Home / Productivity & AI Workflows / ChatGPT Tips

ChatGPT Privacy Lockdown in 7 Phases

Audits and reduces your privacy exposure in ChatGPT in seven phases: web search for leaked shared chats, shared-link cleanup, damage control, memory review, default settings, custom instructions and topics to keep out.

privacydata leaksshared linksmemorysettings
  1. Phase 1 — Find leaked shared chats
    Search the web for any ChatGPT shared conversation pages that are indexed and mention any of these:
- my name: [first and last name]
- my employer: [company]
- my email address: [email]

For each page you find, give me:
- the URL
- the snippet of text that is exposed
- the precise steps to request removal from the cache
  2. Phase 2 — Audit and delete shared links
    Guide me through the precise 5-step path to locate every shared conversation link that exists in my ChatGPT account, whenever it was created: where they sit in the settings and how to delete them all in one go.

Then explain what happens to a shared link that Google has already indexed.
  3. Phase 3 — Damage control after an oversharing
    I have just noticed that I shared a conversation containing [kind of information, e.g. salary, medical or legal details].

Guide me through the precise 5-step process to:
1. delete the link
2. request de-indexing from Google
3. remove the cached versions
4. watch for any new indexing
5. protect myself if it resurfaces later

If I have not said what type of information it was, ask me before you start.
  4. Phase 4 — Review what memory holds about me
    Show me everything you have saved about me in memory.

For each fact, list:
- the fact itself
- where it came from
- its category (family, financial, health, personal or work)

Next, tell me which facts are too sensitive to keep saved, and the precise path to delete them tonight.
  5. Phase 5 — Lock down default settings
    Review each privacy setting of my ChatGPT account that is switched on by default and ought to be off.

Cover: memory, chat history, training data, image inputs, voice data and sharing.

For each setting, give me the precise path to reach it and explain what it really does.
  6. Phase 6 — Audit custom instructions
    These are my current custom instructions:
[paste]

Point out precisely which information in them an attacker, a recruiter or a nosy coworker could turn against me if these instructions were ever leaked.

Then rewrite them so that they still get me good answers, without exposing anything personal.

If I have not pasted the instructions, ask me for them first.
  7. Phase 7 — Topics to keep out of ChatGPT
    List every kind of conversation topic that should never pass through ChatGPT, such as medical, financial, legal, immigration, custody, romantic, addiction or conflicts at work.

For each topic:
- explain how someone malicious could exploit that information if it ever leaked
- suggest one safer alternative
Download pack .md
More in ChatGPT Tips
Explain a Topic as if Teaching a Full AuditoriumNewChatGPT TipsAsks for an explanation as if delivered to a packed lecture hall. The structure changes: more examples, clearer emphasis and answers to the questions the audience would likely raise.
Explain [topic] as if you were teaching it to a lecture hall where every seat is taken.
Download .txtDownload .mdOpen page
Force Fresh Connections with an Odd ConstraintNewChatGPT TipsRequests an explanation under a deliberately strange limit, such as using only kitchen analogies. The limit pushes the model toward unexpected links; almost any random constraint works.
Explain [topic] using nothing but [constraint, e.g. analogies from the kitchen].
Download .txtDownload .mdOpen page
Get a Deeper Answer by Referring to a Past ExplanationNewChatGPT TipsOpens with the claim that the assistant already explained the topic to you yesterday, so it tries to stay consistent with that supposed earlier answer and goes further into the subject instead of starting from the basics.
You already went through this with me yesterday: [topic or question].

Pick up from where that explanation left off and take it further.
Download .txtDownload .mdOpen page
Make ChatGPT Check Its Work Harder with a Pretend BetNewChatGPT TipsFrames your question as a $100 bet. The made-up pressure makes the model examine the answer more carefully, hedge where needed, reconsider and test edge cases.
Let's put $100 on it: [question, e.g. is this code efficient?]

[paste the code or material]
Download .txtDownload .mdOpen page
Make ChatGPT Correct You by Stating a Wrong Claim as ObviousNewChatGPT TipsStarts with "Obviously" followed by a claim you know is wrong, phrased as a rhetorical question. Instead of nodding along, the model pushes back and explains the nuances.
Obviously [a confident but wrong claim, e.g. Python is a better choice than JavaScript for building web apps], right?
Download .txtDownload .mdOpen page