When we introduce Ditto to people, one of the most common questions we get is: “How can an AI possibly understand what people actually want?” It’s a fair question. But it shows that a fundamental insight about large language models gets overlooked in discussions about AI and market research.
Language models don’t learn what people do. They learn what people say.
This sounds like a limitation and in most contexts, it is. But for understanding markets, it is exactly what you need.
The Discourse Layer
Large language models are trained on text. Conversations, articles, reviews, discussions, explanations, debates. They learn the patterns of how humans articulate reasoning, express concerns, justify decisions, and explain choices.
Importantly, they don’t learn from direct experience.
They’ve never felt the hesitation before clicking a purchase button. Never experienced the awkwardness of telling a colleague their recommendation didn’t work out. Never felt the relief of finally solving a persistent problem. They never loved, or grieved.
What they learn instead is how people talk through these moments. The language of decision-making. The vocabulary of doubt and certainty. The phrases people use when explaining themselves, and their innermost emotions, to others and to themselves.
The Same Place Humans Live
For building systems that interact with the physical world, this is a serious constraint. An autonomous vehicle needs to understand physics, not descriptions of physics. A robot needs to know what happens when it moves, not what people say about movement.
But here’s the thing about humans: We’re much better at articulating reasoning than predicting our own actions.
We’re fluent in explaining trade-offs. We can walk through our concerns, using language. We are evolutionary pre-trained to know how to describe what appeals to us and what gives us pause. We’re articulate about constraints, about competing priorities, about what matters in theory versus what matters in practice.
What we fail at is predicting what we’ll do when the moment arrives.
Why This Matters for Synthetic Research
Most market research questions aren’t about predicting specific behaviors. They’re about understanding the cognitive and social context in which decisions happen.
What concerns do people voice? What objections surface? What language resonates? What framing feels manipulative versus clarifying? How do people justify choices to themselves and others?
These aren’t questions about what will happen. They’re questions about what gets said, and what gets said shapes what happens.
Language models and humans both operate in the discourse layer. For most research questions, that’s not a bug—it’s the entire point.
