How to Stop AI Tools From Training on Your Data: Claude, ChatGPT, Copilot, and Gemini Compared

The default is not in your favor

Every time you paste a job cost report into ChatGPT, ask Claude to draft an RFQ response, or feed a spec sheet into Gemini, there’s a question most people never ask: Is this being used to train the model?

For most free and consumer-tier AI tools, the answer is yes.

That doesn’t make these tools dangerous. It makes them tools you need to configure correctly before you put proprietary information into them. This post covers the four major platforms — Anthropic Claude, OpenAI ChatGPT, Microsoft Copilot, and Google Gemini — and gives you the exact steps to lock down your data on each one.

The pattern you need to understand

Across all four platforms, the rule is the same:

Free and consumer tiers use your data for model training by default

Business, enterprise, and API tiers do not

Every platform offers an opt-out — but you have to find it and turn it on yourself

If you’re using a personal account on any of these tools for work, your company’s data is likely being used to improve models that your competitors will also benefit from. That’s not a scare tactic — it’s the default setting.

Anthropic Claude

What gets used: On the free and Pro tiers, conversations may be used for training — particularly when you provide feedback (thumbs up/down), have conversations flagged for safety review, or participate in the Trusted Tester program.

What doesn’t: Claude Teams, Enterprise, and API data is never used for training. Anthropic acts as a data processor, and the customer controls the data.

How to opt out (Free and Pro):

Open claude.ai

Click your profile icon (bottom left)

Go to Settings

Select Data Privacy Controls

Toggle off training data usage

Data retention: Deleted conversations are removed from your history immediately and purged from Anthropic’s backend within 30 days.

Direct link: privacy.claude.com

OpenAI ChatGPT

What gets used: Free, Plus, and Pro conversations are used for model training by default. This is the most widely used AI tool in the world, and the default is opt-in to training.

What doesn’t: ChatGPT Teams and Enterprise conversations are never used for training — this is locked and cannot be overridden. API data is also not used for training by default (since March 2023), with optional Zero Data Retention (ZDR) available.

How to opt out (Free, Plus, Pro):

Open chatgpt.com

Click your profile icon (bottom left)

Go to Settings

Select Data Controls

Toggle off “Improve the model for everyone”

What happens when you toggle this off: Your conversations are still stored and accessible in your history, but they won’t be used for training. OpenAI retains them for up to 30 days for abuse and safety monitoring, then deletes them.

Data retention: Consumer conversations are retained as long as your account is active. API data is retained for 30 days (or zero with ZDR).

Direct link: help.openai.com/en/articles/7730893-data-controls-faq

Microsoft Copilot

Microsoft Copilot is actually two different products with two different privacy models. This is where most people get confused.

Microsoft 365 Copilot (enterprise): Your prompts and responses are explicitly NOT used to train foundation models. Data stays within your Microsoft 365 tenant boundary. Admins can manage it through Microsoft Purview. This is the most locked-down of all the platforms by default.

Consumer Copilot (personal Microsoft account or signed out): Weaker guarantees. Microsoft’s general privacy statement says data may be used to develop and improve products, including AI models. If you’re using Copilot without signing in with a work account, treat it like any other consumer AI tool.

How to tell which version you’re on: If you’re signed in with a work or school Microsoft account, look for the green shield icon — that confirms enterprise data protection is active. No shield means consumer-tier privacy.

Direct link: learn.microsoft.com/en-us/copilot/microsoft-365/microsoft-365-copilot-privacy

Google Gemini

What gets used: Free and Gemini Advanced (Google One AI Premium) conversations are used for training by default. When Gemini Apps Activity is turned on, human reviewers may read, annotate, and process your conversations. Google disconnects conversations from your account before review, but they can be retained for up to 3 years.

That’s worth repeating: up to 3 years.

What doesn’t: Gemini in Google Workspace is not used for training — Google acts as a data processor. Paid Vertex AI API data is also not used. However, the free Gemini API (AI Studio) IS used for training — this catches developers off guard.

How to opt out (consumer Gemini):

Go to myactivity.google.com/product/gemini

Toggle off Gemini Apps Activity

What happens when you toggle this off: New conversations are no longer saved. Existing conversations may still be retained for up to 72 hours for safety purposes, then deleted.

Direct link: support.google.com/gemini/answer/13594961

Side-by-side comparison


What to do right now

If you’re using any of these tools for work — even casually — do this today:

1. Check your tier. Are you on a personal account or a business account? This single distinction determines whether your data is protected or exposed.

2. Toggle the opt-outs. If you’re on a consumer tier, go into settings right now and disable training. It takes 30 seconds per platform. The links are above.

3. Audit what you’ve already shared. If you’ve been pasting customer data, pricing, proprietary specs, or internal documents into a consumer-tier AI tool with default settings, that data may already be in a training pipeline. You can’t undo this, but you can stop it from continuing.

4. Set a policy for your team. Even if it’s informal — a Slack message, a one-paragraph email — tell your team which tools are approved, which tier to use, and what data is off-limits. Most employees are using AI tools without guidance. The risk isn’t that they’re doing something malicious — it’s that nobody told them the default settings are a problem.

5. Consider the API. If your company is building workflows around AI, the API tier on every platform is the safest option. No training, minimal retention, and programmatic control over what gets sent. The cost is marginal compared to the risk of a data leak through a consumer chat window.

The bottom line

None of these tools are inherently unsafe. But every one of them ships with defaults that prioritize model improvement over your data privacy. The fix is simple, it’s free, and it takes five minutes. Do it today.

---

References

Anthropic Privacy Center: privacy.claude.com

Anthropic Privacy Policy: anthropic.com/privacy

OpenAI Data Controls FAQ: help.openai.com/en/articles/7730893-data-controls-faq

OpenAI Enterprise Privacy: openai.com/enterprise-privacy

Microsoft 365 Copilot Privacy: learn.microsoft.com/en-us/copilot/microsoft-365/microsoft-365-copilot-privacy

Microsoft Copilot Privacy & Protections: learn.microsoft.com/en-us/copilot/privacy-and-protections

Google Gemini Apps Privacy: support.google.com/gemini/answer/13594961

Google Gemini Activity Controls: myactivity.google.com/product/gemini