Does Claude Train on Your Data? (Plus ChatGPT & Gemini) — How to Opt Out (2026)

TL;DR. On the consumer plans, the answer is yes, unless you turn it off — Claude, ChatGPT, and Gemini can all use your conversations to improve their models until you switch the setting off. That wasn't always true for every vendor, and it's the kind of policy that shifts, but it's one setting in each. Opting out stops future training and shortens retention, but it isn't a full erase, and the business/API tiers don't train by default. Checking a specific plan or a tool not covered here? Use the training and retention lookup.

"Is the AI reading my stuff to train itself?" is one of the most common — and most reasonable — questions people have. Here's the honest, current answer for the three big tools, plus exactly how to turn it off.

First, what "training on your data" actually means

Two different worries get blurred together, and separating them helps:

1. "They'll use it to train the model." This is the one you can control with a setting (below), or by using a tier that never trains.
2. "A company can see or store it at all." This is true of almost any cloud tool — your email provider, your bank's app, your cloud drive. If your bar is "no company ever processes my data," that rules out most of modern software, not just AI.

Most people are really objecting to #1, which is fixable. So let's fix it.

How to turn it off in each app

All three paths re-verified against Anthropic's Privacy Center, OpenAI's Help Center, and Google's Gemini Apps Privacy Hub on September 6, 2026.

Claude (Anthropic)

Select your name from the settings menu → Settings → Privacy → under Help Improve our AI models, click the button to toggle it off. Same path on mobile.

Worth knowing: in August 2025, Anthropic changed its consumer terms so that Claude Free, Pro, and Max chats (and Claude Code used on those accounts) can be used for model improvement unless you opt out — and opted-in data can sit in Anthropic's training pipelines, de-identified, for up to five years.

Opting out stops both your previous and new chats from going into future training; data already in a completed or in-progress training run stays there. Deleted conversations leave Anthropic's back-end within 30 days. Two more details from the Privacy Center: Incognito chats are never used for training regardless of the toggle (they're still kept for 30 days for safety), and conversations flagged for safety review can still be used to improve Anthropic's safety classifiers. If you've used Claude since the 2025 change and never picked, check this setting.

Anthropic's own pricing page now spells this out in the plan comparison: model training is listed as "Opt-out" on every individual plan (Free, Pro, Max), and Team and Enterprise get no model training on your content by default — Enterprise adds custom data retention controls on top.

ChatGPT (OpenAI)

Settings → Data Controls → turn off "Improve the model for everyone." Or, for a one-off, use a Temporary Chat (the "Temporary" button at the top right of a new chat) — it isn't saved to history or used for training, though OpenAI keeps it for 30 days for safety before deleting it. This applies to the Free, Plus, and Pro consumer plans, which use conversations for training unless you turn this off.

Gemini (Google)

Turn off "Keep Activity" (your Gemini Apps Activity). With it on (the default), Google uses your conversations to improve its services, including training models — and a sample of chats is reviewed by humans. Two honest caveats: after you turn it off, recent chats are kept for about 72 hours, and anything a human already reviewed can be retained for up to three years. Gemini also has a Temporary chat mode for one-offs, which is excluded from training and human review and follows the same 72-hour window.

The escape hatch: business and API tiers don't train

If you want a hard guarantee rather than a toggle, the commercial tiers are the answer. By default they do not use your inputs for training:

• Claude Team / Enterprise (Anthropic's Claude for Work plans) and the Claude API
• OpenAI Business / Enterprise / Edu
• Google Workspace with Gemini (business/enterprise)

This is the main reason regulated professional work (legal, healthcare, finance) belongs on those tiers, not the consumer apps.

What opting out does — and doesn't — do

Be clear-eyed about it: opting out stops future training and usually shortens retention. It is not a full erasure of everything you've ever sent. Data already used in a completed training run can't be pulled back out, and (as above) some retention windows persist even after you opt out.

So the simple, honest rule:

Don't paste anything you'd hate to see leak into a consumer AI you haven't switched off. Turn training off, use a business tier, or redact the sensitive parts.

Working with regulated or client data?

If you're handling protected health information, client financials, or other regulated data, the consumer-tier toggle isn't enough — you need the right tier and the right paperwork. Start with PHI vs PII: what actually counts as protected data when you use AI, then which AI vendors will sign a Business Associate Agreement (BAA).

Sources

• Anthropic: Updates to Consumer Terms and Privacy Policy (Aug 2025)
• Anthropic Privacy Center: Is my data used for model training?
• Anthropic Privacy Center: How do I change my model improvement privacy settings?
• Anthropic Privacy Center: How long do you store my data?
• Anthropic Help Center: Use incognito chats
• Anthropic: Claude pricing — plan comparison
• OpenAI Help Center: Data Controls FAQ
• OpenAI Help Center: Temporary Chat FAQ
• Google: Gemini Apps Privacy Hub