Tag Archives: Donor Profiles

Can You Trust A.I.? Errors, Omissions, Hallucinations… Oh My!

Profiles, Prospect Research, , , , , , , ,

How the Aspire team learned to stop worrying and start working smarter with A.I.

When ChatGPT launched in 2022 — more than three years ago now — I was resistant. There was noise everywhere, good and bad, and the more I learned the more it felt like a burden. Did I want to become a prompt engineer? Not particularly. That title was not anywhere in my five-year plan.

My initial hope was that I could wait it out. Companies would eventually build purpose-built tools for prospect research and data analytics, and I’d just adopt those. And to be fair — there are now many excellent tools that work in many different ways (check out the Aspire AI Link List).

But then I attended a conference with three back-to-back speakers on A.I. in fundraising. That got me interested. What really tipped me over the edge was watching Ryan Woroniecki at Feathr pull out his phone mid-session and use ChatGPT to generate genuinely impressive responses in real time. I wanted to be able to do that too.

So my team and I dove in. We’ve been exploring CoPilot, Claude, Gemini, and ChatGPT — paid and free versions — ever since.

The mindset shift that changed everything

As we started using A.I., I realized I had been through this before. The experience that stands out is finishing my college degree in the late 1990s. The library was offering a training on a brand new resource — EBSCOhost.

The librarian was enthusiastic. We all sat down at terminals and after the tutorial, I ran some searches for a current project. Every search returned thousands of results. Thousands! I gave up, walked over to the card catalog, found five relevant books, checked the table of contents and index, and left with a few of them.

Then came Google. It was somehow easier. It was even fun. I started learning quickly how to get what I needed.

A.I. feels like that moment again. The learning curve going up is steep. But the coasting downhill is worth it.

What we were worried about — and still are

Like many teams in our field, the Aspire team had real concerns: privacy, erroneous information, hallucinated sources, and critical omissions. All of these things do happen. We’ve experienced them firsthand.

But now we have a much better understanding of how and where to incorporate A.I. into our workflow to avoid those trip wires — because we did the work of testing A.I. alongside our traditional methods. It took extra time. It was worth it.

Along the way, we’ve also discovered benefits beyond profiles: email communications, marketing materials, data analysis. A.I. has quietly improved a lot of corners of our work.

How we’ve made it work for profile research

Here’s what we’ve learned so far — the practical moves that help us use A.I. while managing the risks:

Separate the critical pieces. A.I. is genuinely excellent at gathering and summarizing — giving us polished, readable narrative bios. But some information needs to be highlighted and verified completely. Keeping those pieces outside the A.I.-generated narrative helps us focus our attention where it matters most.

Check the sources. It has always been important to recognize when information should be verified and cited. The nuance with A.I. is specific: a link can be real and an article can exist, but that doesn’t mean it confirms the claim being made. And the usual credibility assessment still applies — source reliability, publication date, context. (Test your own skills with our free Solid Intel course.)

Set it down and pick it up again. Fresh eyes help us catch where we need to tweak bio language (bye, “American businessman”), question vague word choices (what does “runs the company” actually mean — CEO? Owner?), and make sure everything in the profile is internally consistent.

Run a wide-net search anyway. Even with A.I.-assisted research, we preserve the wide-net search as a fail-safe. This isn’t just an A.I. concern — public database tools miss things too.

Include the household. Early on, we had a client ask why we’d ignored the wives. It was a fair question. We were so focused on the process that we missed the obvious: A.I. had matched data and written lovely bios on the spouse listed first — and almost all of them were men. That one stung. It also made us better.

Embrace a margin of error. We’ve always been comfortable with capacity ratings that are wrong 100% of the time but excellent for segmentation. We’re now testing A.I.-generated bios delivered without human touch at scale — for use in the database to assist frontline outreach during campaign. Time and experience will tell what that margin of error looks like.

So what’s actually changed?

For the most part, A.I. hasn’t shaved dramatic time off our profile research. What it has done is elevate the quality of the work and give us more space to think strategically. We can deliver more readable, accessible profiles that appeal to our human love of stories. If we had to write polished narrative from scratch every time, we’d never have the bandwidth.

A.I. can also surface connections and answer questions that used to require long, tedious searches — like “does this person or their company have any ties to Dallas, TX?”

More than anything, it has pushed us to be more intentional about where human judgment enters the picture. Because that’s still the scarce resource. A.I. can generate the bio, but it can’t tell you whether the information passes the smell test, or whether the profile is actually useful to the gift officer sitting across from a donor next week.

That part is still ours.

We’re continuing to test and reimagine how research can support frontline fundraisers, and I look forward to sharing more as those experiments mature. In the meantime — what’s your team finding? I’d love to hear how you’re navigating the A.I. learning curve.

Additional Resources