I ran a little experiment a few weeks ago and it rattled me. I asked my AI assistant what it remembered about me, and the list ran way longer than I expected. My city, my job, a side project I’d mentioned once in passing months ago. All of it sitting quietly in memory. So when I spotted this post from a privacy-minded LinkedIn creator laying out a three-prompt routine to get that under control, I read it twice and started running it the same evening.
The original poster opens with a blunt filter: ignore the whole thing if you don’t care about your privacy with AI. But if you want basic privacy, the steps are simple, and they work on both ChatGPT and Claude. Below is the full walkthrough, with my take on why each step earns its spot.
🔍 The three-prompt routine
The whole idea is to run three prompts in order, in the same chat. Each one builds on the output of the last, so don’t skip around.
- Open ChatGPT or Claude. Either works. The author designed the prompts to be tool-agnostic, so pick whichever you use most. If you use both, run the routine in each one, since their memory systems are completely separate.
- Send Prompt 1. This one asks the AI to list your personal details and any accessible memory it holds about you. The rationale is simple: you can’t protect what you don’t know exists. Most people have no idea how much context these tools have quietly accumulated from past chats, and this step surfaces all of it in one place.
- Send Prompt 2. Now the AI scans that list and flags a sensitive detail you probably shouldn’t be sharing. Think of it as a second opinion on your own data hygiene. The point is to identify the one thing that could actually hurt you if it leaked, rather than staring at a big list and shrugging.
- Send Prompt 3. This generates a copy-paste version of your information with identifying details removed. The creator recommends running this before sharing any conversation with anyone, because it produces a separate, cleaned version you can review and copy without touching the original.
That’s the whole thing. The post’s author says this makes it hard for your private chats to end up in a data pile-up or a Google search. I’d add that it takes maybe five minutes, which is a pretty good trade for knowing exactly what these tools hold on you.
🛡️ The step I’d bookmark: Prompt 3
The third prompt stood out to me the most. People share AI conversations all the time. A colleague wants to see how you got a result, a friend asks for the prompt you used, you paste a chat into a group thread. Every one of those shares carries whatever personal context leaked into the conversation along the way.
Prompt 3 solves that by creating a separate, scrubbed version. You review it, you copy it, and the original stays untouched. The expert highly recommends running it before sharing any conversation with anyone, and I think that’s the single most practical habit in the whole post. Make it a reflex: about to hit share? Run Prompt 3 first.
⚠️ One honest caveat
Temporary chats are still retained for safety. The steps above just reduce exposure. They don’t make your chats completely invisible to your company workspace.
I appreciate that the creator says this plainly instead of overselling. If you’re on a work account, your admin can often still see activity regardless of what you clean up on your end. The routine lowers your exposure and gives you a review checkpoint. It’s not an invisibility cloak, and anyone promising one is selling something.
🎁 Bonus prompt: turn off training
The author tacked on a fourth prompt that deserves more attention than a bonus label suggests. Most people have never checked whether their conversations feed model training. This prompt makes the AI walk you through the exact setting, using current official documentation, for your specific app and plan. Fill in the brackets and paste it:
I use [ChatGPT / Claude] on [web / iPhone / Android / desktop], with a [plan name] account. Using current official documentation, show me how to turn off the use of my conversations for model training. Give me the exact setting to check and explain any feedback-related exceptions. Don’t claim you changed anything for me. Wait for me to confirm before continuing.
Two details in that prompt are smarter than they look. First, it asks for feedback-related exceptions, because giving a thumbs-up or thumbs-down on a response can route that chat into training even when the main toggle is off. Second, it tells the AI not to claim it changed anything. AI assistants love to announce a change they never made, and this line forces the tool to hand you the actual steps instead of pretending.
📝 Tips I picked up along the way
- Same chat, same order. The creator is explicit about this. Prompts 2 and 3 depend on what Prompt 1 surfaced.
- Do it in both tools. If you use ChatGPT and Claude, run the routine in each. Their memories don’t talk to each other.
- Repeat monthly. Memory keeps growing as you keep chatting. A quick rerun catches whatever slipped in since last time.
- Pair Prompt 3 with the training toggle. One handles what you share with people, the other handles what the company keeps.
The one who posted it closes with the same honesty they opened with: don’t bother with any of this if you’re fine with your boss reading your private ChatGPT chats. But if that thought makes you a little uneasy, run Prompt 1 tonight. All three prompts are in the full LinkedIn post, so head over there, grab them, and see what your AI has been holding onto. I’d bet it’s more than you think!