This lesson shows you how to keep donor and client information safe when your staff use AI tools. You will sort your data into a few simple categories, choose a plan and settings that limit what the vendor keeps, and set some working habits that stop sensitive details from reaching the tool at all. It matters this week because staff are probably already pasting text into AI assistants, and a donor list or a case note can go in with one keystroke.

The idea: decide what goes where before anyone types

Most data problems with AI tools start with a well-meaning person in a hurry. Someone wants a faster thank-you letter or a tidier case summary, and the quickest route is to paste the whole record in. The fix is not to ban AI. Agree ahead of time which kinds of information may go into which tools, and set up the tools so that the information they do see is handled carefully.

Sort your data into categories

You do not need a complex scheme. Four categories cover most small organizations:

Category	What it includes	Can it go into an AI tool?
Public	Your website text, published reports, event announcements	Yes, in any approved tool
Internal	Meeting agendas, draft policies, staff schedules	Yes, in approved tools on your organization’s plan
Confidential	Grant budgets, partner contracts, board discussions	Only in approved tools with the right settings, and only when needed
Restricted	Donor records with gift amounts, client case notes, health details, payment information	No, unless the data has been carefully de-identified first

Donor and client data sits in the most protected category for a reason. People gave you that information because they trusted you with it. A leaked gift amount can embarrass a donor. A leaked case note can harm a client.

Know what counts as personal information

Personally identifiable information is any detail that identifies a person, either on its own or combined with other details. A name is the obvious one. Less obvious are the details that point to someone when you put them together.

Removing names is often not enough. Picture a case note that says “a 34-year-old woman in a small rural postcode, born in March, caring for a child with a rare condition.” There is no name, but someone who knows that area could probably work out who it is. A postcode, a date of birth and an unusual circumstance can be enough.

So if you want to use AI on records, you need to de-identify them carefully: remove or blur every detail that could point back to a person, not only the name. If you cannot do that with confidence, do not use AI on those records at all.

A worked example from a small organization

Here is an illustration, not a real case. A development officer at a small charity has 200 thank-you letters to send after a year-end appeal. Each letter should mention the donor’s name and their gift. The officer has heard that an AI assistant can write warm, varied letters quickly.

The first instinct is to export the donor list, paste it into the assistant and ask for 200 personalized letters. That would send every donor’s name, address and gift amount to the AI vendor. Depending on the plan, the vendor might keep that text for a while or use it to improve its models.

Instead, the officer does this:

Asks the AI assistant to write a thank-you letter template, describing the appeal and the tone wanted, with no donor details at all.
Asks for placeholders such as [First name], [Gift amount] and [Program supported] where the personal details will go.
Asks for 3 versions of the letter, so donors at different gift levels get slightly different wording.
Copies the finished templates into the organization’s existing donor system and uses its mail merge to fill in the names and amounts.

The result is the same: 200 personal letters, written faster than before. The difference is that the AI tool never saw a single name or amount. The donor data stayed in the system that was built to hold it.

The same pattern works for many tasks. Let the AI handle the words and the structure. Keep the people in your own systems.

Steps to take

Use the right plan and settings

Free and personal AI plans are often set up for individuals, not organizations. Look for a plan where:

Training on your content is off. The vendor should not use what your staff type to improve its models.
Retention is as short as the vendor allows. The less time the vendor keeps your conversations, the less there is to lose if something goes wrong.
An admin controls the settings for everyone. One trusted person should be able to set these choices for the whole organization, so they do not depend on each staff member finding the right setting on their own.

Our tools pages explain, in plain language, what different AI tools do with your data.

Get agreement before recording meetings

AI meeting recorders and note-takers capture everyone on the call, not only your staff. That includes donors, clients and partners. Before you record, tell everyone on the call that an AI tool will record and summarize the meeting, and get their agreement. If someone says no, turn it off. For conversations with clients about their circumstances, think hard about whether a recorder belongs there at all.

Connect only what the task needs

Many AI assistants can be connected to email, shared drives or a donor database. Once connected, the assistant can reach everything in that account or folder, not only the file you had in mind. If a staff member connects their whole inbox to get help with one newsletter, the assistant can now read every donor email in it.

Before connecting anything, ask: what does this task actually need? Connect that folder or that account, and nothing more. Disconnect it when the task is done.

Check your legal duties

If you have legal duties for personal data, they still apply when staff use AI. For example, if you hold data about people in the EU, the GDPR may apply to you. In that case:

Read the vendor’s data processing agreement, which sets out how the vendor handles personal data on your behalf.
Check the vendor’s data region, meaning where your data is stored and processed.
Take advice from someone qualified when you are unsure. Guessing about legal duties is not a safe plan.

Common mistakes

Assuming no names means no risk. As the case note example shows, a few ordinary details can identify a person.
Letting each person pick their own tool and settings. You end up with a mix of personal accounts, unknown retention and training switched on. Choose approved tools and have an admin set them up.
Recording calls without asking. Your staff may be comfortable with a note-taker, but your donors and clients did not agree to it.
Connecting everything for convenience. Linking a whole drive or inbox to save a few clicks gives the assistant far more than it needs.
Writing a policy and never mentioning it again. Staff remember habits, not documents. Show people the template-and-placeholder approach with a real task.
Treating the vendor’s paperwork as someone else’s job. If you have legal duties, the data processing agreement and data region are your business to check.

Try it this week

Set aside 15 minutes.

Pick one common task that touches donor or client data. Good candidates are thank-you letters, appeal emails, case summaries or meeting notes.
Write down, step by step, how a staff member would do it with an AI tool today.
Circle every point where a name, an amount or a case detail would reach the tool.
Rewrite the task so the AI tool never sees any of them. Use placeholders, ask for a template, or describe a made-up situation instead of a real one. Do the personal parts in your own systems afterwards.
Share the rewritten version with the person who does that task most often, and ask whether it would work for them.

When you have done one task, you have a pattern the rest of your team can copy.

Key takeaways

Sort your data into simple categories and treat donor records, client case notes, health details and payment information as restricted.
Removing names is often not enough, so de-identify carefully or keep those records out of AI tools entirely.
Use a plan where training on your content is off, retention is short and an admin controls settings for everyone.
Tell people and get agreement before recording meetings, and connect AI assistants only to what the task needs.
Let the AI write templates with placeholders, and fill in personal details in your own systems.

Protecting donor and client data when staff use AI