# Does Claude train on your chats? A 6-part settings check for personal plans, Team, API and Claude Code

> Whether Claude trains on your chats depends on your account: Free, Pro and Max follow a privacy toggle, while Team, Enterprise and API default to no training. Feedback buttons, safety flags, Claude Code's /feedback command and plain-text local transcripts are the easy-to-miss parts. A 10-minute checklist is at the end.

- URL: https://zero2.perhapxin.com/en/a/claude-data-training-settings-check
- Published: 2026-10-28
- Updated: 2026-10-28
- Language: en

# Does Claude train on your chats? A 6-part settings check for personal plans, Team, API and Claude Code

One setting on your Claude account decides whether new conversations can be used to train future models. Business plans default to no training. Personal plans depend on a toggle. Which rules apply to you depends on whose account you are logged into, so this checklist starts there and then goes through six checks. It takes about 10 minutes.

Every rule below comes from Anthropic's privacy center and product docs, which I read on October 28, 2026. I did not log into claude.ai, so I could not check what the settings screens look like or whether the toggle is on or off by default for a new account. I flag those spots as they come up.

## First: which kind of account do you have

Anthropic splits its products into two sides and documents each in a separate article.

| Your account | Side | Training default |
|---|---|---|
| Free, Pro, Max (including Claude Code logged in with one of these) | Personal | Depends on the privacy toggle |
| Team, Enterprise, Anthropic API, third-party platforms, Claude Gov | Business | Not used for training by default |

The confusing case is "my company pays" versus "I pay." If you sign in to claude.ai with a work account on your company's Team or Enterprise plan, you are on the business side. If you signed up with your own email and paid for Pro or Max with your own card, you are on the personal side, even if you use it for work.

Claude Code does not sit on its own side. It follows whichever account you logged in with. Log in with a personal account and the personal rules apply. Log in with an API key or a company plan and the business rules apply. There is a section on it below.

## Check 1: the "Help improve our AI models" toggle (personal plans)

The official help article lists three situations where your chats and coding sessions may be used to improve models:

1. You chose to allow it
2. A conversation was flagged for safety review
3. You explicitly joined a program, such as the Trusted Tester Program

Only the first is something you control. On desktop and mobile the path is the same: click your name, open Settings, choose Privacy, and find "Help improve our AI models." Claude Code's docs also give a direct address: claude.ai/settings/data-privacy-controls.

I cannot tell you the default. The articles I read say "if you allow it" and never say whether a new account starts with the toggle on or off. Don't trust your memory of what you clicked months ago. Open the page and look.

The scope is the whole conversation: what you pasted, your custom styles, your conversation preferences, and data collected when you use Claude for Chrome. Raw content returned by connectors (Google Drive, for example) and by MCP servers is not included. The docs add one caveat: if you copy that content and paste it into the chat yourself, it can be included. So letting Claude read a file through a connector and pasting the same file into the message box are treated differently for training purposes.

## Check 2: what turning it off does to old conversations

Turning the toggle off does not pull anything back. The official wording has three layers:

- After you turn it off, neither your new conversations nor your previously saved ones are used for future training
- Training that has already started, and models that have already been trained, still contain past data
- Conversations you delete are not used for future training

If a specific old conversation worries you, delete it. Flipping the toggle alone is not the same thing. Deletion also has a delay: the docs say the chat is removed from your history immediately and from backend storage within 30 days.

There is a timing detail that is easy to miss. The 5-year retention period that applies when you allow training (de-identified, inside training pipelines) covers only conversations created or continued after you turned the setting on. Claude Code's data usage page puts the two cases side by side: 5 years if you allow model improvement, 30 days if you don't.

## Check 3: what can still be used after you turn it off

Three things stay in play whatever the toggle says.

**Safety flags.** If an automated trust and safety system flags your conversation, Anthropic may use or analyze it to improve its ability to detect and enforce usage policies, including training models for its Safeguards team. The settings article repeats this warning. Retention differs too: for conversations flagged as violating the usage policy, inputs and outputs are kept up to 2 years, and the trust and safety classification scores up to 7 years. Most people will never hit this.

**Feedback buttons.** If you press 👍 or 👎 and submit the report, the whole related conversation (content, custom styles, conversation preferences) is stored on the backend for up to 5 years. It can be used to analyze the effectiveness of the service, for research, to study user behavior, and to train models. The docs say it is disconnected from your user ID before use and is not merged with your other conversations. This applies on both personal and business plans.

**Explicit programs.** The Trusted Tester Program, for example. You have to join it yourself.

My own advice: the feedback buttons are the easiest one to hit by accident. If a conversation contains client data, don't tap 👎 to mean "this answer was bad." Tell Claude what is wrong in the chat instead.

## Check 4: business plans default to no training, with one switch to manage

For Team, Enterprise, API and Claude Gov, the docs are direct: inputs and outputs are not used to train models by default. The business article I read was dated August 18, 2026.

There is one category of exception: you actively report feedback or a bug (pressing 👍 or 👎, for example), or you choose to allow data use yourself. So on business plans, feedback is the thing to manage.

A Primary Owner or Owner on Team or Enterprise can turn off members' ability to submit feedback for the whole organization. The path is Organization settings, then Data and Privacy, then "Rate chats." I did not verify this screen either, so go by the official wording and check the names against what you see. If you are not an Owner, ask your admin.

One thing worth noting here: the docs describe this as an organization-level setting that disables feedback through 👍 👎 for all members. For a team handling sensitive data, I would turn it off by default and let specific people send product feedback through another channel when needed.

Retention is a separate question from training:

- Anthropic API: inputs and outputs are deleted automatically within 30 days. Exceptions are longer-retention services you control (such as the Files API), a separate agreement with Anthropic (zero data retention, for example), retention needed to enforce usage policies, and legal requirements
- Other business products (Claude for Work, Enterprise, Console): chats and coding sessions stay in the product to give you a consistent experience. When you delete a chat it leaves your history immediately and the backend deletes it within 30 days
- Flagged for violating usage policy: inputs and outputs up to 2 years, classification scores up to 7 years
- Feedback: 5 years

The API default is deletion within 30 days. If you need "stored nowhere," you have to negotiate zero data retention separately.

## Check 5: what Incognito chats actually leave behind

Incognito is available on every plan, Free through Enterprise. The docs say Incognito chats are not used for training even when your personal training toggle is on. They are not saved to chat history or to Claude's memory, Claude does not pull from them when you search past conversations, and they are not counted in the monthly recap.

They do not vanish, though. The docs state that Incognito chats are still retained for 30 days by default, or for the custom period if an Enterprise organization set a longer one. On Team and Enterprise, Incognito chats appear in the organization data that Owners can export, and on Enterprise they fall under the Compliance API.

Two more details. Claude can still read your custom styles and personal preferences inside an Incognito chat. And you cannot convert an Incognito chat into a regular one after it starts.

People often mix this up with turning off Memory. If you switch Memory off from the "+" menu before you start, the conversation still appears in your history. Claude just doesn't use or add memories, and it can still be found when you search other chats. If what you want is "keep this out of my history," use Incognito. If you only want to keep it from affecting memory, turning Memory off is enough. The docs I read don't say how the training rules treat a chat with Memory off, so I'm not going to guess.

## Check 6: where Claude Code fits

Claude Code's data usage page splits it the same way, by the account you logged in with:

- Personal accounts (Free, Pro, Max): when the setting is on, data from Claude Code is also used to train new models. Retention is 5 years with the setting on and 30 days with it off
- Business (Team, Enterprise, API, third-party platforms, Claude Gov): Anthropic does not train generative models on Claude Code code or prompts unless the customer opts in to providing data, for example through the Development Partner Program
- Business retention: 30 days as standard. Zero data retention is enabled per organization by Anthropic and is not included in standard Enterprise plans

Then there are four sources people tend to treat as harmless.

**/feedback, /bug, /share.** Transcripts sent with these commands are kept for 5 years.

**The rating survey.** You may see "How is Claude doing this session?" If you only answer the rating, only the rating is recorded. Dismissing it is the same. A follow-up may ask whether Anthropic can look at your transcript. Choosing Yes uploads the full transcript, any subagent transcripts and the raw session log. Known API key and token formats are masked, but source code and file contents go up as they are. That data is kept for up to 6 months. The docs say this answer does not affect your training preference and cannot be used to train models. To turn the survey off, set the environment variable `CLAUDE_CODE_DISABLE_FEEDBACK_SURVEY=1`.

**Telemetry.** Usage metrics and error reporting can each be disabled with an environment variable, or you can set `CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC` to turn off all nonessential traffic at once. The docs warn that setting `DISABLE_TELEMETRY` or `CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC` also disables feature-flag evaluation, which can make Remote Control unavailable.

**Local transcripts.** This is the one I looked at myself. The docs say Claude Code stores session transcripts in plain text under `~/.claude/projects/`, for 30 days by default so you can resume sessions, adjustable with `cleanupPeriodDays`. I checked in a cloud sandbox (Claude Code 2.1.312, October 28, 2026). The project folder for this task held one `.jsonl` file of about 513 KB, readable and writable only by its owner (`-rw-------`). It is plain JSON you can open directly, and it starts with the prompt for this task. This has nothing to do with Anthropic's training policy. It is a plain-text record on your own computer. If you ever pasted a password, an API key or client data into a session, it may be sitting in that folder. To shorten it, lower `cleanupPeriodDays`. The docs also note that transcripts from sessions started or recently resumed from Claude Desktop or Cowork are not subject to that period by default.

Claude Code running in the cloud, like the environment I used for this task, is different: Anthropic stores the transcript so you can return to the session, and the retention and usage policies for your account type apply. You can delete an individual cloud session, and the event data is then permanently removed.

## Three situations and what to do in each

What follows is my own reasoning from the rules above. It is not official guidance.

**A freelancer on a personal Pro account, summarizing client interview transcripts.** You are on the personal side. You pasted the transcripts into the chat yourself, so the connector exception doesn't help. If the training toggle is on, that conversation falls within what can be used to improve models. Turn the toggle off before doing this kind of work, or start an Incognito chat. Delete the conversation afterward, and the backend clears it within 30 days. If the toggle was already on earlier, training that has begun and models already trained can't be taken back, and deleting the chat doesn't change that.

**A ten-person team on Team, drafting proposals.** You are on the business side, so no training by default. The thing to deal with is feedback. When a colleague thinks an answer was poor and submits a 👎 report, the whole conversation enters 5-year feedback storage. An Owner decides whether to turn off Rate chats, and should tell the team once, in the team channel.

**An engineer running Claude Code on a personal laptop with a personal account.** The training toggle decides whether code and prompts can be used for training, and separately there is a plain-text record in `~/.claude/projects/`. Handle them as two things: the toggle lives in account settings, and the plain-text record is shortened with `cleanupPeriodDays` or cleared on a schedule.

What the three have in common is that the thing that moves data out of your control is often a small habit: pressing 👎, choosing Yes on a survey, pasting a whole document into the chat box.

## The 10-minute checklist (7 steps, covering the 6 checks above)

Go in order and write down what you find at each step.

1. Confirm the account type: are you logged in with your own account (personal) or a work account your company set up (business)? For Claude Code, ask whether you logged in with a personal account, an API key or a company plan
2. Personal: open claude.ai/settings/data-privacy-controls and see whether "Help improve our AI models" is on or off. Turn it off if you don't want your chats used
3. Personal: to pull back specific old conversations, delete them. Don't rely on the toggle alone
4. Business admins: Organization settings, then Data and Privacy, then decide whether to turn off "Rate chats"
5. Everyone: for client data, contracts or internal code, use Incognito or confirm the organization settings, and avoid pressing 👍 or 👎 on those chats
6. Claude Code users: decide whether to set `CLAUDE_CODE_DISABLE_FEEDBACK_SURVEY=1`, then look inside `~/.claude/projects/` and check `cleanupPeriodDays`
7. Compliance needs: if you need the API to store nothing, negotiate zero data retention. Don't assume it is the default

## What I did not verify

- I did not log into claude.ai or an admin console, so I did not check the real settings screens, the default state of the toggle, or whether menu names have changed. The paths here quote the official text. Compare them with what you see before acting
- These documents get updated. The business article I read is dated August 18, 2026. The other pages show no date (the Incognito page says "Updated over a month ago"), so I can only vouch for what they said when I read them on October 28, 2026
- The docs don't describe how turning Memory off relates to training, and I did not infer it
- I read the English pages. Translations exist and their wording may differ
- This is a summary of public documentation, not legal advice. For contracts, compliance or personal data, rely on the Trust Center and the agreement you signed

## FAQ

**I pressed 👎 and submitted it. Can I take it back?** The documentation I read doesn't describe a way to withdraw feedback. You can avoid it from now on, and you can use the privacy rights available where you live. Anthropic's privacy documentation says you can request a copy, object to processing or ask for deletion, and it also says these rights are limited and that processing involving training datasets is complex.

**Can my coworkers see my chats on a company Claude account?** This article only covers training and retention. The docs do say that Incognito chats on Team and Enterprise show up in organization data that Owners can export, so don't assume admins can't see them. For the full access model, check your company's configuration.

**The API deletes data after 30 days. Does that make it the safest option?** Thirty days is the default. Exceptions include conversations flagged for violating usage policy, legal requirements, and long-retention services you use yourself (the Files API, for example). To avoid storage, you need a zero data retention agreement.

## Sources

- Is my data used for model training? (business): https://privacy.claude.com/en/articles/7996868-is-my-data-used-for-model-training
- Is my data used for model training? (Free, Pro, Max): https://privacy.claude.com/en/articles/10023580-is-my-data-used-for-model-training
- How long do you store my data? (personal): https://privacy.claude.com/en/articles/10023548-how-long-do-you-store-my-data
- How long do you store my organization's data? (business): https://privacy.claude.com/en/articles/7996866-how-long-do-you-store-my-organization-s-data
- How do I change my model improvement privacy settings?: https://privacy.claude.com/en/articles/12109829-how-do-i-change-my-model-improvement-privacy-settings
- Use incognito chats: https://support.anthropic.com/en/articles/12260368-using-incognito-chats
- Claude Code data usage: https://code.claude.com/docs/en/data-usage


## FAQ

### If I turn off the training toggle, are old chats pulled back?

No. Per the docs, turning it off stops use for future training, but training already underway and models already trained still contain past data. Deleted chats are not used for future training.

### Do Team or API chats get used for training?

Not by default. The exceptions are feedback you submit (such as 👍 or 👎) or data use you choose to allow. Owners on Team and Enterprise can turn off Rate chats under Organization settings, Data and Privacy.

### Is Claude Code personal or business?

It follows the account you log in with: Free, Pro and Max use the personal rules; Team, Enterprise and API use the business rules. Local transcripts are stored in plain text for 30 days by default, adjustable with cleanupPeriodDays.

## Sources

- [Is my data used for model training? (business)](https://privacy.claude.com/en/articles/7996868-is-my-data-used-for-model-training)
- [Is my data used for model training? (Free, Pro, Max)](https://privacy.claude.com/en/articles/10023580-is-my-data-used-for-model-training)
- [How long do you store my data? (personal)](https://privacy.claude.com/en/articles/10023548-how-long-do-you-store-my-data)
- [How long do you store my organization's data? (business)](https://privacy.claude.com/en/articles/7996866-how-long-do-you-store-my-organization-s-data)
- [How do I change my model improvement privacy settings?](https://privacy.claude.com/en/articles/12109829-how-do-i-change-my-model-improvement-privacy-settings)
- [Use incognito chats](https://support.anthropic.com/en/articles/12260368-using-incognito-chats)
- [Claude Code data usage](https://code.claude.com/docs/en/data-usage)