Home » Guides » Does Claude Train on Your Data? Plan by Plan, From the Source

Does Claude Train on Your Data? Plan by Plan, From the Source

Does Claude Train on Your Data? Plan by Plan, From the Source

Short answer

Does Claude train on your data? Whether Claude trains on your data depends on the account, and CoworkStrategy.com sets out each one below from Anthropic’s own privacy pages. For commercial products such as Claude for Work and the Anthropic API, Anthropic’s commercial privacy article says that by default Anthropic will not use your inputs or outputs to train its models, with exceptions for feedback and explicit permission.

This guide is for the owner or operations lead of a small company who is deciding which client documents and company files can go into Claude. The job you want handled is a clear, sourced answer you can write into a staff rule. Claude is the tool the work runs on, so I stick to what Anthropic’s own pages say, name the page behind each claim, and say so where a page is silent.

I sit down with owners, map which accounts their people actually use, and leave a one-page data rule the owner keeps. The checklist near the end of this guide has the same shape.

Key factsDetail
Commercial defaultAnthropic says it will not use inputs or outputs from its commercial products to train its models by default, per Anthropic’s commercial privacy article.
Consumer settingFor Free, Pro, Max and Ultra, Anthropic’s consumer privacy article says chats and coding sessions are used if you choose to allow it, among other listed cases. I state no default for it, because that page states none.
Feedback exceptionIf you report feedback or bugs, for example with the thumbs up or down button, Anthropic says it may use your chats and coding sessions to train its models, per Anthropic’s commercial privacy article.
What the owner can switch offA Primary Owner or Owner of a Team or Enterprise plan can disable member feedback with the Rate chats setting, per Anthropic’s commercial privacy article.

Does Claude Train on Your Data? Start With the Account You Have

Anthropic splits its answer by product family. Anthropic’s commercial privacy article says it covers commercial products such as Claude for Work and the Anthropic API, and it points consumer readers to a separate article for Claude Free, Pro, Max and Ultra. Anthropic’s article on team data describes the Claude for Work plan as the Team or Enterprise plans.

So the first question is never about Claude in general. It is which account the person typed the document into. A founder who pastes a client contract into a personal account and an employee who pastes it into the company’s organization are on different terms, and Anthropic’s article on team data says the consumer terms and privacy policy do not apply to the Claude for Work plan where Anthropic acts as the data processor.

Decision diagram: Claude Free, Pro or Max accounts follow consumer terms where training depends on your setting, while Team, Enterprise and API follow commercial terms, not used for training by default
Anthropic publishes a separate article for each side, so the first step is knowing which account was used.
Three colleagues around a conference table reviewing client folders and deciding what is safe to put into Claude
Decide which account the work goes into before any client file does.

Consumer Plans Versus Commercial Plans: The Two Answers

Start on the consumer side. Anthropic’s consumer privacy article says Anthropic will use your chats and coding sessions to improve its models if you choose to allow it, if your conversations are flagged for safety review, or if you otherwise explicitly opt in to training, such as by joining a program like its Trusted Tester Program. The same article says Incognito chats are not used to improve Claude, even if you have enabled Model Improvement in your Privacy Settings.

I do not tell you what that setting is set to, because that article does not say. Open your own Privacy settings and look. Anthropic’s article on changing the setting says that on a computer you select your name, open Settings, then Privacy, and toggle the model improvement button off or on. It also says that turning it off means new chats and coding sessions are not used for future model training, while data already in training that has started, or in models already trained, stays included.

Here is how the account types line up, with the page behind each cell.

Account typeTraining defaultWhat can change itRetention or owner controls
Free, Pro, Max (consumer)Used if you allow it, if flagged for safety review, or if you otherwise opt in. No default stated on Anthropic’s consumer privacy article.Your own Model Improvement toggle in Privacy settings, per Anthropic’s article on changing the setting.You can delete conversations at any time, per Anthropic’s consumer retention article.
Team (commercial)Not used for training by default, per Anthropic’s commercial privacy article; Anthropic’s pricing page lists no model training on your content by default on its Team card.Reported feedback or bugs, or choosing to allow it, per Anthropic’s commercial privacy article.A Primary Owner or Owner can disable member feedback with the Rate chats setting, per Anthropic’s commercial privacy article.
Enterprise (commercial)Not used for training by default as a commercial product, per Anthropic’s commercial privacy article.Same exceptions: feedback, bugs or explicit permission.Rate chats setting as above, plus custom data retention controls, which Anthropic’s pricing page lists as Enterprise only.
Anthropic API (commercial)Not used for training by default, per Anthropic’s commercial privacy article.Feedback or explicitly choosing to allow it.Inputs and outputs are automatically deleted from Anthropic’s backend within 30 days, with listed exceptions, per Anthropic’s retention article.

Free, Pro, Max and Ultra sit on the consumer side of that table. Team, Enterprise and the API sit on the commercial side, and the two sides do not share terms.

If you want this handled when the account is created, see setting up Claude Team so client data stays private.

What the Commercial Default Says for a Team

Anthropic’s commercial privacy article states the default plainly: by default, Anthropic will not use your inputs or outputs from its commercial products to train its models. I would repeat that wording to staff, and not stretch it into a promise the page does not make.

Anthropic’s article on team data adds the control side. It says your organization’s designated Primary Owner manages the Claude for Work account and its data, including requesting data exports, and that use of the account is governed by the agreement between Anthropic and your organization. If a client asks who decides how your company’s chats are handled, that is the page to read.

A manager pointing at a shared document on dual monitors while two colleagues review a data-handling choice for Claude
A short review of who uses which account catches most surprises.

Three boundaries are worth writing down:

  • Training is one question. Retention, deletion and who inside your company can see chats are separate questions, and the next sections take them in turn.
  • Connectors have a caveat.Anthropic’s commercial privacy article says feedback data does not include raw content from connectors such as Google Drive, though data may be included if it is directly copied into your conversation with Claude.
  • Safety and law still apply.Anthropic’s retention article says that in all cases Anthropic may retain chats or sessions as required by law or as necessary to combat violations of its Usage Policy.

The Feedback Exception: What a Thumbs Up Changes

The exception that surprises owners is the thumbs up or down button. Anthropic’s commercial privacy article says that if you explicitly report feedback or bugs, for example through that button, or otherwise choose to allow it, Anthropic may use your chats and coding sessions to train its models.

Two details from that article matter. It says that when you use the button, Anthropic stores the entire related conversation in its secured back-end for up to 5 years, and Anthropic’s retention article gives the same 5 years for data associated with a feedback submission. It also says Anthropic de-links feedback from your user and customer IDs before use and does not combine it with your other conversations with Claude.

Hub diagram of what can change the commercial default on Claude training: the thumbs up or down button, reported bugs, and explicitly choosing to allow it
Per Anthropic, these are the ways commercial conversations can end up used for training despite the default.

So the real question when an employee clicks thumbs up is what is in that conversation. I would not click it on a chat that holds client details, and I would tell staff the same.

What an Owner Can Switch Off, and What Stays Outside Your Control

Anthropic’s commercial privacy article says a Primary Owner or Owner of a Team or Enterprise plan can disable members’ ability to submit feedback through the thumbs up or down button, using the Rate chats setting under Organization settings > Data and Privacy.

Flow diagram of how an owner switches off member feedback in Claude: Owner or Primary Owner, Organization settings, Data and Privacy, then the Rate chats setting
The path Anthropic’s article gives for disabling the thumbs up and down button for members.

On Enterprise, there is a second control. Anthropic’s article on custom retention says custom data retention controls are available to Enterprise plan customers, need a Primary Owner or Owner role, and live under Organization settings > Data and Privacy. It also says data is retained indefinitely by default unless a custom retention period is set. Anthropic’s pricing page likewise lists custom data retention controls as Enterprise only. If you are weighing that move, my guide to setting up a Claude Enterprise account covers what changes from Team.

Some things stay outside the owner’s hands. Per Anthropic’s retention article:

  • A deleted conversation is removed from your chat history immediately and deleted from Anthropic’s back-end storage within 30 days.
  • Incognito chats are automatically deleted within 30 days unless flagged as a Usage Policy violation.
  • Inputs and outputs from chats flagged by automated trust and safety systems as violating the Usage Policy may be retained for up to 2 years.
  • Feedback submissions are retained for 5 years.
A business owner and a colleague talking across a desk with a client file between them, discussing Claude data privacy
Write the data rule down so staff are not guessing.

A written rule helps staff more than a link to a help article. This prompt drafts one, and it works in a Claude session that has just opened, because it needs only what you type in.

Prompt: draft a staff rule for what goes into Claude

Write a one-page staff rule for a small company about what may and may not be pasted into Claude. Cover: which account people must use for company work (the company's organization, not a personal account), what to leave out entirely, and who to ask when unsure. Keep it under 250 words in plain language. Do not describe how Anthropic handles data and do not state retention periods or setting names; where a line depends on Anthropic's policy, write "confirm on Anthropic's privacy pages" instead.

[list the kinds of client or company information you never want pasted]

You get a draft rule you can edit. Check that it names the account staff must use, and that every line that depends on policy points back to Anthropic’s pages instead of stating a policy of its own.

A Checklist Before You Put Client Data In

Use this as the one-page rule I mentioned. Each step comes from a page cited above.

  1. Find the account. Ask each person whether they use a personal Free, Pro, Max or Ultra account or the company’s organization. The two follow different terms.
  2. Read the personal toggle yourself. On a personal account, open Privacy settings and read the Model Improvement setting rather than assuming it.
  3. Name the Primary Owner. On the company organization, confirm who holds that role, because per Anthropic’s article on team data that person manages the account and its data.
  4. Decide on feedback. If members should not send thumbs up or down feedback, an Owner or Primary Owner can disable it with the Rate chats setting.
  5. Brief staff on the button. Tell them not to click thumbs up or down on chats that hold client details.
  6. List what never goes in. Write down what stays out of Claude entirely, and ask Anthropic or your own counsel about anything with contractual or regulatory duties.

If you would rather not rebuild this from scratch, this prompt turns what you know into a to-confirm list.

Prompt: turn my Claude account facts into a data check

Below I list which Claude accounts my team uses (personal, company organization, or API), who owns the company account, and whether anyone uses the thumbs up or down feedback button. Turn this into a one-page table with three columns: what I know, what I still need to confirm on Anthropic's privacy pages, and who should confirm it. Do not state what any plan's training or retention terms are; list each one as a question for me.

[paste your accounts, the account owner and your feedback habits]

The result is a three-column table. Treat every row in the middle column as an open item and close it against Anthropic’s pages before a client contract depends on it.

Worth watching

Anthropic’s pages describe defaults and exceptions, not guarantees. Anthropic’s commercial privacy article carries both a default and a feedback exception, and each article shows its own date at the top, so re-read the page for your account type before you rely on it in a client conversation.

Key takeaways
  • Consumer and commercial accounts follow different terms, so identify the account first.
  • Commercial products are not used for training by default, per Anthropic’s commercial privacy article, with feedback and explicit permission as exceptions.
  • For Free, Pro, Max and Ultra, training use depends on your choice and other listed cases, per Anthropic’s consumer privacy article.
  • An Owner can disable member feedback with the Rate chats setting.
  • Deleted chats leave your history immediately and back-end storage within 30 days, per Anthropic’s retention article.

If you want someone to map your accounts and write the data rule with you, the next step is a conversation, not a purchase: Claude for teams setup on Strategized.ai is where I scope a team rollout. If you are still reading up first, the free Claude guides on Strategized.ai cover the basics. Two of my own guides pair with this one: what data Claude can access through connectors and the owner’s security settings checklist. The rest of the library is in all CoworkStrategy.com guides.

Frequently Asked Questions

Does Claude train on your data on a Team plan?

Not by default. CoworkStrategy.com relies on Anthropic’s commercial privacy article, which says commercial inputs and outputs are not used for training by default, with exceptions for feedback or explicit permission.

Does Claude use your data for training on Free, Pro, Max and Ultra?

Only in the cases Anthropic lists. CoworkStrategy.com cites Anthropic’s consumer privacy article: if you choose to allow it, if chats are flagged for safety review, or if you otherwise opt in.

Is my data used for model training in Claude if an employee clicks thumbs up?

It may be. CoworkStrategy.com cites Anthropic’s commercial privacy article, which says reporting feedback this way means Anthropic may use your chats and coding sessions to train its models.

Does Claude train on your data pulled in through connectors?

Anthropic says feedback data excludes raw connector content. CoworkStrategy.com cites Anthropic’s commercial privacy article, which adds that data may be included if it is directly copied into your conversation.

Is Claude data privacy the same on the API as on Team?

The training default is the same, but retention is described differently. CoworkStrategy.com cites Anthropic’s retention article, which gives API inputs and outputs a 30-day automatic deletion baseline with listed exceptions.

Does deleting a chat change Claude data privacy for my team?

It removes the chat from history immediately. CoworkStrategy.com cites Anthropic’s retention article, which says back-end deletion follows within 30 days, while feedback data is kept for 5 years.

See also: Does Claude Keep Your Data Private? What an Owner Can Check.

Michael Robichaud

Written by

Michael Robichaud

Founder of Strategized.ai and creator of the Claude AI Strategy Stack. Michael sets Claude up for client teams and runs his own five-site content operation on Claude, so what he writes here is tested in real work first.

 ·  Connect on LinkedIn →

Not Sure Where to Start?

Read It

Start with the free guides on this site, then the Learn Claude guides at Strategized.ai when you want the longer walkthroughs.

Learn Claude guides

Install It

The free Foundation Plugin teaches Claude the basics of your work in one download. No application, no card.

Get the free Foundation Plugin

Have It Set Up for You

For your business, Strategized.ai sets Claude up for the whole team. For yourself, CoworkConsultant.com builds your own setup or a custom plugin.

How we help businesses
Just you? See how we help

Two Ways to Get Claude Working for You

Do it yourself with the free guides and plugin, or have it set up for you. Either way you end up in the same place: a Claude workspace configured around the way you work.

Do It Yourself

The free Foundation Plugin, the guides on this site, and the Learn Claude library at Strategized.ai. No application, no card.

Get the free Foundation Plugin

Have It Set Up for You

For your business: Strategized.ai sets Claude up for the whole team, configured for every role, trained on their own work and handed over.

How we help businesses

For yourself: CoworkConsultant.com builds your own Claude setup, or a custom plugin around the job you repeat most.

How we help individuals
The Foundation Plugin: free for business owners and professionals Get it free → Instant download. No account.