Do ChatGPT, Claude and Gemini Train on Your Data?

Do ChatGPT, Claude and Gemini train on your data? The answer depends more on the channel than on the vendor. In apps used with a personal account, OpenAI, Anthropic and Google can use conversations to train their models until the user turns this off. In the paid API and in business plans, all three vendors do not train on customer content by default.

No training does not mean no storage, though. Even in the API, prompts stay with the vendor for a set period, usually up to 30 days, and in a few situations much longer. Below are the rules of all three vendors based on their official documentation (as of 7 November 2026), a list of exceptions and a checklist for companies.

Do ChatGPT, Claude and Gemini train on your data? It depends on the channel

Each of the three vendors has two sets of rules: consumer rules for apps used with a personal account and commercial rules for APIs and business plans. The same data may or may not end up in training, depending on whether an employee pasted it into a personal chat or a company application sent it through the API.

In practice, it helps to tell three channels apart:

• Consumer app: ChatGPT Free, Plus and Pro, Claude Free, Pro and Max, the Gemini app on a personal Google account. Training is possible and has to be switched off in settings.
• Business plan: ChatGPT Business and Enterprise, Claude Team and Enterprise, Gemini in Google Workspace. No training by default, and chat history follows the organization's settings.
• API: the OpenAI API, the Claude API, the Gemini API and Agent Platform. No training by default (the exception is the free Gemini API outside the EEA) and short retention for abuse detection.

Two terms come up throughout this article. Abuse monitoring means checking whether the service is used in breach of its usage policies; it is the reason vendors keep logs even when they do not train. Zero Data Retention (ZDR) is a mode in which the vendor does not store prompts or responses after returning the result.

ChatGPT and the OpenAI API: what goes into training

ChatGPT Free, Plus and Pro

On personal accounts, OpenAI enables data sharing for model improvement by default (Data Controls FAQ). You switch it off in Settings → Data controls → "Improve the model for everyone". Your chat history stays, and new conversations are no longer used for training. The setting applies to the whole account on every device.

A Temporary Chat is not used for training and does not appear in your history, but OpenAI may keep it for up to 30 days for safety purposes. Deleted conversations and temporary chats are removed from OpenAI's systems within 30 days (OpenAI, update of 22 Oct 2025).

ChatGPT Business, Enterprise and the API

OpenAI states that by default it does not use data from ChatGPT Enterprise, Business, Edu, ChatGPT for Healthcare, ChatGPT for Teachers or its API platform, including inputs and outputs, to train or improve its models (OpenAI Business data). For the API, this rule has applied since 1 March 2023.

No training does not mean nothing is stored. According to the API documentation, OpenAI creates abuse monitoring logs by default and keeps them for up to 30 days. Some features keep data longer: the Responses API stores response state for 30 days by default (the store parameter), and files, threads and Assistants objects are kept until deleted. Modified Abuse Monitoring and Zero Data Retention both require prior approval from OpenAI.

Since February 2025, companies in the EU can choose to store data in Europe for new API projects and new ChatGPT Enterprise and Edu workspaces. Existing projects cannot be moved (OpenAI, data residency).

Claude and the Anthropic API: what goes into training

The Claude app (Free, Pro, Max)

On 28 August 2025, Anthropic changed the terms for consumer accounts: conversations and Claude Code sessions can be used for training if the user allows it in the privacy settings (Anthropic). With consent, data is kept for up to 5 years; without it, the 30-day period applies.

Incognito chats are not used for training, even when Model Improvement is turned on in settings. A deleted conversation disappears from your history immediately and from Anthropic's back-end systems within 30 days (Privacy Center).

Two exceptions apply regardless of your settings. When you rate a response, Anthropic stores the entire conversation for 5 years and may use it, among other things, to train its models (Privacy Center). Conversations flagged by safety systems as violating the usage policy are kept for up to 2 years, and their classification scores for up to 7 years.

The API and business plans

According to the Claude API documentation, data retained by Anthropic is never used for training without the customer's express permission. Anthropic automatically deletes API inputs and outputs within 30 days of receipt or generation. The exceptions are a ZDR agreement, usage policy enforcement and legal requirements (page updated 7 Nov 2026).

ZDR is enabled by Anthropic's sales team, separately for each organization. It does not cover, among others, the Claude Console, the Files API, batch processing (29-day retention), the code execution tool (up to 30 days) or Claude Managed Agents. Since June 2026, Claude Fable 5 and 5.1 and Claude Mythos 5 and 5.1 require 30-day retention and are not available under ZDR without separate authorization from Anthropic. Content flagged by safety systems may be kept for up to 2 years even under ZDR.

Gemini, the Gemini API and Agent Platform: what goes into training

The Gemini app

For adult users, the Keep Activity setting (formerly Gemini Apps Activity) is on by default, and activity is auto-deleted after 18 months. You can change this to 3 or 36 months or turn auto-delete off (Google Help).

According to the Gemini Apps Privacy Hub (updated 26 Oct 2026), some conversations are read by human reviewers who rate the responses. Google keeps reviewed conversations for up to 3 years and does not delete them along with your history. Google itself asks users not to enter confidential information they would not want a reviewer to see.

With Keep Activity off, and in temporary chats, conversations stay with your account for 72 hours. They are not used for training unless you send feedback, which Google keeps for up to 3 years.

The Gemini API: free vs paid

In the free Gemini API and in Google AI Studio, Google may use prompts and responses to improve its products, and human reviewers may read them. The terms explicitly advise against submitting sensitive, confidential or personal information there (Gemini API terms). For users in the EEA, Switzerland and the UK, the same terms apply the paid-service rules to all services, including the free ones.

In the paid Gemini API, Google does not use prompts or responses to improve its products, but it logs them for a limited period to detect policy violations. Grounding with Google Search stores prompts and responses for 30 days. Google itself points to Agent Platform for workloads that need guaranteed ZDR (Gemini API, ZDR).

Gemini Enterprise Agent Platform and Google Workspace

In Gemini Enterprise Agent Platform (formerly Vertex AI), Google commits not to train models on customer data without the customer's permission. Full ZDR, however, requires action on the customer's side: disabling data caching, requesting an exception from abuse monitoring logging and avoiding features that store data, such as Grounding with Google Search (Agent Platform documentation).

In Google Workspace, prompts count as customer data under the Cloud Data Processing Addendum and are not used for training without the customer's permission (Workspace Help). According to the Workspace Privacy Hub, content is also not reviewed by humans outside the customer's domain without permission.

"We don't train" does not mean "we don't store"

Turning off training closes only one of the ways data stays with a vendor. The others also apply to APIs and business plans:

• Abuse monitoring. OpenAI keeps API logs for up to 30 days, Anthropic deletes API data within 30 days, and Google logs paid Gemini API data for a limited period.
• Flagged content. Anthropic keeps conversations that safety systems flag as policy violations for up to 2 years, even under ZDR.
• Response rating buttons. A thumbs up or down sends the whole conversation to the vendor. Anthropic keeps such feedback for 5 years, Google for up to 3 years.
• Stateful features. Files, batch processing, assistant threads and agent sessions keep data longer than a single request and are usually not covered by ZDR.
• Litigation. In May 2025, a court in the case brought by The New York Times ordered OpenAI to preserve ChatGPT and API logs, including those deleted by users. According to OpenAI, this obligation ended on 26 September 2025.
• Models with mandatory retention. Anthropic's latest Mythos-class models require 30-day retention on every platform, including AWS and Google Cloud.

How to send company data to an LLM safely: a checklist

The steps below follow directly from the vendor documentation described above.

1. Company data only through a business plan or the API. Personal ChatGPT, Claude and Gemini accounts have training on by default or available, and Gemini also has human review. Your company policy should state this explicitly.
2. Check which mode your Gemini API project runs in. A project without active billing is a free service. Outside the EEA, that means possible training and human review.
3. Turn off training on any personal accounts in use. ChatGPT: Settings → Data controls. Claude: privacy settings (Model Improvement). Gemini: Settings & help → Activity → turn off Keep Activity.
4. Do not rate responses in conversations with confidential data. At Anthropic and Google, feedback bypasses the training settings.
5. Use temporary or incognito mode for one-off sensitive questions, keeping in mind that the vendor still keeps such a chat for 72 hours (Gemini) or up to 30 days (ChatGPT).
6. If you need ZDR, apply for it and check the list of exclusions. At every vendor, files, batch processing or caching fall outside ZDR.
7. Consider EU data residency. At OpenAI, you choose it when creating a new API project or a new ChatGPT Enterprise workspace.
8. Review the terms every quarter. Anthropic changed its consumer terms in 2025, Google published new Gemini API terms in March 2026, and Anthropic's retention page was updated on the day this article was written.

Summary

It is worth replacing the question "does this model train on my data" with "which channel am I using to send data, and what does that channel store". In business plans and the paid API, the answer to the first question is "not by default" at all three vendors. The differences lie in retention, exceptions and ZDR conditions, and those are what you need to check before connecting an LLM to customer data.

If you are planning to integrate an LLM with company data and want to choose the right channel and retention settings, book a free consultation with the Midapt team.