Human Before AI: Why Human Review Matters in Nonprofit Work

I use the tools. I also review the work.

I use AI every day. It helps me organize information, develop workflows, and take on projects that would have been difficult to manage otherwise. Some tasks that might have taken days can now come together in an hour.

I also review the work.

Accepting an AI response as completely factual without checking it is poor professional practice. The fact that something reads well doesn’t mean it’s accurate, complete, or appropriate for the decision someone needs to make.

And in nonprofit work, those decisions affect people.

I understand why we need the help

Many of us wear multiple hats. We’re asked to take on responsibilities outside our title, job description, and sometimes our experience. An event needs preparation. Leadership needs a briefing. A colleague needs help making sense of information before a meeting.

We step in because we care about the people we work with and the communities we serve.

That’s why I’m excited about AI. Used thoughtfully, it gives us room to accomplish work that might otherwise stay on the someday list.

But the time we save needs to include time to review. If a workflow produces more information than we can reasonably verify, we need to adjust the workflow.

A source link is the beginning of verification

When I prepare a research profile or report, I need evidence supporting the factual claims I include. A link beside a sentence gives me somewhere to look. I still have to check what it says.

In March 2025, Columbia University’s Tow Center for Digital Journalism published an investigation testing eight AI search tools on their ability to identify and cite news articles. Researchers documented incorrect attribution, fabricated links, and confident answers when the tools could not reliably identify the source. Those findings describe a specific test of the tools available at that time, rather than an error rate for every AI task today. They still illustrate why citations need inspection.

For example, a source might confirm that someone serves on a nonprofit board. That alone doesn’t establish their giving capacity or interest in supporting our organization. Those are separate questions requiring additional evidence and judgment.

For prospect research, that means asking practical questions:

• Is this the right person?
• Does the source support the claim?
• Is the information current enough for how we’re using it?
• Have we clearly separated confirmed information from our interpretation?

• Use the follow-up research prompts (PDF)

We still have to do the thinking

A study published at CHI in April 2025 surveyed 319 knowledge workers about their use of generative AI. Higher confidence in AI was associated with less reported critical thinking. The researchers also found that AI shifted critical thinking toward checking information, integrating responses, and overseeing the work. This was a self-reported survey, so it doesn’t establish that AI causes people to lose critical-thinking ability.

My takeaway is practical: we need to stay actively involved, especially when an answer looks convincing.

In my work, AI sits alongside other research tools, records, sources, and professional experience. It can help construct a useful starting point. I’m responsible for deciding what belongs in the finished product.

Sometimes that means correcting a detail. Sometimes it means removing a claim because the evidence isn’t strong enough. Sometimes it means saying we don’t know yet.

• Explore Prospect Research & Intelligence

Human review needs to be part of the process

Apra’s July 2025 update on AI use in fundraising reinforces the need for ongoing oversight. It also encourages practitioners to understand the type of AI they’re using, evaluate vendors, and follow their organization’s acceptable-use policies.

For a busy nonprofit team, I would start with a simple expectation: before AI-assisted work informs a decision or reaches another person, someone needs to own its review.

That person needs enough time and context to check it. A quick glance at a polished report won’t necessarily catch a mistaken identity, an outdated role, or a conclusion the evidence doesn’t support.

Human review cannot promise perfection. It does give us a process for questioning the work, correcting errors, and explaining what we know.

That’s what human before AI means to me.

Use the tools. Build things you didn’t have time to build before. Give yourself room to focus on the people and relationships that brought you into this work.

And when someone asks, “How do we know this?” be ready to walk them through the evidence.