Nonprofit AI field guide
Reviewed July 21, 2026Write the human review boundary
Make it clear where AI may assist, where a person must decide, and where the workflow must stop.
Separate assistance from authority
Describe the allowed role with a verb such as summarize, classify for review, draft, translate for review, or suggest questions. Avoid language that implies the system owns a decision, relationship, or professional judgment.
Name the person accountable for the final result. Human review is meaningful only when the reviewer has time, context, authority, and a practical way to reject or correct the output.
Define what review must check
Create a short review rubric tied to the use case: factual support, missing context, tone, accessibility, prohibited content, protected information, bias, citations, and required organizational policy.
Use representative difficult cases, not only easy demonstrations. Include ambiguous input, missing information, multilingual content, accessibility needs, and examples where the correct response is to abstain or escalate.
Give people a stop path
Write the conditions that pause the workflow and route it to the responsible human process. Examples include safety concerns, legal rights, eligibility, employment action, safeguarding, medical content, material financial impact, or uncertain identity.
Tell affected staff how to report a problem without fear of being blamed for the tool. The feedback path should reach an owner who can suspend use, preserve evidence, notify responsible parties, and revise the boundary.