LLMs with free-text answers expose what really drives choices, researchers say

Why do people make the choices they do? That question sounds philosophical until you run into the practical problem: behavioral data tells you what happened, not why it happened. A new approach reported by researchers from the Center Synergy of Systems (SynoSys) at TUD Dresden University of Technology, the Max Planck Institute for Human Development, and the University of Basel tries to close that gap. Their method combines observed choices with participants’ own descriptions of their decision processes. The goal is simple to state and hard to pull off in practice: study human behavior in greater detail than is possible with behavioral data alone.

Here is the key promise, straight from what they built: you do not just watch what people pick, you also collect what they say they were thinking. Then, critically, free-text answers and LLMs are brought together to reveal the hidden reasons behind human choices. In other words, the researchers treat decision-making like a process you can both observe and partially reconstruct. That matters because, in many real-world settings, “what people did” is only half the story. The other half is the reasoning that produced it, including context people may not realize they are using or may not be able to express in a forced-choice format.

To see why executives should care, zoom out to how decisions get made in the first place. Product teams run experiments and analyze outcomes. Marketing teams segment audiences and optimize conversion. Policy teams design rules and then measure compliance. In each case, teams typically rely on behavioral signals, such as what people clicked, bought, chose, or avoided. Those signals can be powerful, but they can also be misleading when incentives, constraints, or internal mental models are not aligned with what the metrics measure.

This is where the new approach becomes interesting. The researchers are not replacing observed choices. They are combining observed choices with participants’ own descriptions of their decision processes. That pairing acts like an extra lens. Behavioral data can show correlations. Free-text reasoning can add the mechanism, at least more than behavioral data alone. And when LLMs are used with free-text answers, the system can potentially translate messy human explanations into patterns researchers can study.

There is also a second-order effect that decision-makers should not ignore: improving “reason discovery” can change how teams interpret experiments and user research. If you learn that people picked an option for reasons that differ from your assumptions, you can stop optimizing the wrong lever. Boards and executives already worry about strategy drift, where leaders read dashboards but miss the underlying driver. A method that surfaces hidden reasons could tighten the feedback loop between plan and reality. It could also make it easier to explain to stakeholders why certain strategies worked or failed, because the “why” comes from participants’ own descriptions rather than only from statistical inference.

From a research-to-industry perspective, the regulatory backdrop is also relevant. When you collect free-text explanations and process them with LLMs, you are in a world where privacy, consent, and data governance become central. Even though the source does not spell out specific regulatory plans, the approach itself points to the direction many organizations are moving: blending qualitative inputs with machine-assisted analysis. That combination tends to raise questions about how participant data is handled, how sensitive information is protected, and how results should be interpreted responsibly. Executives who treat “reason data” as just another dataset may get surprised later. Teams that plan for governance early, with clear controls around what gets collected and how it is processed, will have an easier time scaling.

Finally, consider what this means for decision-making beyond research labs. In finance, hiring, healthcare, and consumer tech, leaders constantly face the same mismatch: metrics show outcomes, but stakeholders still ask for explanations. “People chose X” is not the same as “people chose X because Y.” When researchers can study decision processes in greater detail, they can generate better models of behavior. That, in turn, can inform how organizations design interventions, communication, and product features to fit how people actually think and decide.

So the strategic stakes for peers in similar roles are clear. If hidden reasons behind human choices can be extracted by combining observed choices with participants’ own descriptions, then companies that rely only on behavior may be missing the levers that truly move outcomes. The researchers’ approach from SynoSys, the Max Planck Institute for Human Development, and the University of Basel is not just an academic tweak. It is a reminder that understanding people requires more than watching their actions. It requires listening closely enough that the reasons behind those actions become visible.