Share the minimum information needed for the task. Remove secrets and identifying details, and check the tool’s current data and privacy settings before sharing sensitive material.
Privacy choices depend on the tool, account and your organisation’s rules. A useful prompt often needs the situation and the goal rather than a full personal record or a confidential document.
A practical workflow.
Check permission first
Ask whether you are allowed to share the material with the service. Work, client, school or third-party information may have rules that apply even when you remove a name.
Use placeholders
Replace names, account numbers and other unnecessary identifiers. Keep the relationships and context the task needs without retaining details that identify someone.
Review files and screenshots
Look at the whole file, including headers, comments and visible background information. A screenshot can expose more than the button you wanted to discuss.
Check current settings
Read the service’s privacy guidance and your account’s available controls. Do not assume that a setting in one product applies to another tool or account type.
READY TO MAKE YOURS
Start with this prompt.
Help me make this situation safe to describe in a prompt: [non-sensitive summary]. List categories of information I should remove or replace with placeholders. Do not request passwords, verification codes, account numbers or confidential records. Preserve only the context needed for [task].
Replace the brackets. Leave out private details. Check the response.
ILLUSTRATIVE EXAMPLE
A clearer request.
A VAGUE STARTHere is the complete client record.
A CLEARER BRIEFHelp me draft a reply using this anonymised situation: [summary]. Names and identifiers are removed. Ask only for additional information essential to the wording.
The task retains useful context while reducing unnecessary disclosure.
Before you use the result.
- Passwords and verification codes are excluded.
- Sharing is allowed by the relevant rules.
- Unnecessary identifiers are removed.
- Current tool settings are reviewed.
A few useful answers.
Is replacing a name always enough?
No. Other details can identify a person or reveal confidential information. Review the full context and follow the rules that apply to the material.
Can I use AI for confidential work?
Check your organisation’s approved tools and policies first. The availability of a chat box does not establish permission to upload confidential material.
PUT THE GUIDE TO WORK
Keep the useful context and leave out the private details
A good prompt does not need every fact in the original document. It needs the facts that change the answer. Before sharing a message, screenshot or file, decide what the task requires and create a smaller version for the tool. These AI privacy prompts help you separate useful context from identifying detail and keep a record of what you chose to share.
Describe the task before deciding what to upload
If you want help writing a polite reminder, the model may need the agreed deadline and the action requested. It probably does not need an account number, a home address or the entire email thread. Start with a blank brief: purpose, relevant facts, constraints and desired output. Add only what changes those fields. If you cannot explain why a detail is needed, leave it out of the first version. A short fictional example can be enough to learn a technique; you can apply the technique to the real material privately.
Use placeholders consistently and review the whole file
Replace a person with [client A] everywhere, not with a different label in each paragraph. Preserve the relationship that matters, such as who requested the change and who must approve it. Remove unnecessary dates, contact details and identifiers from the material you intend to share. A screenshot can contain private notifications outside the area you meant to discuss. A document can contain comments or extra pages you did not need. Make and inspect a separate copy. Redaction is a human review task; a prompt asking for redaction still shares whatever you put into that prompt.
Settings and the content you share are separate decisions
Check the actual provider’s current data controls and your organisation’s rules before using it. For ChatGPT, OpenAI says that switching off “Improve the model for everyone” stops new conversations being used for model training; those chats may still remain in history. Temporary chats have their own handling, described in the official guide below. These controls do not make every kind of information appropriate to share. Decide whether you are allowed to use the material, whether a smaller example will do and how you will keep the useful result.
Open a card, copy the prompt and replace the brackets with your own details. Each card includes a follow-up and a check before you use the answer.
SITUATION 01 ↗
Find the minimum context for a task
Build a useful brief without starting with the whole document.
Open the prompt +
I need help with [task] and want [output]. Here is a non-private summary [summary]. List the minimum information needed to give a useful answer, the information that is optional and the details that do not appear necessary. Ask for roles or placeholders instead of personal identifiers. Do not request the full source document by default.
Refine the next draft
Using only these approved details [details], draft the task brief. Put any genuinely missing requirement in a question rather than filling it in.
Before you use it
Decide yourself whether sharing each requested detail is permitted. A model cannot approve your organisation’s data-handling rules.
SITUATION 02 ↗
Practise with a fictional example
Learn the editing method before using it on private material.
Open the prompt +
Create a clearly fictional practice example for [task type]. I want to practise [skill]. Use these invented constraints [constraints], and do not ask me to upload my real document. Show a rough version, a revised version and the reason for each important edit. Label the scenario fictional.
Refine the next draft
Give me a short checklist for applying the same editing method to my real material privately. Keep the checklist about the method, not about the invented names or numbers.
Before you use it
Do not present invented scenarios as real experience or evidence. The practice result is a learning aid.
SITUATION 03 ↗
Check an already redacted brief for clarity
Keep the relationships understandable after replacing names.
Open the prompt +
Review this manually redacted brief [text]. The role labels mean [non-identifying role key]. Check whether the labels are consistent and whether removing detail has made a necessary relationship unclear. Suggest clearer wording using the same placeholders. Do not reconstruct identities or ask for real names.
Refine the next draft
Rewrite the brief so the requested action, responsible role and deadline are clear. Preserve the facts supplied and list any ambiguity separately.
Before you use it
Manually inspect the version before sharing it. The model sees the text you submit, so do the initial privacy review beforehand.
SITUATION 04 ↗
Make a provider-check list
Turn a privacy concern into specific questions to verify.
Open the prompt +
I am considering [use] with material of this type [non-private category]. Using this provider documentation [official text], make a checklist of controls and conditions I should verify. Separate training use, saved history, sharing and any stated retention. Cite the supplied section for each statement and mark missing details unknown. Do not declare that the service is suitable for confidential material.
Refine the next draft
Which unanswered questions affect whether I should share this material at all? Write them for the relevant provider or my organisation’s authorised contact.
Before you use it
Read the current original documentation and follow the rules that apply to your account. A general summary does not settle a confidential-data decision.
The examples below are original, illustrative edits using fictional situations. They show the checks to make, rather than promising what an AI tool will produce.
Your minimum-context sharing brief
Task: [what I need help with] Output I want: [format / audience] Facts that change the answer: [relevant facts] Details the task does not need: [identifiers / extra history] Placeholder key kept privately: [label → real item] Material I will share: [small text / separate copy] Material I will keep out: [items] Permission or workplace rule checked: [rule / owner] Provider controls reviewed: [source / date] Final manual review: [what I checked] Where I will keep the approved result: [location]
WORKED EXAMPLE
A reminder needs a deadline, not an account number
An original fictional example: a freelancer wants help writing a reminder about feedback on a draft.
FIRST DRAFTHere is the entire client thread, their phone number, billing address and private project folder. Write a reminder.
AFTER CHECKINGHelp me write a brief reminder to [client A]. Feedback on [draft] was agreed for [date], and I need a decision on [specific section] before the next step. Use a polite tone and do not add a new deadline or commitment.
What changed
The smaller brief retains the task, agreed timing and dependency. The extra identifying details do not help write this reminder. The writer adds the real recipient privately after reviewing the wording.
WORKED EXAMPLE
Keep role labels consistent after redaction
An original fictional example: a draft mixes three different labels for the same reviewer.
FIRST DRAFTClient A approved the outline. The manager has not replied about the examples. That person needs to sign off tomorrow.
AFTER CHECKING[Reviewer A] approved the outline. Approval of the examples is still pending from [reviewer A]. I need confirmation of the review date before treating tomorrow as an agreed deadline.
What changed
The revised version uses one label for one role and separates an approved outline from pending examples. Redaction should remove unnecessary identity details without disguising who is responsible for the next decision.
Try it with your own task.
Try the minimum-context brief on a low-risk writing task. Create the version yourself before opening the tool, then compare how much original material you actually needed. Keep a private note of the facts you removed.
Check the official instructions.
For settings and safety guidance, use the original provider's current help. Menus and account options can change.
GO DEEPER / 1 FREE TEMPLATES
Prompts for this part of the task.
Use a template for the step you are working on. Add the required inputs, then check the result against your source material.
Browse the task-guide directory ↗
Use a small workflow that you can repeat.
- Use a small task you understand well enough to review.
- State the goal, context, limits and output format.
- Look for missing facts and uncertain claims before asking for more detail.
- Keep the useful parts, refine one instruction and compare the new result with the first.
Keep in mind: Never assume the tool can browse, see a file or run code just because it says it can. Ask what it actually used and verify important claims yourself.
A private-by-default task brief
Use placeholders before sending information.You get: Anonymised brief with unnecessary details removed+
COPY. PERSONALISE. CHECK.
Help me design a template for asking AI about [task]. Do not ask me to paste the real private information. List what kinds of context are necessary, which can be replaced by placeholders and which should be left out. Give a fictional example of a useful brief using made-up names and non-sensitive facts. Explain where I must follow my organisation’s rules.
Example input
A worker wants help drafting a message without exposing a customer’s account details.
Illustrative scenario, not a generated or verified AI result.
Then ask
Reduce the brief to the minimum information needed to draft a useful response.
Before using the answer
Check your tool settings and rules; removing a name alone may not anonymise the context.
Keep learning from original sources.
These sources inform the workflow and checks; the templates above are original LifestyleGlide prompts.
