MeKat field guide
What to leave out of an AI prompt.
Before pasting information into an AI tool, decide what the task actually needs. A request to improve the tone of a message rarely needs someone’s full name, contact details or the entire conversation.
Separate the task from the identifying details
Here is a fictional, prepared example. The placeholders describe information someone might otherwise paste. The task is to write a clear schedule-change message; a phone number and home address do not help with that. The real name and internal meeting details may be unnecessary too.
Rewrite this message to [coworker’s full name]. Their personal number is [phone number]. They live at [home address]. Tell them our private team meeting has moved from Tuesday to Wednesday.
Write a friendly message to a coworker saying a team meeting has moved from [OLD DAY] to [NEW DAY]. Keep it under 50 words. Ask them to confirm whether the new day works. Leave the placeholders in the draft.
Fill in the details yourself afterward
Review the wording, then add the real details in your own message editor. This prepared draft keeps the useful writing task intact without sending identifying information to the AI tool.
Hi [NAME], our team meeting has moved from [OLD DAY] to [NEW DAY]. Please let me know whether the new day works for you. Thanks!
Review the whole thing you are sharing
Attachments and screenshots may contain more than the paragraph you want help with. Look for names in comments, email addresses in headers, visible notifications, account details and unrelated messages.
For a simple writing task, consider typing a short, generalized description instead of uploading a full document. Remove information before submitting it. Asking an AI tool to remove private details after uploading them means those details have already been sent.
Replacing a name does not necessarily make a document anonymous. A distinctive job, location, schedule or combination of events may still identify someone.
Check the service, not just a setting’s name
Providers and account types have different rules for storage, access, training and connected tools. Read the current information for the service you use. For work or school material, follow the organization’s approved process.
The UK’s National Cyber Security Centre advises keeping sensitive information out of prompts submitted to public AI services. Its guidance also emphasizes understanding the service’s terms and privacy arrangements.
A training preference is not a promise that information is never stored. For example, OpenAI explains that turning off model training does not remove ordinary conversations from chat history. Check the provider’s current documentation rather than assuming every tool works the same way.
Your turn
Try it yourself.
You want help drafting a reminder for a neighborhood cleanup. Your notes contain the activity, a proposed date, volunteers’ phone numbers, their home addresses and a private gate code. What should go into the prompt?
Show answer guidance
Describe the activity and request a reminder template. Use placeholders for the date and meeting point if needed. Leave out the phone list, home addresses and gate code; add necessary logistical details through the appropriate private channel afterward. Before submitting, ask: “Does this detail change the help I need?” If it does not, leave it out.
Sources and further reading
The worked examples are original, prepared exercises. These primary sources explain the underlying guidance:
- NCSC: public AI and sensitive information
Keep sensitive information out of public prompts and understand the service’s privacy arrangements.
- OpenAI: Data Controls FAQ
Training preferences and conversation history are separate controls.