When AI writes your nonprofit’s story, who remains accountable?

When a nonprofit publishes a story, someone should be able to explain why its claims deserve to be believed. I would keep that responsibility with a named editor, including when AI helps produce the draft. The editor needs access to the evidence, time to review it, and authority to say that a story is not ready.

This matters most when the material describes someone else’s life. A participant’s account might become a newsletter, a donor appeal, and a social post. Each version makes choices about what to include, what to leave out, and what the experience means. I want those choices to remain visible.

Start with the evidence

NIST’s Generative AI Profile describes confabulation: generated content can sound confident while being false. It also identifies information integrity and over-reliance on AI as concerns. The profile accompanies NIST’s voluntary risk-management framework.

My application is straightforward: give the editor the source material alongside the draft. A writing tool’s explanation of where a sentence came from is something to check. It cannot substitute for the interview, approved program record, or original report.

I would ask the tool to flag missing information rather than complete the story. If the notes do not explain what happened after a program ended, the draft should leave that question open. An unfinished paragraph can be a useful reminder to make a phone call.

A small editorial example

Consider this invented example, offered as a review exercise rather than a client case. An approved program record says that twelve people completed a four-week budgeting workshop. An AI-assisted appeal describes twelve families becoming financially secure.

The second sentence introduces a different unit, a longer implied time horizon, and an outcome the record does not establish. An editor could retain the completion figure and describe what the workshop covered. A claim about financial security would need its own evidence.

I would mark the draft with three simple labels: supported, needs checking, and remove. Each disputed sentence gets a short note explaining what is missing. That gives the next reviewer a decision to make instead of another polished paragraph to reread.

Let people recognize themselves

The Society of Professional Journalists’ Code of Ethics calls for verification, context, respect for people’s privacy, and accountability for corrections. It is voluntary guidance written for journalism. I find those commitments useful when thinking about nonprofit communications, while recognizing that fundraising has a different purpose.

For a participant story, my proposed editorial conversation includes where the piece will appear, which name or photograph it will use, and what the person is comfortable sharing. I would offer a chance to check how their experience is represented. Permission for one newsletter should not silently become an assumption about every future campaign.

That conversation also helps keep the person’s own language in view. I would preserve a quotation as recorded, or make a clear paraphrase without quotation marks. A sentence created to sound like someone is not their testimony. When a story contains an illustration or an invented scenario, I would label it where the reader encounters it.

Make review part of the draft

For a small team, I would attach this short review note to the working document:

• Evidence: Which records support the numbers and outcome claims? Link them beside the relevant sentences.
• Representation: Which passages are direct quotations, paraphrases, or editorial interpretation?
• Permission: What use of the person’s account, name, and image has been agreed? Record any limits.
• Decision: Who approved this version, when, and for which channel? List anything still unresolved.
• Correction: Who will receive a concern, check it, and update the published piece?

A label saying that AI helped write something can explain part of the process. It does not resolve those questions. I would also describe the tool’s role when that information would help a reader understand the work, without implying that a disclosure replaces editorial care.

The useful question for me is whether a team can stand behind the story after it leaves the draft folder. I want software to make the sources and decisions easier to find, so the person signing off can read closely and mean the approval they give.