Which tool to choose
You don't have to choose: the guide covers both, because those who care about privacy often use one for one thing and the other for another. The useful difference to know:
- ChatGPT has the more granular controls over memory (you can see, modify and delete the individual saved memories).
- Claude has the narrower collection range and a clean Privacy section; the memory, where present, is more transparent.
In both, the setting that weighs the most is the same: training on your data, on by default.
How to do it
The three controls to sort out are identical in the two services. From a browser or from an app the path is equivalent, only the labels change.
Training on data (the most important).
- ChatGPT: profile, Settings, "Data Controls", turn off "Improve the model for everyone".
- Claude: name at the bottom left, Settings, "Privacy", turn off "Help improve Claude". Check: reopen the screen and verify the switch stayed on off.
Memory. It's the feature that makes the system remember facts about you from one chat to another (people, projects, preferences). Convenient, but it's an archive of personal information.
- ChatGPT: in the data controls or in "Personalization" you find the memory management; there you can disable it entirely or open the list of saved memories and delete individual entries.
- Claude: in the memory settings you can see what's been noted and disable it. You decide: if you use it for work continuity keep it, but review every now and then what it has memorized.
Temporary chat. For every conversation with confidential data, enable it (icon at the top, label like "Temporary chat"): it doesn't stay in the history, doesn't update the memory, isn't used for training.
Export and delete. Both let you download a copy of your conversations and delete them. Use the deletion to clean up the history you no longer need.
If a label isn't where you look for it because the interface has changed, ask the AI in the box: "Where are the privacy settings now and the option to not use my chats in training?".
The working syntax to have it generate your periodic review checklist:
Make me a short privacy checklist to recheck every three months on my AI account.
Include: training on data, memory/saved memories, temporary chat for sensitive data,
history cleanup. One line per item, in English, no frills.
A concrete example
Davide uses ChatGPT for work and Claude for writing. On both he turns off training. On ChatGPT he opens the list of memories and discovers that the system had memorized the name of his company and a client: he deletes the two entries and leaves the memory active only for style preferences. For contracts, from that day he always opens a temporary chat. In ten minutes he put two accounts in order.
When it does NOT work (and how to fix it)
If you disable the memory but the old memories remain
Turning off the memory stops new saves, it doesn't empty the existing archive. Fix: go into the list of memories and delete them by hand, or use the command to wipe the entire memory if available.
If you can't find the Privacy section on Claude
The entries move between "Privacy", "Account settings" and "Memory" depending on the updates. Fix: search by keyword inside the settings, or ask the AI for the updated path as above.
A tip from someone who actually uses it
Set yourself a quarterly reminder to review these settings. Not because something breaks, but because companies release new features (usually on by default) and update the terms. Privacy isn't a one-off action, it's a light but regular maintenance.
Frequently asked questions
Does changing the settings on one device also apply to the others?
Yes. Training and memory are tied to the account, not to the device: you turn them off once and they apply on web, phone and tablet connected to the same account.
If I delete the memory do I lose the conversations?
No, they're two distinct things. The memory is the summary of facts about you; the conversations are the chats in the history. You can delete the memory and keep the chats, or vice versa.
Does configuring privacy make the AI less useful?
It's the fear that stops many. In practice no: turning off training doesn't change the answers you get, and disabling the memory only removes the continuity between different chats, not the quality within the single conversation. The price of privacy here is almost nil; what you lose is marginal compared to the control you gain.