Keeping client information safe when staff use ChatGPT and similar tools
A plain-language guide for staff on what to keep out of an AI tool, such as names, contact details, case notes, health or financial information, passwords and unpublished plans, and how to ask for what they need using made-up examples or removed details instead. It also covers checking the privacy and data settings of any tool before using it, and writing down which tools are approved. The same ideas work for a small business that holds customer records or confidential plans.
What it is
This is a plain-language guide for staff on what to keep out of AI tools. It covers names, case notes, health details and passwords, and how to ask for help without them.
Who it is for
Nonprofits, and small businesses that hold customer records, can use it to build safe habits early.
What you get
You get written guidelines, a short training plan and a list of approved tools.
A real-life example
A small nonprofit's staff want to use AI, but the director worries about client privacy. She builds a short guide with a helper. It lists what never goes into an AI tool, like names, case notes and passwords. It also shows how to ask a question without them. The guide is part of every new staff training.
This is a made-up scenario. It does not describe a real organization.
How it works
- You: List the kinds of information your group must protect.
- The AI: Drafts a short, plain guide for staff.
- The AI: Adds examples of safe and unsafe questions.
- The AI: Suggests a short training plan and a list of approved tools.
- You: Check the guide against your policies and your laws.
- You: Share it and review it with staff each year.
Time saved
Typical estimate
By hand: about 1 day.
With AI: about 2 hours, plus your review.
These are rough guesses, not measured results. Your own times will vary.
Sample output
Example Safe and unsafe example
Do not paste: "Maria Lopez, age 34, was evicted and has diabetes." Safe version: "Write a kind note to a client who lost housing and needs a health clinic referral. Do not use a name." Never share: names, addresses, health details, case notes or passwords.
Made-up names and numbers, shown only to give you an idea. Always read AI output before you use it.
What it involves
- Tools / helper type
- Written guidelines + a short staff training + an approved-tools list
- Rough effort
- Low (Low = an afternoon or less; Medium = a few days with some setup; High = a project)
Who should look it over
Rules for client, donor and member data come from your funders, your policies and the law, so have your leadership or an attorney review the guidelines. Read each tool's own privacy terms and settings yourself rather than assuming, and ask the vendor when anything is unclear. Remind staff that when in doubt, leave it out, and review the approved-tools list from time to time.
Related use cases
Drafting a first pass of a grant application from past reports
The helper reads your earlier reports and program descriptions.
The result: You get a draft to reshape, so you start with words instead of a blank page.
Read more about Drafting a first pass of a grant application from past reportsWriting donor thank-you notes with a personal touch
You give the helper a donor's giving history and notes on what that person cares about.
The result: You get drafts that a staff member edits and signs.
Read more about Writing donor thank-you notes with a personal touchOrganizing volunteer shift reminders and thank-yous
The helper reads the shift schedule.
The result: You get a batch of reminders and thank-yous to review and send.
Read more about Organizing volunteer shift reminders and thank-yousExamples are for general information. Tools and features change, so check with the vendor, and keep a person in the loop for anything involving money, private information or people in your care.