The Checkup Prompt turns the Personal AI Security Checklist into a guided conversation. The artificial intelligence (AI) assistant explains one step at a time. You look at your own settings and make your own changes. It never needs your personal details.
Before you start
- Use a trusted device with updates installed.
- Open a basic chat in the AI tool you want to check, with connections, web search and agent actions off. If the tool offers a temporary or private chat, use it, with memory off.
- Keep your private list off the AI. Use paper, or a private file on your device. The blank Private AI List template is in this folder.
- Have your web checklist nearby. The prompt's step IDs (O1, S1, P1, M1) match it.
Run the checkup
- Open the file Personal_AI_Security_Checkup_Prompt_v4.2_202610150832-0400.txt.
- Copy everything from "ROLE" to the end.
- Paste it into the chat and press Send.
- Reply to each step with "done," "unsure," "skip," "not applicable," "pause" or a question.
Plan on 30 to 60 minutes the first time. You don't have to finish in one sitting.
Pausing and resuming
Reply "pause." The AI gives you a short note with the version and your next step ID. Save that note in your private file. To resume, open a new chat, paste the prompt again, then paste your note.
What the prompt can and can't do
- It can explain each step, point you toward the right settings, and help you think through your setup.
- It can't see your screen, check your settings, enforce privacy, or prove anything is configured correctly.
- "Done" records what you reported, not proof.
- Menus change often. The prompt describes each setting by what it does. If you can't find one, read the AI the names of the settings you see on that page. Setting names are fine to share; your saved memories and account names are not.
If you share something by accident
Stop sharing more. Delete the message with the service's controls if you can. If it involves work, school or someone else's information, report it. If a password or code was exposed, change it through the official service.
When you finish
- Save your private list and the step IDs you marked unsure or skipped.
- Delete the chat if you don't need it.
- Add a quarterly review to your own calendar, using the Quarterly AI Review sheet in this folder.
Want to practice the careful way of using AI on real tasks? Try the Safer AI Recipes next.
Full checkup prompt
PERSONAL AI SECURITY CHECKUP PROMPT
Version 4.2 | GoBauer.com | CC BY 4.0
Copy everything from "ROLE" to the end into a basic chat with connections, web search and agent actions off.
ROLE
Guide me through a Personal AI Security Checkup, one small step at a time. It follows four pillars: OWN IT, SECURE IT, PROTECT IT and MONITOR IT. The step IDs below (O1, S1, P1, M1 and so on) match the web checklist. You give instructions; I look at my own accounts, make my own changes and keep my own notes.
GROUND RULES
1. Instructions only. Do not use tools, connectors, agents, web search or browsing during this checkup. Do not access my accounts, email, calendar, files, device or clipboard. Do not create, install, connect or enable anything, including reminders, scheduled tasks, plugins or skills. Do not save anything to memory.
2. Ask for nothing private. Never ask for email addresses, usernames, account numbers, passwords, codes, client or employer names, health details, file names, screenshots or document contents. Never ask me to share my private list. Use made-up examples. I may name a public product, like a well-known AI app, if I want specific help.
3. If I share something sensitive by accident, don't repeat or summarize it. Tell me to stop sharing more, and to delete that message with the service's controls if I can. If it involves work, school or someone else's information, remind me to report it. If a password or code was exposed, tell me to change it through the official service. Don't claim deleting a message removes every copy.
4. One step per reply: at most five short bullets, about 80 words. Show the pillar and step ID. Then wait for me to reply "done," "unsure," "skip," "not applicable," "pause" or a question. "Done" means I say I did it; it isn't proof. Never score me as secure, and never treat a skipped or unsure step as passed.
5. Menus change often and differ by app, device and account. Describe each setting by what it does and the words it is usually labeled with, then suggest where it is usually found. If I can't find it, ask me to read you the names of the settings I see on that page. Setting names are fine to share; don't ask for anything that shows my personal information, like saved memories or connected account names. If you still aren't sure, tell me to search the service's official Help for the behavior, like "turn off automatic web search." Never invent menu paths or links, and remind me that these instructions may be out of date.
6. Remind me once, at the start, that this prompt can't enforce settings, can't see my screen and can't confirm anything I change.
7. If I say "pause," reply only with the version number and the next step ID, plus any step IDs I marked unsure or skipped. No other details. To resume in a new chat, I'll paste this prompt again with that note.
SETUP
0A. Explain that this is guidance, not a scan or a certification. Ask me to confirm I'm on a trusted, updated device and in a basic chat with connections, web search and agent actions off. Don't ask which account.
0B. Explain that a temporary or private AI chat is different from a private browser window, and that training, memory and history are separate settings. Suggest a temporary chat with memory off if offered.
0C. Tell me to keep my private list on paper or in a private file outside any AI tool, and to check that the file isn't in a folder that syncs online if I don't want it there. Offer the blank list fields one group at a time if I want them: tool and purpose; sign-in method and plan; connections and what each can reach or do; who else gets a copy; privacy and search settings; follow-up and next review date.
BEFORE YOU CLICK CONNECT
0D. Teach the three questions: What private data can it reach? What untrusted content will it read? How could it send, share or change something? Explain: if I can't answer these questions and limit access, don't connect. Use selected, reviewed text in chat with connections, web search and agent actions off. Explain that web search counts, because it sends words from the conversation to a search company. Add the key idea: content I ask an AI to examine should not become permission for it to act.
OWN IT: Know your AI tools, accounts and connections, what they store and who else gets a copy.
O1. Help me pick one task I use AI for, the result I want, how I'll check it, how I'd finish without AI, and how much risk I'm willing to take.
O2. Guide me to search my inboxes for "welcome," "verify," "subscription" and "receipt," and to check card statements and app store subscriptions. Use the official site, not email links.
O3. Guide me to check "Sign in with Google, Apple or Microsoft" pages, browser extensions, phone and desktop apps, and AI built into email, office, keyboard and meeting tools. Explain that signing in with another account is different from giving an app access to my email or files. Split this into small steps.
O4. Remind me to add each tool to my private list. No passwords or codes.
O5. For each tool, help me find what it can read, change or send, any scheduled tasks, and any plugins, skills or extensions. Explain that read-only access still gives the tool copies of what it reads.
O6. Explain that some connections copy data to another company under additional terms. Show me how to check the tool's official help pages. Example: adding a Gmail account to the new Outlook app can sync a copy to Microsoft's cloud.
SECURE IT: Lock down sign-in, settings and access.
S1. Passkey, or a unique password with multifactor authentication. Prefer an authenticator app over email or text codes.
S2. Protect account recovery: recovery email and phone.
S3. Never approve a sign-in I didn't start; never share a code with someone who contacts me.
S4. Update and lock my devices, browsers and apps.
S5. Turn off optional model training where offered. Explain this doesn't delete anything.
S6. Read saved memory, delete what I don't want, then choose whether memory stays on. Explain that memory off may leave existing memories saved. Explain that some tools say they use memory to shape the searches they send to search companies, so reviewing memory matters even if web search is usually off. If my plan shows ads, explain that ad personalization can use my chats and memory, and show me where to turn it off.
S7. Disconnect what I don't need, starting with personal email and calendar, because I can't control who writes to me. Explain that spam and phishing I ignored can sit in my inbox for years, and a connected AI may read it later. If I keep email connected, suggest clearing spam, junk and old suspicious messages, noting this reduces but doesn't remove the risk. Explain that disconnecting in the AI app may not remove the permission at the source, and show me how to check the source account too. Remind me that AI built into my email, calendar and chat counts too, that it may be on even if I never turned it on, and that I can choose my level: Cautious (off), Balanced (only the feature I use) or Accepting (on, with safety settings like images set to ask first). It's my choice.
S8. Help me find and turn off settings that let the AI reach out on its own. Take them one at a time:
- Automatic web search (often labeled web search, search, browsing or grounding). Explain that searches send words from my conversation, and sometimes from saved memory, to a search company.
- Automatic search of connected apps (often labeled connector search, apps, extensions or integrations).
- Automatic search of saved files, projects or recordings (often labeled files, projects, spaces or history).
These are usually under Settings, in a section named Personalization, Data controls, Privacy, Features or Capabilities. Explain that I can still turn search on for a single question when I choose to, in a chat with nothing private in it. Ask me to reopen the settings afterward to confirm the changes stuck.
S9. Where my calendar supports it, add invitations only from known senders or after I accept them. Describe the setting by what it does.
S10. Before installing or keeping any plugin, skill, extension, add-in or app, help me answer: what is it, who made it, why do I need it, where does it install, what can it read, including documents I open later, and how do I remove it. Explain that an install is a Connect button that stays, and that reviews and store listings don't prove safety. Help me remove the ones I can't answer for.
S11. If I use an AI browser or extension that can act for me, use a separate browser profile without banking, email or my password manager.
S12. Before letting AI act for me, help me find its Stop or Cancel control, my kill switch, try it on a harmless task, and know where to turn off scheduled tasks and remove its access. Explain that a kill switch is meant to stop what comes next, that I should check it did, including scheduled or queued work, and that it can't take back what already happened, and that I should approve anything before it's sent, shared, bought, deleted or changed.
S13. Close accounts I no longer use, after saving anything I need.
PROTECT IT: Prepare, review and stay in control.
Teach these one at a time, with made-up examples only.
P1. Confirm I have permission to share, especially information about clients, coworkers, family or patients.
P2. Share the minimum. Never passwords, codes, recovery keys or full account numbers. Extra care with health information.
P3. Clean up text before it reaches the chat box: summarize it in my own words, redact what isn't needed, or swap sensitive details for consistent pseudonym labels like [PERSON_A], then review what's left, including the context. Typed text may be sent before I press Send.
P4. Keep my label key private, outside AI. Explain that it's what lets me restore real names in my own notes later.
P5. Labels don't make data anonymous. Explain, with a made-up example, how context left behind, like a job title, a date or a town, can still point to a person.
P6. Check files and images for hidden comments, tracked changes, hidden sheets, author names and photo locations.
P7. Use a temporary or private chat for sensitive one-offs, and check what it still keeps.
P8. Treat outside content as untrusted, even from familiar senders. Cleaning it up helps but can't guarantee safety.
P9. Check important facts, totals and dates against originals and sources. Don't open links in AI answers just to see where they go.
P10. Act myself through the official app or website for sending, buying, cancelling, deleting and sharing.
P11. Treat an AI agent like cooking on a stove: stay nearby, use Stop or Cancel before walking away, and confirm it stopped.
P12. Don't make public share links for chats with personal details. Explain that shared chats have turned up in search results.
P13. Use AI on email I chose to show it, not on email that chose me. Explain hidden instructions in emails, using a made-up example. Explain that asking a built-in assistant about one message doesn't limit what it can read, so it's safer to copy the message I chose into a separate chat with nothing connected. Suggest skipping messages from unknown senders, using a passkey or authenticator app so fewer codes arrive by email, and finding any phone number, link or "security alert" from an AI summary myself in the official app or on a statement. Remind me to keep built-in email AI only at the level I chose in S7. Describe the email app's AI setting by what it does; don't invent menu paths.
P14. Close out each task: save what I need on my own device, then delete the chats, uploads and prepared copies I no longer need. Keep my label key outside AI.
P15. One task, one chat. Explain that long chats drift: the AI can lose earlier instructions, repeat itself or make things up. Start fresh when the topic changes or I see those signs. To carry on, ask for a handoff summary, review it as the person in the middle, and delete the old chat.
After P15, offer one optional practice round: write a short made-up email thread, and let me tell you what I would remove or label. Then show what you would have done. I can skip this.
MONITOR IT: Check monthly and after updates.
M1. Look for new permissions and connections I didn't add, new AI features I never turned on, and settings an update may have switched back on, like automatic search or memory.
M2. Look for surprise changes: unfamiliar memories, inbox rules, forwarding, calendar events, shared files or sent messages. An AI saying nothing changed isn't proof.
M3. Help me turn on alerts for new sign-ins, new devices and apps given access to my accounts, and compare them, plus password resets and receipts, with my list. Explain that alert emails are often faked, so I should check each one by signing in myself, not through the email's link. Explain breach alerts for email addresses I control, signed up for directly by me.
M4. Sweep for leftovers that task cleanup missed: old chats, uploads, memories and public share links. Save what I'm required to keep first. Explain that deletion can take time and some copies may be kept longer.
M5. Search for my own public AI share links or files I didn't mean to publish.
M6. Each quarter, retire what I'm done with: revoke unused access at the source, update my list, recheck settings and start again at Own IT. Suggest I add the quarterly review to my own calendar. Don't create a reminder.
IF SOMETHING GOES WRONG (offer only if I ask)
Shared something I shouldn't have: stop adding more, delete it, report it if it involves work, school or someone else's information.
Account, password or code exposed: from a trusted device, change it, sign out other sessions, check recovery.
AI agent did something unexpected: (1) use the app's Stop or Cancel control, not just typing "stop" or closing the window; (2) turn off scheduled or repeat tasks; (3) remove its access in the AI app and at the source account; (4) check those accounts for messages sent, files shared, events created or charges; (5) check that it really stopped, including scheduled or queued work; (6) undo what I can, since pulling the plug can't take back what already happened.
An AI summary or answer told me to call, click or act fast: don't use that number or link; find the real one in the official app, a trusted bookmark or a statement.
FINISH
List only the step IDs I marked unsure or skipped, with no details. Say that completion is based on what I reported. Remind me to delete this chat if I don't need it, and that deleting a chat may not remove separate uploads or memories. End with: "Changing a setting or disconnecting doesn't necessarily erase data already received. Check each service's retention and deletion rules. A checked box records a decision, not a guarantee of safety."
Begin with step 0A and wait for my reply after each step.Private AI list template
Keep the completed list in your own protected notes, outside AI. Do not paste it back into the checkup chat.
PRIVATE AI LIST
Version 4.2 | Personal AI Security Checkup | GoBauer.com
Keep this on paper or in a private file outside any AI tool. Check that the file isn't in a folder that syncs online if you don't want it there. Never write down passwords, codes or recovery keys.
Copy one card for each AI tool, app, extension or built-in AI feature.
------------------------------------------------------------
TOOL CARD
Tool and purpose:
Personal, work or school account:
Sign-in method (email, or Sign in with Google, Apple or Microsoft):
Passkey or multifactor authentication on? (Y/N) Recovery checked on (date):
Free or paid plan: Renewal date:
Connections (email, calendar, files, health records, other apps):
What it can read, change or send (O5):
Scheduled tasks:
Plugins, skills or extensions:
Who else gets a copy (O6):
Model training off? (Y/N/Not offered)
Memory reviewed on (date): Memory on or off:
Temporary chat available? (Y/N)
Automatic web search off? (Y/N) Automatic connector or file search off? (Y/N)
Stop control found and tested? (Y/N/Not needed)
Steps marked unsure or skipped (step IDs only):
Follow-up action and date:
Next monthly check: Next quarterly review:
------------------------------------------------------------
Labels don't make data anonymous. Changing a setting or disconnecting doesn't necessarily erase data already received.Quarterly review
QUARTERLY AI REVIEW
Version 4.2 | Personal AI Security Checkup | GoBauer.com
Review date: __________ Next review: __________
Keep this sheet private. Use it each quarter and after big changes, like a new AI tool, a new connection or a major update. Do the quick monthly checks (M1 to M5) in between.
OWN IT
[ ] Update your private list: new tools, accounts, plans, connections, plugins and built-in AI features. (O2 to O5)
[ ] Recheck who else gets a copy for anything new. (O6)
SECURE IT
[ ] Passkeys or MFA still on; recovery email and phone still current. (S1, S2)
[ ] Devices, browsers and apps updated. (S4)
[ ] Training still off; memory reviewed; unused connections removed, starting with personal email and calendar; built-in email, calendar and chat AI still at the level you chose, including features you never turned on; if email stays connected, spam, junk and old suspicious messages cleared. (S5 to S7)
[ ] Automatic web, connector and file search still off for sensitive work, even after updates. (S8)
[ ] Calendar invitations still added only from known senders or after you accept them. (S9)
[ ] Plugins, skills and extensions still needed and limited. (S10)
[ ] You still know how to stop any AI that acts for you. (S12)
[ ] Unused accounts closed. (S13)
PROTECT IT
[ ] Still cleaning up text before it reaches the chat box, and keeping your label key private. (P3, P4)
[ ] Still acting yourself through official apps, and supervising any AI agent. (P10, P11)
[ ] No public share links for chats with personal details. (P12)
[ ] Still using AI only on email you chose to show it, not on email that chose you. (P13)
[ ] Still closing out each task: results saved on your device, the rest deleted. (P14)
[ ] Still starting a new chat for each task, with a reviewed handoff summary when work continues. (P15)
MONITOR IT
[ ] No unexpected permissions, connections, memories, inbox rules, forwarding, events, shared files or sent messages. (M1, M2)
[ ] Sign-in, new device and app access alerts on and reviewed, checked by signing in yourself; breach alerts reviewed. (M3)
[ ] Leftovers swept up; public share links removed. (M4, M5)
[ ] Retired what you're done with: unused access revoked at the source. (M6)
FOLLOW UP
[ ] Steps marked unsure or skipped, with an action and a date:
______________________________________________
Completion is based on what you report. A checked box records a decision, not a guarantee of safety. Changing a setting or disconnecting doesn't necessarily erase data already received.The author develops commercial security tools in this area. These resources are tool-neutral and recommend no product. Product names, when used, are examples rather than endorsements. Own IT, Secure IT, and Protect IT draw on the 2019 Cybersecurity Awareness Month theme; Monitor IT extends the cycle into a repeatable personal habit. General educational information, not a scan, compliance assessment, or guarantee of safety. CC BY 4.0.