The first stage of design thinking is the one nonprofits are about to quietly delete.

You can see it coming. A program team wants to know why families keep dropping out of an after school program. Nobody has time to sit down with families. So someone opens a chatbot and types: "Act as a single mother of three working two jobs. Why might you stop bringing your kids to our program?"

The answer comes back in about four seconds. It's articulate. It's plausible. It mentions transportation, schedule conflicts, and feeling judged at pickup. It reads like research.

It isn't. It's a guess in a costume.

The Shortcut Looks Like Research

Synthetic personas are everywhere now. They've become the default answer whenever design thinking and AI come up in the same meeting. Vendors sell them. Webinars teach them. The pitch is simple: why spend three weeks on interviews when a model can simulate your audience in an afternoon?

I get the appeal. I've spent twenty years in this sector and I have never met a program team with spare time. If a tool promises to compress the slowest part of the process, people are going to reach for it.

But think about where that simulated mother came from. She's an average of everything ever written about women like her. News stories. Grant narratives. Blog posts by people like me. She knows what the sector already believes about her, and that's all she knows.

So when you ask her why families leave, she tells you what you already thought. Transportation. Scheduling. Stigma. You nod, because it matches your assumptions, and the match feels like confirmation.

It's an echo.

What Selwyna Knew

When I was in the Peace Corps, I taught at a school called C.A.R.E. One of the teachers there, Selwyna McKetney, adjusted something in every single class. A lesson would start one way, and by the halfway mark it had shifted. Sometimes twice.

At first I thought she was disorganized. Then I started watching more closely. She was reading the room the entire time. Who had gone quiet. Who was pretending to follow along. Which kid had walked in hungry, or tired, or carrying something from home that had nothing to do with the lesson. She noticed, she changed course, and nobody in that room ever had to explain themselves to her.

No lesson plan told her to do that. No survey would have caught what she caught. She caught it because she was there.

That's empathy in the design thinking sense. Not a warm feeling. A practice of paying close attention to real people in real conditions, close enough to notice what they would never think to tell you.

A language model can't stand in that room. It has never been in any room.

Four Things a Simulation Can't Give You

When design thinking and AI get blended carelessly, these are the first things to drop out.

The detail nobody predicted. Real conversations surface things nobody would think to guess. The family that stopped coming because the new pickup door faces a street the kids are scared to cross. The parent who quit because a staff member mispronounced her daughter's name for three months and she got tired of correcting it. A model trained on the average gives you the average problem. Your problem is rarely average.

The words people actually use. In my last post on design thinking I wrote about listening for your community's vocabulary, the "Thursday class" instead of whatever the program is officially called. A synthetic persona doesn't talk like your families. It talks like a grant report, because grant reports are what it learned from.

The relationship. Some of the most useful things I've learned in this work came out in a second or third conversation, after someone decided I was worth being honest with. Trust is part of the data. You can't simulate the moment someone finally tells you the real reason.

The people who are missing. Models know the most about the people the internet writes about most. The communities nonprofits serve are often the ones least represented in that data. A simulation fills those gaps with stereotypes and hands them back with total confidence.

I'm Not Anti AI. I'm Anti Skipping.

I use these tools. And the prototype stage is a different story entirely. Ten versions of a flyer in ten minutes? Great. A rough draft of a new intake form you can put in front of five real people tomorrow? Even better. I'll write more about that soon.

AI is genuinely useful after you've listened. It can help you sort interview notes, spot patterns across thirty conversations, or draft better questions for next time. It can make the empathy work easier to organize.

It can't make the empathy work unnecessary.

That's how I think about design thinking and AI working together. The line I'd draw is simple. Use AI to work with what you learned from people. Never use it to replace learning from people.

Put It in Writing

If your organization is drafting an AI policy right now, add one sentence: no synthetic persona substitutes for talking to a real person we serve.

Then make the real version easy. Five conversations. Five people who use the program, or left it, or never signed up in the first place. Twenty minutes each. Buy the coffee. Ask what their week looks like, not what they think of your services. Then listen for the thing you didn't expect.

If you really want to bring AI into it, use it as a foil. Before your interviews, write down what the synthetic persona tells you. Afterward, put that list next to what real people said. Every gap between the two lists is a place where assumptions have been running your program.

The Room Hasn't Changed

Selwyna never needed a framework. She needed to be in the room and to care enough to notice. That's still the job. The tools changed. The room didn't. Whatever happens with design thinking and AI over the next few years, that part stays human.

The fastest way to understand people still looks like the slowest one. Show up. Ask. Stop talking. Change something because of what you heard.

When was the last time a decision about your program changed because of something a participant told you face to face?

Frequently Asked Questions

What is a synthetic persona?
A synthetic persona is an AI generated stand in for a real audience member. You prompt a model to "act as" a donor, volunteer, or program participant and ask it questions. The answers sound realistic, but they reflect patterns in the model's training data, not the lived experience of the people you actually serve.

How do design thinking and AI fit together for nonprofits?
AI is strongest in the prototype stage and helpful around the edges of the empathize stage. AI can help you prepare interview questions, transcribe and organize notes, and find patterns across many conversations. What it can't do is replace the conversations themselves. The insight comes from being there.

Isn't a synthetic persona better than no research for a small nonprofit?
Not really, because it tends to confirm what you already believe, and that can feel like evidence when it isn't. Five short conversations with real people cost almost nothing and will tell you more than any simulation. If time is the barrier, start with three.

How does this fit into a nonprofit AI policy?
Add a clear line saying AI may support research but may not substitute for direct input from the people you serve. It protects community voice, keeps program decisions grounded, and gives staff a simple test for when a tool has crossed the line.