In This Guide:
What Actually Happens to Your Data
When you type a message to an AI chatbot, three things generally happen:
- Your message is sent to the company's servers for processing — this is necessary for the AI to respond
- The conversation may be stored — retained by the company for safety review and, in some cases, model training
- Humans may occasionally review conversations — a small sample of conversations are read by employees to improve safety and quality
Important nuance: "May be used for training" doesn't mean your specific conversation will definitely be used. It means the company reserves the right to use it. In practice, large-scale automated processes, not individual conversations, drive model improvements.
Data Practices by Provider
| Provider | Default Data Retention | Training Opt-Out | Paid Tier Difference |
|---|---|---|---|
| ChatGPT (OpenAI) | Stores conversations; may use for training | Yes — Settings > Data Controls | ChatGPT Team/Enterprise: no training on your data by default |
| Claude (Anthropic) | May retain conversations for 30 days for safety review | Limited on free tier | Claude Pro: stronger privacy commitments; Enterprise: no training |
| Google Gemini | Conversations stored; reviewable by Google | Yes — Gemini Activity in Google Account settings | Google Workspace (business): governed by Google's data processing terms |
| Microsoft Copilot | Varies by account type (personal vs. M365) | Enterprise accounts: no training by default | M365 Copilot: governed by Microsoft's commercial data protection |
The pattern: Free consumer tiers have the weakest privacy protections. Paid business/enterprise tiers universally offer stronger data protection. If you use AI for work with sensitive information, the enterprise tier is worth it.
What to Share vs Never Share
🚫 Never Share
- Passwords or login credentials
- Social Security or national ID numbers
- Credit card or bank account numbers
- Medical records with your full name
- Confidential client data or trade secrets
- Private information about others without consent
- Passport or driver's license numbers
- Biometric data
✓ Freely Share
- General context about your job or industry
- Anonymized or generic scenarios
- Your first name (not full name + identifier combo)
- Questions about public topics
- Draft text you wrote yourself
- General business questions
- Public information
- Creative ideas and concepts
The simple rule: Don't share anything you wouldn't put in an email to a stranger. That mental filter covers 99% of situations.
How to Turn Off Data Training
ChatGPT
Click your profile icon (bottom left)
Go to Settings → Data Controls
Toggle off "Improve the model for everyone"
Google Gemini
Go to myaccount.google.com
Data & Privacy → Web & App Activity
Find "Gemini Apps Activity" and turn off
Note: Turning off model training may reduce some personalization features. It does not delete past conversations — you need to do that separately in each app's history settings.
Using AI for Work Safely
The biggest privacy risk with AI is using free consumer tools for sensitive work tasks. Here's the practical framework:
- Client names and details — use initials or "Client A" instead of real names. AI doesn't need the real name to help you.
- Confidential projects — describe the situation generically. "A tech company" instead of naming your employer's client.
- Legal and financial documents — if you must paste content, redact identifying information first. Or use an enterprise tier with appropriate data agreements.
- Medical situations — you can describe symptoms and situations without including your full name and date of birth together.
5 Privacy Rules to Follow Always
- Never share credentials. No AI tool needs your passwords for any legitimate purpose.
- Anonymize before pasting. Before pasting any document with real names, replace them with generic labels.
- Turn off conversation history when discussing sensitive topics — or start a new chat you manually delete after.
- Match the tier to the sensitivity. Sensitive work = enterprise account. Personal tasks = free tier is fine.
- Treat AI like a searchable tool. Assume a determined person could eventually access your conversations. If you wouldn't want them to see it, don't share it.
The bottom line: AI privacy risks are real but manageable with basic precautions. The vast majority of everyday AI use — writing help, research, brainstorming, learning — involves no meaningful privacy risk at all.