Clean Up Your Claude and ChatGPT Privacy Settings

Your private AI chats are not sitting on Google. What can leak is anything you shared or published, and most people have created more of those links than they remember. Then there is the quieter layer: what the assistant remembers about you, and whether your chats help train future models. All of it is under your control, and a full cleanup takes about fifteen minutes.

I use Claude every day for work, so I run this audit every few months. Here is the order I do it in.

Step 1: search for yourself

Start from the outside, the way a stranger would find you. Paste these into Google with your real name, your company, or a phrase only you would have typed:

site:chatgpt.com/share "Your Name"
site:claude.ai/share "Your Name"
site:claude.ai/public/artifacts "Your Name"
site:chatgpt.com/share "your company or project"

Two things to keep in mind. First, most shared chats never mention your name, so search a project name or a distinctive phrase too. Second, an empty result does not mean private. A share link is a working public page for anyone who has the URL, whether Google lists it or not. Do the cleanup below either way.

Step 2: clean up Claude

Shared chats

Go to Settings, then Privacy, and click Manage next to Shared chats. You will see every chat you have shared, with its title and date. Click Unshare on anything you do not want public. You can also open a single chat, open its Share menu and set it to Private. The link stops working immediately.

According to Anthropic's help page on public links, shared chat pages carry a noindex instruction that tells search engines not to list them. That helps, but it does not cover copies that archive or scraping services may have saved, and it does not stop anyone you sent the link to.

Published artifacts

Published artifacts are separate from shared chats and do not show up in that list. Find them in the Artifacts section of the sidebar, open each one and click Unpublish. Read the fine print first: per Anthropic's artifact help page, you cannot publish the same artifact again after unpublishing, and any storage data it used is permanently deleted. Save a copy if you might need it.

Memory

Open Settings, then Memory. Here you can review what Claude remembers, edit or delete individual items, and turn off Generate memory from chats or Search and reference chats. If you want a full reset, there is an option to delete all memory. By default Claude leaves sensitive topics like health and politics out of memory, and there is a separate toggle for that. The details are in Anthropic's memory guide.

For a one off conversation you do not want saved at all, start an incognito chat with the ghost icon. It will not appear in your history or feed memory.

Model training

On Free, Pro and Max plans, go to Settings, then Privacy, and find the toggle under Help improve our AI models. Turning it off means new chats and coding sessions are not used for training. The exact steps are on Anthropic's privacy center. Team and Enterprise plans work under commercial terms and are not used for training.

Step 3: clean up ChatGPT

• Shared links: Settings, then Data controls, then Manage next to Shared links. Delete individual links, or use More actions to delete all of them. Deleting a link stops future access but does not delete the original conversation, and it cannot remove a copy someone already saved.
• Memory: Settings, then Personalization, then Memory. Open Manage to read and delete saved memories one by one or all at once. Depending on your account you will see toggles such as Reference saved memories and Reference chat history. Deleting a chat does not delete a memory that was saved from it, so check both.
• Training: Settings, then Data controls, then turn off Improve the model for everyone. Your chats still appear in your history, they just stop being used for training going forward.
• Temporary Chat: for anything you want kept out of history and memory. OpenAI keeps these for up to 30 days for safety reviews and then deletes them.

Step 4: ask what it knows about you

Settings pages show you the list. A direct question shows you how the assistant actually sees you. Run this in a normal chat, not an incognito or temporary one:

List everything you remember or can reference about me from previous
conversations. Split it into two lists: things I told you directly, and
things you inferred. Quote exact wording where you have it. Do not soften
or summarize. If you have nothing stored, say so plainly.

Treat the answer as a rough mirror. For an actual record, request a full export: in Claude it is Export data under Settings, then Privacy, and ChatGPT has an equivalent in Data controls.

Step 5: push old results out of Google

Deleting a link at the source is the fix. Google may still show the old result for a while. Once the page is gone, submit its URL to Google's Refresh Outdated Content tool. You do not need to own the site, just a Google account. Then set a Google Alert for your name in quotes so you hear about new results without searching.

Habits that keep it clean

• Before you share a chat, reread it top to bottom. Pasted emails, client names and phone numbers hide in the middle.
• For sharing with one colleague, prefer a private or workspace share over a public link when your plan supports it.
• Put a reminder in your calendar to run Step 1 every three months.

Do the search in Step 1 right now, before you change any settings. What it finds tells you exactly how much of the rest you need.