When to Tell Participants You Used AI: A Practical Disclosure Framework

One of the most common questions we hear from nonprofit leaders: “Do we have to tell people we used AI?” The honest answer is that there is no single rule — disclosure norms are still forming, and most participants have not been asked what they expect. But that doesn’t mean every organization has to invent its own approach from scratch.

Here’s a three-tier framework that organizations can adopt as a working policy. It treats disclosure as a spectrum, not a binary, and it gives staff a clear rule to follow rather than asking them to make individual judgment calls under pressure.

Tier 1: Silent Use

No disclosure required. Use AI as you would use spell-check, search, or a calculator. The output is fully reviewed and edited by a human, the AI’s contribution doesn’t change the substance, and a reasonable participant wouldn’t expect to be told.

Examples. Spell and grammar checking. Generating subject-line options for an email you then write yourself. Summarizing internal meeting notes for your own use. Brainstorming program-name ideas. Cleaning up the formatting of an existing document.

Tier 2: Footer Disclosure

A standing disclosure in your site footer, your AI use page, or your annual report. The AI played a substantive role but a human reviewed and is responsible for the output, and the disclosure level fits the context.

Examples. First drafts of public-facing blog posts. First drafts of routine donor communications. AI-assisted translation of program materials, with bilingual staff review. Summaries of public reports for your newsletter. Standardized case-note formatting from staff-dictated content.

A simple working footer: “Some content on this site is drafted with the help of AI tools and reviewed by our team before publication. Our AI use policy is available at [link].”

Tier 3: Explicit Disclosure

A specific, in-the-moment disclosure to the affected person. The participant has a direct stake, the AI’s contribution affects something about them, or a reasonable person would want to know before continuing.

Examples. Anything that influences a decision about the participant (eligibility, placement, services, referrals). Anything that includes their personal story in fundraising or grant material. Chatbots or AI assistants the participant interacts with. AI-generated translations of legal or rights-bearing documents. Communications a reasonable person would assume came from a specific staff member.

The Test Questions

When staff aren’t sure which tier applies, four questions usually resolve it.

Does this affect a decision about the participant? If yes, Tier 3.
Could a participant feel deceived if they later learned AI was involved? If yes, Tier 3.
Is the AI doing something a reader would assume only a human did? If yes, Tier 2 minimum, often Tier 3.
Is a human reviewing every output before it leaves the building? If no, you have a process problem, not a disclosure problem. Fix the process first.

What This Framework Doesn’t Cover

This is an ethics and trust framework, not a legal framework. State, federal, and funder rules may impose additional disclosure obligations — particularly around AI use in eligibility determinations, automated decisions affecting protected classes, and the use of participant data in AI training. A disclosure tier does not substitute for compliance review where it applies.

It also doesn’t cover staff-internal questions like attribution of work product, performance evaluation, or grant-narrative authorship. Those need their own organizational norms.

Putting It Into Policy

The simplest implementation: paste the tiers and test questions into your AI use policy, add the standing footer to your site, and brief all staff in a single thirty-minute meeting. The next time someone asks “should we say we used AI for this?”, they have a tier to apply, not a personal call to make.

StrategicalAI.org