Beyond the Hype: Using AI to Discover Personas, Not Invent Them.

Over the last year, I’ve seen “AI-powered personas” that were really just LLMs role-playing as UX researchers. We’d prompt them to “create three personas for a banking app,” and get back “Frugal Fiona,” “Investor Ivan,” and “Tech Averse Tom.” It looks impressive for five seconds until you realize it’s just polished guessing. That’s not using AI. That’s outsourcing imagination.

The real value of AI in UX isn’t about inventing users. It’s about discovering them. As a cognitive psychologist, I’ve always struggled with how personas are often built, three or five tidy archetypes we decide in advance, forcing messy reality into neat categories. Real people aren’t that simple. This is where machine learning changes the game. Instead of forcing patterns, it helps us find them.

Every product team is sitting on a mountain of behavioral evidence: clickstreams, task times, error rates, feature use, even the way people move through a form. That’s data waiting to be explored. When you clean it, scale it (please, scale your data), and run real unsupervised models like clustering, UMAP, or HDBSCAN, something genuinely insightful happens. Patterns emerge.

You don’t invent “The Efficient User.” You discover them, a clear segment of users who finish tasks quickly, explore features deeply, and never touch the tutorial. You don’t imagine “The Struggling Novice.” You find them, users who hit high error rates, revisit the help center, and stall at the same step in your funnel. These aren’t characters. They’re statistically distinct groups defined by shared behaviors.

And that changes what we do as researchers. The machine can tell us what patterns exist, but it can’t tell us why. That’s our job. We’re the detectives, looking at Cluster 2 and asking, “Why are these users behaving this way? What in our design is driving this?” We take the cold data and bring it back to life with context, story, and empathy.

That’s the real synthesis, personas that are grounded in evidence but enriched with human understanding. So maybe it’s time to stop asking AI to write stories for us and start using it to uncover the ones already written in our data. Let’s move from invention to discovery, and start creating personas that evolve with our users instead of sitting still on a slide.

If you want to see how to do this in practice, I’ve shared a detailed step by step guide with real data and code here: Machine Learning for Personas