The Nonprofit AI Almanac

Guide

Fairness, consent and ownership

Please note: This site is only a guide to help people understand. Consult your IT, security and/or privacy department for clarification.

Bias and unfair treatment

AI tools learn from huge amounts of writing and images made by people. That material has unfair patterns in it, so a tool can repeat bias without anyone meaning it to.

Outputs can repeat stereotypes about communities, disability, race, gender, age or poverty. A summary might leave out the people who said the least. A translation or rewrite might make someone sound less capable than they are.

So ask two questions about every draft. Who might be left out? Who might be shown in an unfair light? Ask people with lived experience to review public material before it goes out, and pay or thank them for their time.

Never use AI to make decisions about people. It must not decide who is eligible for help, rank or prioritize clients or beneficiaries, or screen, hire or fire staff and volunteers. A person decides; AI can only help draft.

When you read a draft, watch for these:

  • Words that define people by a problem, such as "addicts" or "the homeless". Try "people who use drugs" or "people experiencing homelessness", or the words your community prefers.
  • Stories that show one group only as needing help, never as helpers, experts or leaders.
  • Guesses about someone's background, family, faith or abilities that were not in your facts.
  • Summaries that drop the views of smaller or quieter groups, such as people who answered in another language.

Read where AI must not decide

Consent and dignity in storytelling

Stories and photos of the people you serve are powerful. They also belong, first of all, to those people.

Get informed consent before you share someone's story or photo. Informed means they know where it will appear, who will see it and how long it will stay up. That includes pasting their story into an AI tool, because the text usually goes to an outside company. If you have not asked, remove names and details first, or leave the story out.

People must be free to say no, or to change their mind later, without losing any services. Tell them how to withdraw their consent. If they do, take the story down from anything you still control. For a child, ask a parent or guardian, and ask the child too.

Let people review their own story before it goes out. If you changed details to protect someone, or combined several people into one story (a composite), say so in the piece.

Avoid pity framing. Show people as whole people with goals and strengths, not only as problems to solve. Their dignity matters more than a moving appeal.

Do not create AI images that pretend to show real clients or real events. If you use an AI image, label it, for example "Illustration made with AI."

A sample line for your story form: "I agree that [ORGANIZATION] may share my story in [WHERE] until [DATE]. I [agree / do not agree] that staff may use an AI writing tool to help edit it. I can say no or change my mind at any time, and it will not affect my services."

Before you share a story, check each of these:

  • They agreed to this use, in this place, and you kept a record.
  • They know they can say no or withdraw later, and that it will not affect their services.
  • They saw the final version, or chose not to.
  • Names, faces and identifying details are removed, unless they said yes to each one.
  • Changed details and composite stories are labeled.
  • Nothing that identifies them went into an AI tool without their consent.

Check before you paste

Copyright and permission

Copyright protects creative work like writing, photos, music and art. Do not paste other people's copyrighted work into an AI tool beyond what you have the right to use. That includes paid reports, licensed photos, a partner's materials, and stories or art that donors or community members sent you, unless they said yes to this use.

AI output can look or sound a lot like existing work. Before you publish, check that a draft does not closely copy a known text, logo or image. If something seems familiar, rewrite it or leave it out.

Who owns AI-generated material is still being worked out in law. Below is a short summary of what the US Copyright Office has said. It covers US law only, and it may change. Laws differ by country, and the Office notes that other countries are working through these questions under their own laws.

  • In a January 2025 report, the Office concluded that copyright does not extend to purely AI-generated material, or to material where a person did not have enough control over the expressive parts.
  • It said that using AI as a tool to assist human creativity, rather than stand in for it, does not affect copyright protection. A person's own original expression is still protected even if the work also includes AI-generated material.
  • It said that, with the technology generally available today, prompts alone do not give enough control to make the user the author. A person's creative selection, arrangement or changes to the output can be protected.
  • It said that whether a person contributed enough to count as the author must be decided case by case, and the Office said it will keep watching for changes in technology and law.
  • Its 2023 registration guidance says that if you register a work, you have a duty to disclose AI-generated content in it and briefly explain what the human author contributed.
  • The Office has also studied how copyrighted works are used to train AI. Its report on that topic was released as a pre-publication version in May 2025.

Official sources for this section:

What copyright means for your team

If your organization wants to own something, like a logo, a curriculum or a book, have people do the creative work and keep notes on what they did. Talk to a lawyer before you rely on owning anything made mostly by AI.

Also check your funder's and publisher's rules on AI use. A grant maker, journal or platform may have its own rules about AI-written applications, images or posts, and they can be stricter than the law.

This is not legal advice. For a question about a specific work, talk to a lawyer.

Translation and accessibility limits

Machine translation gives you a fast first draft. It can get tone, dialect, legal and medical words, and cultural meaning wrong. A polite phrase in one language can sound cold or rude in another.

Have a fluent speaker review anything important before it reaches your community. It helps if they know the community's own way of speaking and the topic.

Never rely on AI as the only channel for critical information, such as safety, legal or health messages. Use a qualified human translator or interpreter for those, and give people a way to ask a person questions.

AI can draft alt text (the short description of an image for people who use screen readers), captions and plain-language rewrites. A person needs to check each one against the original. Does the alt text describe what matters in the image? Do the captions match what was said, including names? Did the simpler version keep every fact?

Check the reading level of plain-language versions too. Many writing tools can show a reading level score. Then ask someone from your audience whether it is clear. Making text easier to read helps accessibility, but only if the meaning stays the same.

Before translated or simplified material goes out, check each of these:

  • A fluent speaker read the whole thing, not just a sample.
  • Names, dates, numbers, addresses and phone numbers match the original.
  • Legal, medical and money words were checked by someone who knows them.
  • The tone fits: respectful, not childish and not stiff.
  • Alt text, captions and simpler versions were checked against the original by a person.
  • Readers have a way to ask a person if something is unclear.

A quick fairness check before you publish

Answer yes or no. If any answer is no, fix it or ask someone before it goes out.

  • Did everyone whose story, photo or words appear here agree to this use?
  • Would the people described recognize themselves, with pride?
  • Did someone read it for loaded words and stereotypes?
  • Do we have the right to use every text, image and quote in it?
  • If it was translated or simplified, did a fluent speaker or a member of the audience check it?

Official sources

Every fact above about a standard or a law comes from one of these pages.

Keep going