Artificial intelligence (AI) is useful when you know the result you want and how you will check it. The supplied examples cover family planning, household tasks, learning, and other repeatable work. Start with general descriptions, not real records or account access.
Choose a starting point
- If you already know the steps, use Safer AI Recipes.
- If the task is unfamiliar, use the Guided Planner below.
- If you want to examine every checkpoint, use Plan an AI Task.
Before using any prompt, check the real settings. A request to avoid tools does not disable them. Keep connections, web search, browsing, and agent actions off for the prepared-input workflow. Decide whether a basic chat is appropriate for the information at all.
Guided Planner prompt
Copy and review this request. Describe the task in general terms. Keep names, records, addresses, exact family routines, credentials, and other private facts outside the planning conversation.
You're going to help me plan how to use AI safely for one task, following four pillars: Own IT, Secure IT, Protect IT and Monitor IT. Walk me through it like a friendly interview.
How to run this:
• Ask one question at a time, in plain words, with a short example answer. Wait for my answer.
• I'll describe my task in general words. Never ask for real names, addresses, account numbers, birth dates, logins or documents. If I share something like that, tell me and don't repeat it.
• Don't search the web, connect to anything or take any action.
• Suggest answers when I'm stuck, but I make the decisions.
Ask in this order:
OWN IT
1. What's the task, in one sentence? What result do I want at the end, and who will use it?
2. What are the steps, start to finish? For each step: where does the information come from, what goes in, what do I do, what comes out, and who uses it?
3. For each step: what must be true for it to be done right (requirements), and what could go wrong or change (exceptions)?
4. Which AI account will I use? Ask whether this task and its data belong in a personal, school or work account. For a personal task, suggest the account whose settings I've already checked; paying for a plan doesn't make it private or approved. Whose information is involved, and do I have permission to share it?
5. How private is the information (low, medium or high), and what could a wrong answer cost (low, medium or high)? Suggest the highest setup level with this rule: low and low, up to Level 3 (web search on purpose, nothing private in that chat); medium on either, Level 1 or 2 (basic chat with cleaned-up text); high on either, Level 1 only, or do it without AI.
SECURE IT
6. Help me check my settings by what they do: a temporary chat, training off, memory, web search and connected apps. Should web search be off for this task?
7. May the AI act, or only draft? Suggest draft only. Where is my Stop button, and when should I stop (stop conditions)?
PROTECT IT
8. For each step, what can I redact (leave out) or replace with pseudonym labels like [PERSON_A]? Build my "keep out of AI" list.
9. How will I check the result against the original, and which actions will I take myself?
10. When the task is done, what will I save on my own device, and what will I delete from the AI tool?
MONITOR IT
11. Which security alerts should I turn on, like new sign-ins, new devices and app access, and how will I check them safely, by signing in myself rather than through an email's link? When will I review memory, connections and settings again?
When we're done, give me:
A. A one-page summary in this order: task, account, how private, cost of a wrong answer, highest setup level, web search, act or draft, kill switch, stop conditions, cleanup, answers to the three questions (What private data can it reach? What untrusted content will it read? How could it send, share or change something?), and the keep-out list.
B. A table of the steps with these columns: step, suppliers, inputs, process, outputs, consumers, requirements, exceptions, AI can help with, keep out.
C. A Start request I can paste into a new chat to do the task, with the rules built in.
Use labels only, and keep everything short and easy to read. Start with question 1.Requests to use while doing the task
- Define the result: Explain the task, the useful output, and the checks it must pass.
- List missing information: Ask which facts are needed, then supply only the minimum you can share appropriately.
- Work from reviewed material: Label the source rows or messages and ask the AI to cite them.
- Expose uncertainty: Ask it to mark unknowns and assumptions rather than fill gaps.
- Check the response: Compare totals, dates, responsibilities, and important facts with the originals yourself.
- Prepare the handoff: Put real names back only in your own notes, review again, and act through the official service.
- Close out: Save the useful result, remove unnecessary copies, and review any access you granted.
A small worked example
Task: review recurring household charges. Prepare a table of service labels, amounts, currencies, billing periods, and renewal dates. Leave out card numbers, addresses, account identifiers, and unrelated purchases. Ask for monthly and annual totals, with unknown billing periods clearly marked. Recalculate the totals yourself and verify each subscription in the official service. You decide whether to cancel and make that change yourself.
For children's schedules, medical questions, serious allergies, or other high-consequence tasks, a confident response is not a sufficient check. Use a qualified person or the original authoritative source and keep the manual fallback available.
Keep going
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.