The "AI-Assisted" Disclosure: Where Nonprofits Should Label AI Use and Where They Shouldn't
A growing number of nonprofits have responded to the AI debate by labeling everything: appeal letters, thank-you notes, social posts, board reports. The instinct is honest, but the result is often worse than no labeling at all. This guide draws on emerging law, donor research, and recent behavioral studies to show where AI disclosure is required, where it protects trust, and where over-disclosing actively undermines the very credibility it is meant to defend.
A nonprofit communications director sends two end-of-year appeal emails. The first is signed by the executive director and ends with a small footer that reads "Drafted with AI assistance and reviewed by our team." The second is identical in content but has no AI label. Behavioral research published in 2025 in Organizational Behavior and Human Decision Processes suggests something uncomfortable about the result: the first email, the transparent one, will be trusted less than the second. The same study found clients placing roughly twenty percent less trust in service providers who disclosed AI assistance, even when the underlying work was identical and the audience was familiar with AI.
That finding sits in tension with a parallel one. The Fundraising.AI Donor Perceptions of AI report, published in 2025, found that the overwhelming majority of donors say it is important that nonprofits disclose where and why AI is used, how humans remain in control, and what evidence shows it works. Roughly half want the ability to opt out of AI-driven interactions altogether. Donors say they want transparency, then they punish it when they see it. Both findings can be true at the same time, and the tension is the central puzzle every nonprofit leader has to navigate.
Layer in the law. The EU AI Act's Article 50 transparency obligations take effect on August 2, 2026, with several specific disclosure requirements for AI-generated content. California's chatbot disclosure law took effect at the start of 2026 with a private right of action attached. Utah, Illinois, Colorado, and New York have added their own disclosure rules. The Federal Trade Commission has signaled a 2026 enforcement push on AI-related advertising claims. Disclosure is no longer just a values question, it is increasingly a compliance question, and the compliance scope varies dramatically by jurisdiction and content type.
This article works through where nonprofits should disclose AI use, where they should not, and how to write disclosure language that meets the legal threshold without triggering the trust penalty. For the broader transparency posture this work sits inside, see building a strategic plan for AI and the August 2026 EU AI Act deadline.
The Transparency Paradox: Why Donors Say One Thing and Do Another
Survey data and behavioral data tell different stories about AI disclosure, and the gap between them is not noise. It is structural. When donors are asked directly whether nonprofits should disclose AI use, the answer is almost universally yes. When the same donors encounter an AI-labeled appeal in their inbox, they trust it less than an unlabeled one. The pattern is consistent enough across studies that nonprofit leaders should treat it as a fact about human psychology rather than a research artifact.
The mechanism researchers point to is something called a legitimacy discount. When a person sees an AI label on a communication, they interpret it as a signal that the human author was either less competent than they should have been, less invested in the relationship than the message implies, or less authentic than the warmth of the language suggests. None of those interpretations is necessarily fair, but all of them happen automatically. The label triggers a question the recipient was not previously asking, and that question itself reduces the persuasive force of whatever the message was trying to do.
What donors say in surveys
- The vast majority say nonprofits should disclose where and why AI is used
- About half want the ability to opt out of AI-driven interactions
- Roughly half want third-party audits of nonprofit AI systems
- Top concern: AI bots presented as humans representing a charity
What behavioral studies show
- AI labels can reduce trust by roughly twenty percent in expressive contexts
- The effect persists even with audiences familiar with AI tools
- AI labels can make misinformation seem more credible (machine heuristic)
- The strongest backfire occurs in personal, relational communication
The implication is not that nonprofits should disclose less. It is that they should disclose differently. The disclosure that protects trust is calibrated to the materiality of the AI's role, the audience encountering it, and the legal context. Blanket disclosure of every word that touched an AI tool treats spell-check the same as a wholly generated appeal, which is both unhelpful to the reader and corrosive to credibility. The work is to draw lines that match how thoughtful people actually think about AI involvement, not to label everything as a defensive reflex.
Where Disclosure Is Legally Required or Strongly Expected
Before talking about discretionary disclosure, it is worth being precise about where the question is no longer discretionary. Several categories of AI use now carry explicit legal disclosure requirements in major jurisdictions, and the patchwork is expanding rather than consolidating. Nonprofits operating in multiple states or with any European presence should treat these as floors rather than ceilings.
Chatbots and conversational AI
Required almost everywhere the question has been answered
California's SB 243 took effect at the start of 2026 and requires operators of companion or human-like chatbots to clearly and conspicuously disclose that the user is not talking to a human. The law carries a private right of action with statutory damages starting at one thousand dollars per violation plus attorney fees. Utah's AIPA, the EU AI Act, and several other state laws layer additional requirements, with higher bars in mental-health and other high-risk contexts.
For a nonprofit running a donor service chatbot, a beneficiary intake bot, or any conversational interface that a member of the public might mistake for a person, persistent disclosure is now table stakes. The disclosure should be visible at the start of every session, not buried in a privacy policy three clicks away.
AI-generated images, audio, and video
Mandatory under the EU AI Act, increasingly expected by donors
Article 50 of the EU AI Act requires deployers of AI-generated or manipulated image, audio, or video content to disclose that the content is artificially generated. The deadline is August 2, 2026, with a small carve-out for clearly artistic, satirical, or fictional use. New York's AI content disclosure law for advertising, signed January 2026 with effect from June, adds significant penalties for undisclosed AI content in advertising.
Beyond the law, the nonprofit sector is converging on a strong norm following the controversies around AI-generated humanitarian imagery in late 2025. Visual content that depicts beneficiaries, scenes of need, or program outcomes should carry a clear AI label when it is generated rather than photographed, and ideally should not be used at all in those contexts. The reputational risk of being caught with undisclosed AI imagery in a fundraising appeal exceeds any plausible cost saving.
Consequential decisions about people
Disclosure rules expanding state by state
Colorado's revised AI Act, effective January 2027, requires pre-use notice when automated decision-making technology is used in consequential employment decisions, plus post-adverse-outcome disclosure within thirty days. For nonprofits, this matters most in hiring, volunteer screening, and any beneficiary-eligibility decision where AI plays a meaningful role. Quebec's Law 25 already requires notification when decisions are made solely by AI.
Even where no specific law applies, sector norms now treat AI involvement in grant decisions, beneficiary screening, and program eligibility as decisions that warrant explanation to the person affected, including what role the AI played, what the human reviewer did, and how to challenge the outcome. The framework is explored further in the right to explanation for nonprofit AI.
AI-generated text on matters of public interest
Specific to EU exposure but a sector signal
Article 50 also covers AI-generated text published on matters of public interest, which means a nonprofit publishing AI-drafted research summaries, policy briefs, or news commentary into an EU audience needs to be careful. The disclosure requirement is that the content is artificially generated or manipulated, which is a different and looser standard than wholly generated by AI.
For U.S.-only nonprofits, this is currently a sector-signal rather than a legal requirement, but it is worth aligning with given how often advocacy content crosses jurisdictional lines.
Where Over-Disclosure Hurts More Than It Helps
The mirror image of the legal floor is the practical ceiling. Disclosing AI involvement in uses where it is genuinely incidental, where a reasonable person would not consider it material, or where the disclosure invites a misreading that the AI did more than it did, is the most common trust mistake nonprofits make when they begin to think seriously about transparency. The labels were intended to demonstrate honesty, and they end up signaling either incompetence or evasion instead.
Spell-check, grammar, and routine polishing
Grammarly, the spell-checker in your word processor, and the autocomplete in your email client all involve AI, in some sense, in nearly every document a modern nonprofit produces. Disclosing this is unhelpful in two directions. It implies the AI did more than it did, which can erode trust unnecessarily, and it desensitizes audiences to AI disclosure in the cases where it actually matters.
The materiality test that works: did the AI shape meaning, voice, or persuasion in this content. If the answer is no and a thoughtful reader would not consider its involvement consequential, no disclosure is warranted.
Drafts that humans substantially rewrote
A development director who asks an AI for five subject-line variations, picks none of them, and writes her own based on the ideas they triggered has used AI. Disclosing this in the email itself misrepresents the actual authorship and triggers the legitimacy discount in exchange for very little informational value to the recipient.
The same goes for AI-suggested outlines, brainstorming sessions, and research prompts that informed but did not produce the final copy. The rule of thumb: if a human substantively wrote the words that ship, the human is the author.
Background analytics and segmentation
AI-driven donor segmentation, predictive giving models, lapsed-supporter scoring, and most of what goes on inside a modern nonprofit CRM is invisible to the donor and not something they expect to be told about. Disclosing these uses on a per-communication basis is both unworkable and uninformative.
The right place for transparency on these uses is a public AI use statement page, where the nonprofit explains its analytics and modeling practices in plain language and offers opt-out paths for donors who want them. The Fundraising.AI finding that around half of donors want opt-out options applies here more than to inline labeling.
Personal, relational communication
Thank-you notes, condolence messages, and personal donor outreach are the contexts where AI disclosure backfires most sharply in the research. The implication is uncomfortable but important: these are the contexts where AI probably should not be drafting the substance in the first place. A handwritten thank-you note from the executive director is worth keeping handwritten precisely because what it signals is human attention, and AI assistance erodes that signal even when disclosed.
The honest move in this category is not better disclosure language. It is to use AI for the logistical scaffolding around relational communication, including reminders, contact lists, and timing prompts, and to leave the words themselves to the humans whose relationships those messages represent.
Disclosure Patterns That Hold Up in Practice
Between the legal floor and the trust-eroding ceiling lies a band of communications where disclosure is genuinely helpful and where the form of the disclosure determines whether it builds trust or destroys it. Five patterns are emerging as workable defaults across the nonprofit sector in 2026. They are not mutually exclusive, and most organizations end up using several at once.
Public AI use statement page
The single most important disclosure surface
A dedicated page on the nonprofit's website that explains which categories of AI tools are used, what they are used for, what data is excluded from AI tools entirely, how human review works, what opt-out options are available, and when the statement was last updated. This is where the donor who actually wants to understand AI involvement can find a substantive answer, and where regulators and journalists will look first.
The page does not need to be long. It does need to be honest, specific to the organization, and refreshed as the practice changes. Treating it as a static legal page rather than a living statement is the most common failure mode.
Persistent chatbot badge
Required by law in several jurisdictions
Every chatbot interface used by the public should carry a visible label that names the bot as a bot, ideally at the start of every session and persistently while the conversation is active. Language like "You are chatting with an AI assistant. For complex questions, please reach a team member at..." satisfies the legal requirement in California and several other states while also offering a clear escalation path that builds rather than erodes trust.
For helpline and mental-health-adjacent contexts, additional disclosures are required, and many sector experts argue the bot itself is the wrong tool. The risk framework in why crisis hotlines should not use generic chatbots applies here.
Visual provenance via Content Credentials
Cryptographic labels for images and video
The Coalition for Content Provenance and Authenticity (C2PA) standard, implemented as Content Credentials, embeds cryptographic provenance information in images and video. Adobe, Microsoft, OpenAI, Google, and most major camera manufacturers now support it. For nonprofits producing visual content, adopting Content Credentials gives a verifiable way to show whether content is authentic photography, AI-generated, or AI-edited.
One caveat that has surfaced in 2026 user research: many viewers misread the Content Credentials icon as marking AI-generated content rather than verifying authentic origin. Educating your audience about what the icon actually means, especially in donor and major-gift communications, is part of the work of adopting it.
Inline label on substantively generated content
When the AI did the heavy lifting and a human reviewed
For long-form content where the AI produced most of the words and a human reviewed and signed off, a short inline or footer notice is the right pattern. Language that has tested well in donor research includes "Drafted with AI assistance and reviewed by our team" and "AI-assisted, human-edited." The disclosure is honest without overstating the AI's autonomy.
This pattern is appropriate for research briefs, AI-drafted social posts that ship largely as written, AI-translated content where the translation was not extensively re-edited, and similar uses. It is not appropriate for personal communications, condolence notes, or thank-you messages, where the trust penalty exceeds the transparency benefit.
Process disclosure for consequential decisions
Specific to decision-making contexts
When AI plays a role in a decision that affects an individual, including hiring, volunteer placement, grant award, or program eligibility, the disclosure should explain what the AI did, what the human reviewer did, and how the affected person can challenge the outcome. This is more than a label, it is a short paragraph in a notice or letter, and it is increasingly required by law in employment contexts.
The good news is that process disclosures, when well written, generally increase rather than decrease trust because they answer the question the affected person is most likely to ask: was this decision made fairly. The pattern works best when there is a real human review step to describe, not when AI made the decision and a person rubber-stamped it.
Disclosure Language That Has Tested Well
The specific words matter more than nonprofit leaders sometimes assume. Disclosure language that sounds bureaucratic, evasive, or preachy can do more reputational damage than no disclosure at all. Language that is plain, specific, and located near the relevant content tends to hold up better in donor research and reads naturally to non-specialist audiences.
For a public AI use statement
"We sometimes use technology, including AI-assisted tools, to help our small team draft communications, summarize non-confidential insights, and improve how we serve our community. We do not use AI to replace human decision-making, and we do not enter sensitive personal information into public AI tools. Our team reviews donor-facing communications before they are sent."
For an inline AI label
"Drafted with AI assistance and reviewed by our team."
"AI-assisted, human-edited."
"This summary was prepared with AI tools and reviewed for accuracy by [name or team]."
For a chatbot session start
"Hi, I'm an AI assistant from [organization]. I can help with general questions and routing. For sensitive or complex matters, I'll connect you with a team member."
For a process disclosure on a decision
"This decision was reviewed using an AI-assisted screening tool to help our team manage a large applicant pool. A staff member reviewed the AI's output before any decision was made. If you would like to know more about how this decision was reached, or you believe an error was made, please contact us at..."
Avoid language that overpromises or moralizes. Phrases like "we use AI responsibly" or "our ethical AI" without substantive backing trigger skepticism more often than reassurance. So does language that sounds like it was written by a lawyer for a privacy policy. The disclosure should sound like the rest of the organization's voice, slightly more careful.
Edge Cases Every Nonprofit Will Eventually Hit
The framework above covers the central ground. The edge cases are where most actual policy work happens, and where nonprofit leaders should think carefully before a real incident forces a rushed answer.
Volunteer-authored content
Many nonprofits rely on volunteers for drafting, social media, and event communications. The disclosure standards should track the role of the content, not the employment status of the author. A volunteer-drafted appeal letter that used AI for the heavy lifting deserves the same disclosure as a staff-drafted one. The implication for governance is that AI policies must apply to volunteers, which requires both training and a way to find out when volunteers depart from the policy.
Different audiences, same tool
A nonprofit chatbot serves both donors and beneficiaries. The disclosure obligations differ. Beneficiaries interacting with an eligibility chatbot are protected by stricter privacy laws including GDPR and Quebec's Law 25, while donor interactions are governed mainly by sector norms and FTC advertising principles. The right answer is usually a unified disclosure that meets the strictest applicable standard, then layered notices for specific contexts.
Staff disclosing what leadership has not
Staff who believe AI is being used in ways the organization is not disclosing will eventually face a choice between loyalty and public honesty. Without a clear policy and an internal escalation path, this can spill into public whistleblowing that damages the organization more than disclosure would have. A working AI policy should name a decision-maker for unclear cases and explicitly invite staff to raise concerns through that path before they become external complaints.
Content Credentials misread as AI labels
The Content Credentials icon was designed to mark authenticated origin, including for legitimate human-taken photography. Recent user research has found that many viewers misread it as marking AI-generated content. Nonprofits adopting the standard should plan for audience education, particularly in major-gift and donor stewardship communications where the misread could undermine credibility.
A Compliance Posture That Will Hold Up Through 2026 and 2027
The disclosure landscape is moving quickly, and any specific list of obligations will be partly out of date by the time it is implemented. A more durable approach is to adopt a compliance posture that translates new rules into action as they appear, rather than trying to anticipate every change in advance.
Six commitments that travel well across jurisdictions
- Always label AI chatbots and generated visual content, regardless of where the audience is
- Disclose AI involvement in consequential decisions about individuals, with a real process explanation
- Maintain a public AI use statement that is reviewed at least annually and refreshed when practice changes
- Provide opt-out paths for AI-driven interactions where feasible, especially for donors and beneficiaries
- Apply AI policies to volunteers and contractors, not just staff, and provide training that matches
- Treat AI-related communications as a board-visible topic so leadership can see practice and outcomes together
For the EU exposure piece specifically, our coverage in the EU AI Act August 2026 deadline walks through the additional steps U.S. nonprofits with European operations need to take. The compliance landscape for AI-related advertising claims is covered in AI and charitable solicitation compliance.
Conclusion: Disclose for Trust, Not for Theater
The temptation when nonprofit leaders first encounter the AI disclosure question is to label everything, on the assumption that more transparency is always better. The research shows it is not. Labels carry meaning, and meaning has trust consequences. A label that misrepresents the AI's role, or that signals more autonomy than the AI actually had, costs credibility in exchange for very little informational benefit to the reader.
The discipline that holds up across jurisdictions and audiences is to disclose where the law requires it, where a reasonable person would consider the AI involvement material, and where process transparency genuinely helps the person affected. Stop short of labeling routine tools and substantially human-rewritten drafts, where the disclosure does more harm than good. And reserve personal, relational communication for actual human authorship, because no disclosure language repairs the trust damage from labeling a thank-you note as AI-assisted.
The work of getting this right is one of the more underappreciated parts of building an AI-mature nonprofit. It requires policy, training, communications discipline, and a willingness to revisit decisions as the law and the technology change. Done well, it is invisible: donors trust the organization, beneficiaries understand the systems they encounter, and the disclosures fade into the background of a credible, modern nonprofit operation. Done poorly, it becomes the story, and the story is rarely good for the mission.
For the policy-writing companion to this article, see how to create an AI policy for your nonprofit. For the broader transparency posture, our coverage of AI in charitable solicitation compliance covers the advertising and fundraising specifics in more depth.
Need Help Drafting Your AI Disclosure Posture?
We help nonprofits write public AI use statements, calibrate inline disclosure language, and align their practices with the EU AI Act, state laws, and the donor-trust research that should shape how disclosure is actually delivered.