An AI privacy checklist is a short set of settings and habits that limits how your conversations with ChatGPT, Claude and other AI assistants are stored, shared and used to train models. The one below has 10 steps, and most people can work through it in about ten minutes.
It covers the plan you use for work, the training and memory switches, the share links you have forgotten about, the extensions in your browser, and the information that should never go into a consumer AI tool at all.
We publish AI systems like this for 300,000+ senior professionals at AI Central, and more of them live in the AI Central Library.
Why most AI users have a privacy gap
Most people use AI like a search bar: type a question, read the answer, move on. But a chat is not a search. Every conversation is stored, some are reviewed, and depending on your plan and settings, some are used to train the model.
Several settings start in the less private position. Training may be on until you turn it off, memory keeps context between sessions, and a share link you created for a colleague months ago is still a public URL today.
Each fix is a single setting or habit, and the 10 steps below show where each one lives.
How to run an AI privacy check-up in 10 steps
1. Choose the right plan for work
Keep company and client work on a Business plan. Free and Personal plans often use your conversations for training; Business plans usually do not. Before you sign up, check the seat minimum and read the data policy, so you know whether the plan uses your data for training.
2. Turn off model training
Open your account's privacy or data settings and disable model training. By default, your chats may be used to improve the model. Switching it off keeps your conversations out of the training data.
3. Skip feedback on sensitive chats
The thumbs up and thumbs down buttons often send the whole thread for review along with your rating. On any chat that contains sensitive context, don't rate the response.
4. Keep sensitive chats private
A share link is a public URL. Anyone who has it can read the conversation, and it can be forwarded or end up indexed. Don't create share links for personal or sensitive conversations.
5. Use temporary chat for personal topics
Temporary Chat in ChatGPT and incognito chats in Claude don't appear in your history, don't create memories and aren't used for training, so you can ask about personal matters without leaving a trail in your account. Both providers can still keep a copy for up to 30 days for safety, so these modes reduce what is kept rather than guaranteeing nothing is.
6. Turn off memory
Memory keeps context across sessions, which means it also keeps your data. Disable it for more privacy, then review and delete the memories already saved, so no long-term profile is built from your conversations.
Share links stay live long after you have forgotten them. Find and switch them off here:
Claude: Settings → Privacy → Shared chats → Manage, then Unshare
ChatGPT: Settings → Data Controls → Shared Links, then delete
8. Check your Codex settings
Coding tools have data settings separate from your main account. If you use Codex, review its own training controls, including the one that covers full-environment data, so your codebase and environment aren't used for training without your knowledge. Our guide to setting up Codex in ChatGPT covers the rest of the setup.
9. Audit your AI extensions
Browser extensions that add an AI sidebar can see everything you type, including passwords and private data. Remove any you don't trust, and check which of the remaining ones can read your chats.
10. Keep real secrets out
Some information should never enter a consumer AI tool, whatever the settings: health, legal and client matters, and unreleased work. Keep it out of personal accounts and use an approved business setup instead. For the rules on what is allowed at work, Chapter 9 of the AI Bootcamp covers how to use AI at work without breaking policy.
What actually changes
Before the check-up, your chats may be used for training by default, old share links stay public, memory builds a profile from your conversations, and extensions can see what you type. After it, work sits on a plan built for it, training and memory are off or reviewed, the old links are closed, and the most sensitive information never enters the chat.
If you would rather set memory up deliberately than switch it off, our guide to AI custom instructions and memory shows how. For more checklists like this one, see the AI Central Library.
Frequently Asked Questions
How do I stop ChatGPT or Claude from training on my chats?
Open your account's privacy or data settings and turn off model training: in Claude it sits under Model Improvement in your Privacy Settings, and in ChatGPT under Data Controls. The change applies to new conversations. Business and Enterprise plans from OpenAI and Anthropic don't train on your data by default.
Yes. Anyone who has the link can read the conversation, without an account. To switch old links off, go to Settings → Privacy → Shared chats in Claude and click Unshare, or Settings → Data Controls → Shared Links in ChatGPT and delete them. Switching a link off does not remove copies or screenshots someone has already made.
Does temporary chat delete everything?
No. Temporary Chat in ChatGPT and incognito chats in Claude stay out of your history, don't create memories and aren't used for training, but both providers can keep a copy for up to 30 days for safety. They keep a conversation out of your account, not off the provider's systems entirely.
Is it safe to put client data into ChatGPT or Claude?
Not in a personal or consumer account. Health, legal, client and unreleased work belong in an approved business setup, because some data should not touch a consumer tool whatever the settings.
How long does an AI privacy check-up take?
About ten minutes for all 10 steps.
