1970s
Preference reversals showed that choice and pricing questions can produce different answers for the same options.Research foundation.
Crystallize is built on more than forty years of preference research.
The core idea is simple but important: for complex choices, people often do not retrieve a finished preference. They form one through the act of thinking, comparing, reacting, and putting words around what matters.
The old assumption.
Many products are built as if preferences already exist in stable form. The user is expected to know the destination; the software helps them search, filter, compare, or optimize.
That works when the decision is simple. It breaks down when the choice is high-stakes, identity-laden, and unfamiliar: a first career move, a risky life change, a housing decision, a financial posture, or any moment where advice arrives before self-knowledge has settled.
Crystallize starts from a different premise. In many important decisions, the user does not begin with a complete answer. They begin with fragments: values, aversions, images of a possible future, pressure from other people, constraints, and half-formed intuitions.
The research base
Decision science has been saying this for decades.
Research on preference reversals, constructed preferences, adaptive decision strategies, miswanting, stated-choice bias, and user modeling all points toward the same conclusion: the way a question is asked can change the preference that becomes available to answer it.
1980s
Decision and valuation research showed how expertise, framing, and hypothetical contexts affect stated values.1990s
Constructive processing research described preferences as assembled through task demands, context, and cognitive strategy.2000s
Adaptive interface and recommender-system research made user modeling central to digital decision support.2020s
AI alignment and human-feedback research made the relationship between model behavior and human preference newly urgent.The Crystallize model.
Crystallize treats preference formation as a structured process. A session is not trying to guess the right answer. It is trying to create the conditions under which the user can recognize, test, correct, and stabilize their own answer.
The system works with what the user already has: value fragments, emotional reactions, negative-space knowledge, analogies, social signals, constraints, and aspirations. None of these is yet a full preference. Together, they are the raw material from which a preference can form.
How preferences form
Externalization.
A vague thought becomes visible enough to inspect. Seeing it outside your head makes reaction possible.
Contrast.
Preferences often become clearer through comparison than through abstract self-description.
Articulation.
Putting felt preference into language stabilizes it enough to be corrected, carried, and used.
Product translation
The product turns research into a quiet loop.
Crystallize uses concrete questions, short reflections, correction, and exportable artifacts to help a user move from raw material to usable language. The interface is not a passive container. It is part of the formation environment.
Probe
A small, concrete question or prompt gives the user something real to react to.Mirror
The session reflects back what it heard in plain language.Correct
The user can reject, sharpen, or redirect the interpretation before moving on.Artifact
The final output is something the user can save, revisit, or bring into a real conversation.Why careers first.
Early career decisions are unusually good tests of preference formation. They are high-stakes, socially loaded, multidimensional, and often made before the person has much lived experience with the domain.
Job boards, recommendation engines, personality tests, and advice from well-meaning people usually operate downstream. They help execute a preference that is assumed to exist. Crystallize works one layer earlier: helping the user form language around what kind of work, environment, manager, pace, and tradeoff may actually fit.
Ethical commitments
A formation system has to be careful.
Any system that helps people form preferences has power. Crystallize is designed around user authorship: the system can ask, mirror, and structure, but it should not steer the user toward a conclusion that serves the product.
- Narrative sovereignty
- The user owns the story that emerges. The system does not rank or judge their life direction.
- Correction as design
- A wrong reflection is not a failure. It is a useful moment if the user can correct it easily.
- Designed to end
- The goal is a clearer user, not a longer session. The artifact remains after the product stops.
Selected references.
Alba, J. W., & Hutchinson, J. W. (1987). Dimensions of consumer expertise. Journal of Consumer Research.
Ariely, D. (2008). Predictably Irrational. HarperCollins.
Fischhoff, B. (1991). Value elicitation: Is there anything in there? American Psychologist.
Gilbert, D. T., & Wilson, T. D. (2000). Miswanting. In Feeling and Thinking.
Green, P. E., & Srinivasan, V. (1990). Conjoint analysis in marketing. Journal of Marketing.
Kahneman, D., & Tversky, A. (1979). Prospect theory. Econometrica.
Lichtenstein, S., & Slovic, P. (1971). Reversals of preference between bids and choices. Journal of Experimental Psychology.
Louviere, J. J., Hensher, D. A., & Swait, J. D. (2000). Stated Choice Methods. Cambridge University Press.
Payne, J. W., Bettman, J. R., & Johnson, E. J. (1992). Behavioral decision research: A constructive processing perspective. Annual Review of Psychology.
Payne, J. W., Bettman, J. R., & Johnson, E. J. (1993). The Adaptive Decision Maker. Cambridge University Press.
Pennebaker, J. W. (1997). Writing about emotional experiences as a therapeutic process. Psychological Science.
Resnick, P., & Varian, H. R. (1997). Recommender systems. Communications of the ACM.
Slovic, P. (1995). The construction of preference. American Psychologist.
Thaler, R. (1980). Toward a positive theory of consumer choice. Journal of Economic Behavior & Organization.