Before you share · A fictional practice exercise
What Does AI Need?
Before pasting a whole note into ChatGPT or another AI tool, decide which details actually help with the job.
Try a made-up customer update or delivery complaint. Keep, remove or generalise each detail and see the prompt change. You’ll learn when less information is enough, and when removing a fact makes the task harder.
No AI account needed. Nothing to upload. This is practice with supplied examples, not a service for cleaning your real documents.
All details below are made up. Work only with this example; there is no box for your real information.
Prepare a calm, short update explaining the current problem and the next planned step, without promising a fix.
- Explain what is still unresolved
- Keep the next step separate from a promised result
- Use placeholders for customer details
3. Choose what goes in
Keep includes the exact detail. Remove leaves it out. Generalise replaces it with broader wording. Start with a name or an unrelated detail, then try removing a fact the task needs.
Your task: Draft a customer update
The prompt you’re building
This preview is assembled from your choices. No AI has written or checked an answer.
Left out: Nothing yet. Try removing a detail that does not help the task.
Check what changed ↓4. Does it still do the job?
Information the task is missing
The selected details retain this case’s stated task requirements. That does not establish permission to share them.
What could still identify someone?
Names, addresses or reference details still included: Customer name, Case reference.
- Exact dates can help someone link this event to another record.
- The only overnight laboratory at a named location can identify a customer without their name.
Removing names does not guarantee anonymity. This exercise can explain selected clues; it cannot certify that a combination of details is anonymous.
Extra detail still included without helping this task: Distinctive customer detail, Unrelated personal note.
Try a different purpose in step 2 without changing your choices. Watch what happens to the timing feedback. Removing everything is not the goal; keeping enough for the particular task is.
Take the habit with you
Before using a real tool with real information, check:
- Is this tool and destination approved for this information? At work, check the organisation’s rules and authorised recipients.
- What does this particular service do with inputs, including retention, access and training settings? Check the actual product and account settings.
- Can I do the task with less information, with placeholders filled directly at the destination, or without AI?
- Removing names does not guarantee anonymity. Dates, places and unusual details may still identify someone together.
Choices last only for this page session; refreshing starts again. Copy or download deliberately to keep the card. The card contains the selected fictional prompt and the checks above. No real-data intake or model call is part of this exercise.
About this case and the guidance
Fictional case. All names, references, places and dates are hypothetical teaching details. Founder Reality editorial exercise. Case version 1.0, reviewed 2026-10-12. These are constructed examples, not accounts of George’s experience or tests of a particular AI product.
The generalisations and feedback are authored teaching choices. They do not grant employer permission, certify privacy, establish claim eligibility or provide legal advice. The underlying information may still be confidential even when names are removed.
- NCSC: the hidden risks of shadow AI. Check the service and approved organisational arrangements before sharing work information.
- ICO: data minimisation. Enough information for a defined purpose, without unnecessary detail. Its guidance is under review; this is a design principle, not a compliance verdict.
- Google PAIR: data collection and evaluation. Combinations of apparently ordinary details can reveal personal information.
Case history: version 1.0 adds these two fictional records and purpose-specific feedback. Revisit when a case, explanation, source or accessibility issue changes; the review date does not make the hypothetical details current facts.