Last verified: October 2026.
This is general information, not legal advice. Privacy and data laws differ between countries, and between US states and Canadian provinces. Check the rules where you work, and talk to a lawyer before you rely on this for anything with legal risk.
Before staff put anything into an AI tool, it helps to know what kind of information it is. We suggest sorting everything your organization holds into five tiers and agreeing what to do with each one. Staff then only need to answer one question before they paste or upload something: which tier is this?
The fourth tier only applies to organizations covered by the EU’s General Data Protection Regulation (GDPR) or the UK’s version of it. That usually means you are based in Europe or hold data about people there.
| Tier | Examples | What we recommend |
|---|---|---|
| 1. Public information | Your website, published reports, press releases, social posts | Use it in any AI tool |
| 2. Private information that is legal to share with AI | Budgets, financial statements, fundraising plans, board papers, with personal details removed | Use it on an organizational plan, and delete the chat when you are done |
| 3. Personal data about supporters | Donor records, volunteer records, email subscriber lists | Don’t use it, or anonymize it before use |
| 4. Personal data covered by GDPR | Records about supporters, staff, or volunteers in the EU or UK | Get the legal basics in place before any of it goes in |
| 5. Legally sensitive material | Details of undercover investigations, confidential sources, legal strategy, health information | Keep it out of AI tools |
Tier 1: Public information
Anything already published, or written to be published, can go into any AI tool. That covers your website, published reports, press releases, newsletters, and social posts. A lot of the everyday value of AI sits here. You could paste last year’s annual report into ChatGPT and ask for a month of social posts, or ask Claude to turn a long report into a one-page summary for donors.
Tier 2: Private information that is legal to share with AI
This tier covers information you would not want made public, but where nothing in the law stops you from putting it into an AI tool. Budgets, financial statements, fundraising plans, draft strategy, and board papers usually sit here, once any names and other personal details are taken out.
We think it is fine to use this kind of information in AI, as long as you use an organizational plan rather than personal accounts. Business plans from the main AI companies don’t train on your data by default, and our training FAQ explains how to check yours. Much of the risk comes from how staff use AI, which our guide to using AI safely at work covers.
Some organizations still don’t trust AI companies with this kind of information and keep it out of AI tools. We respect that choice, though we don’t think putting it in is wrong in itself.
One habit makes this tier safer: delete the chat once you have finished. When you delete a chat in ChatGPT or Claude, it disappears from your history straight away, and the company permanently deletes it from its systems within 30 days. There are some exceptions, for example where a company has to keep a chat for legal or safety reasons. Gemini works differently and can take longer. Our guide to law enforcement access has a table of each company’s rules.
You can also skip the deleting step by starting a temporary chat in ChatGPT or an incognito chat in Claude. Neither appears in your history or is used for training. OpenAI keeps a copy for up to 30 days for safety reasons, and Anthropic keeps one for 30 days by default. Whichever route you take, don’t rate these chats with a thumbs up or thumbs down, because a rated conversation can be kept for longer, even after you delete it.
One exception to deleting: once your organization learns of an investigation, a lawsuit, or a request to preserve records, stop deleting anything that could be relevant and get legal advice. Deleting records at that point can count as destroying evidence.
Tier 3: Personal data about supporters
This tier covers information that identifies people: names, email addresses, phone numbers, home addresses, donation histories, and notes about volunteers. Whether you are allowed to put it into an AI tool depends on the privacy laws where you work and on what your own privacy policy promises supporters.
Even where it is allowed, we recommend that you either don’t use it or anonymize it before you use it. We say this because of the risk to your reputation. Many donors would be upset to learn that their details had gone into an AI tool, even if doing so was legal, secure, and low risk.
Most tasks don’t need to know who anyone is. AI can spot which donors are likely to stop giving from donation amounts and dates alone. Before you export a spreadsheet, delete the name, email, phone, and address columns and add a simple ID column (Donor 1, Donor 2, and so on). Keep the version that links IDs to people in your own systems, so you can act on what the AI finds.
Removing names is not always enough. A rare job title, a small town, or an unusual donation amount can still point to one person, so remove any detail the task doesn’t need. Do this in your spreadsheet or database. If you ask a cloud AI tool to do it, you have already shared the original with the AI company.
If a task needs to know who people are, such as drafting personal thank-you notes, check your privacy policy and the law first.
Tier 4: Personal data covered by GDPR
GDPR applies to organizations based in the EU, and the UK has its own near-identical version. GDPR can also apply to organizations elsewhere, including in the US and Canada, if they offer goods or services, paid or free, to people in the EU, or monitor their behavior. If you have supporters in Europe, check whether it covers you. The European Commission’s guide explains who it applies to in plain language.
You can use AI with personal data under GDPR, but you need at least these three things in place first:
- A lawful reason. GDPR calls this a “lawful basis” for using the data this way, and your privacy notice should tell people about it.
- A data processing agreement with the AI company. This is a contract that sets out how the company handles personal data on your behalf. When you sign up for a business plan with one of the main AI companies, this is put in place for you automatically. OpenAI, Anthropic, and Google (for Gemini on a Google Workspace business plan) all build it into their business terms.
- A legal route for sending data outside Europe. The main AI companies usually process data in the US, and GDPR has rules for that.
Anonymizing data, as in tier 3, helps, but it doesn’t always take data outside GDPR. If anyone at your organization can link the IDs back to people, the data usually still counts as personal data. Only data that can no longer identify anyone falls outside the law. The details depend on your situation.
Tier 5: Legally sensitive material
Some advocacy work carries legal risk for your organization or the people you work with. Details of undercover investigations, the identities of confidential sources, and your legal strategy usually belong here, along with anything that could become evidence in a criminal case or lawsuit. So does health information about anyone, and the other data GDPR treats as especially sensitive, such as religion, ethnicity, and sexual orientation.
Keep this tier out of AI tools, and tell staff so clearly. AI companies hand over conversation content to law enforcement when they receive valid legal process, such as a subpoena or a warrant, and OpenAI and Anthropic both report having done so. A chat can also be kept after you delete it, for example if a safety system flagged it. Our guide to law enforcement access explains the risk in more detail, including what it means for AI meeting recorders.
Keeping material out of AI means choosing not to create a new record. It never means destroying records you already have. If your organization learns of an investigation, a lawsuit, or a request to preserve records, stop deleting anything relevant and get legal advice.
If you have a strong reason to use AI on this kind of material, get specialist advice first. Some setups, such as AI models that run entirely on your own computer, keep data off AI companies’ servers, but they take technical effort. Our data security guide explains the tradeoffs.
Adding the tiers to your AI policy
The tiers work best with your own examples. List three or four things your organization holds under each tier, name one person staff can ask when they are unsure, and add the result to your AI policy. Our AI policy worksheet has a section for data rules.
Where to go next
- Should you worry about law enforcement accessing your AI conversations? covers what AI companies hand over, how deletion works at each one, and what it means for meeting recorders.
- Is our data used to train AI models? explains how to check your plan’s training settings.
- Data security and privacy basics for AI tools covers what to check before you trust a vendor, connectors, and local models.
- Use AI safely at work is a guide for staff on what to share, what to connect, and how to check AI’s work.

