"Is my data safe?" came up in about seven of my client calls recently, often from people in legal, finance and manufacturing. My answer is always the same: privacy here is mostly about five minutes of settings and one good habit.
Step 1: Turn off training on your chats
Checked October 2026. Menu names change, so look for similar wording.
ChatGPT
- Click your profile, then Settings.
- Open Data controls.
- Turn off Improve the model for everyone.
Claude
- Open Settings, then Privacy.
- Turn off Help improve Claude. If you leave it on, Anthropic says it can keep your chats for up to five years for training.
Gemini
- Open Settings and help, then Activity.
- Turn off Keep Activity (previously called Gemini Apps Activity). Google still keeps chats for up to 72 hours to run the service.
These settings apply to new chats from now on. They do not delete old chats, so delete those separately if needed.
Step 2: Use temporary chats for sensitive one-offs
- ChatGPT: Temporary Chat. These are not used for training and are deleted within 30 days.
- Claude: Incognito chat. Not used for training.
- Gemini: Temporary Chat. Kept for up to 72 hours and not used to train models.
Step 3: Know what never to paste
- Passwords, OTPs, API keys
- Aadhaar, PAN, bank or card numbers
- Client contracts or health records, unless your company has approved the tool
- Anything under an NDA
Remove names and numbers first. "Client A, ₹X" works just as well for most tasks. This is the habit I push hardest, because it protects you even when a setting changes without you noticing.
Step 4: For businesses, use business plans
ChatGPT Business and Enterprise, Claude Team and Enterprise, and Gemini in Google Workspace do not train on your data by default, and they add admin controls. Pair that with a one-page policy: which tool, which data, who approves.
Step 5: For the most sensitive work, go local
A local model runs on your own laptop, so prompts don't go to an AI company. See my guide on how to install Ollama.
Watch out for connectors and agents
When you connect email, Drive or WhatsApp, the AI can read more than you paste. Start read-only. More in is it safe to let an AI agent use your computer.