney.

What about ChatGPT?

OpenAI's answer has the same shape. On ChatGPT Business and Enterprise: "By default, we do not use your business data for training our models," unless someone opts in to share it, through feedback for example. Data from connected apps isn't used for training by default either.

On Free, Plus and Pro, OpenAI says it plainly: "data sharing is enabled by default." Each person can turn it off in their data controls, and Temporary Chats aren't used for training.

One thing OpenAI's pages leave unclear is how long ChatGPT Business keeps chats. Its privacy page says your admins can control it. Its help centre, on what happens when someone leaves your workspace, says their chats and files are kept indefinitely. If that matters to you, ask OpenAI in writing.

The real risk: your team on personal accounts

If even one person on your team uses AI for work on a personal account, that's where the gap is.

On a personal account, the training setting is that person's, not yours. The chats live in their account, so you can't see, export or delete them, and when they leave, the guest details they pasted leave with them. And none of it is covered by a contract between your business and the AI company.

What to do this week

Put everyone on one business plan

Anyone who uses AI for work should use it on the business's account, not their own. That's what puts a contract between your guests' data and the training. Claude Team plan vs ChatGPT Business helps you choose.

Switch off feedback, and keep approval on

An owner can turn feedback off for the whole organisation, in its data and privacy settings. Switch on only the connectors your team needs, and leave approval before each action on, so nothing goes out in your name unread.

Connect a folder, never the drive

In Cowork, connect one working folder for the task. Keep financial documents, ID copies and full guest lists out of it, as Anthropic itself advises.

Write the rule down

One line everyone knows: what never goes into AI. ID and passport numbers, card details and health notes, for a start. Put it in your team's project as well, so Claude reads it too; here's how to set one up.

Frequently asked questions

Does Claude Pro train on my data?

Yes, unless you opt out. Pro is a personal plan, so Anthropic's consumer terms apply: your chats may be used to improve its models unless you opt out in your privacy settings. Even then, chats flagged for safety review and anything you send as feedback can still be used. If you use Pro for work with client details in it, check that setting today.

Is Claude safe to use for business?

On a business plan, Claude doesn't train on your content, and that's written into Anthropic's commercial terms, with a data processing addendum beside them. The rest depends on how your team uses it: what they paste in, which tools they connect, and whether a person approves what it sends. Those three are yours to set, whichever plan you're on.

Does the Claude Team plan protect our data?

On training, yes: by contract, Anthropic may not train on your team's content, and an owner can switch feedback off too. What Team doesn't give you is an audit log of what people paste in, or your own retention period; those are Enterprise. Here's what Claude Enterprise really costs, or book a call and we'll look at what your team actually needs.

Does ChatGPT train on your data?

On ChatGPT Business and Enterprise, not by default. On Free, Plus and Pro, data sharing is on by default, and each person can turn it off in their data controls. Temporary Chats aren't used for training.

The takeaway

The plan decides the answer. On a business plan, Claude doesn't train on your data, and the contract says so. On a personal plan, it's one person's setting.

But not training isn't the same as safe: the four steps above are what keep your guests' details safe. If you'd like that set up properly for your team, here's what's in every build.

Sources, read on 12 November 2026

Up next

Claude Team Plan vs ChatGPT Business: Which One Fits Your Team?

Claude Team plan vs ChatGPT Business: the same price per seat, different strengths. What each gives a team, what admins can see, and what neither does.

10 minute read

Manie Eksteen, founder of Tidastra

Written by

Manie Eksteen

Manie Eksteen spent twelve years in diving and hospitality before founding Tidastra in Noordhoek, Cape Town.

He now builds AI systems that give service businesses their hours back, from admin to sales, and writes these guides from what he builds.

One call. We'll tell you straight if it's a fit.

We'll look at your business, show you the machine running ours, and map what it would do for you.

Book a call →