As AI becomes better at predicting what people want, product designers face a growing challenge: preserving the sense of agency that makes choices feel rewarding.
Personalisation is now one of the central promises of consumer technology, but its logic contains a tension that product teams often understate: the better a system becomes at anticipating preference, the less room it leaves for the user to choose. That trade-off matters because choice is not merely instrumental. Research by Lauren Leotti and Mauricio Delgado showed that even the anticipation of making a decision activates reward-related brain circuitry, which helps explain why a recommendation engine can feel efficient yet still leave a user oddly less satisfied than a decision they made themselves.
The deeper issue is that systems which decide on a person’s behalf can quietly remove some of the motivation that makes an experience feel owned. Studies from Rutgers and work within self-determination research point in the same direction: people do not only value the result, but also the act of deciding. Once a product turns a live decision into a silent prediction, it can reduce the sense of control that supports engagement, attachment and long-term satisfaction. That is why a highly accurate system may still feel flat if it leaves the user as a passive recipient rather than an active participant.
Evidence from music streaming illustrates the practical consequences. In research published by Spotify, algorithm-driven listening was associated with less diverse consumption, as recommendation systems tended to reinforce established tastes rather than broaden them. The same work linked that narrower listening pattern to retention and premium conversion, which shows why the commercial case for automation is so strong. But it also highlights the strategic risk: if a platform optimises too hard for immediate relevance, it can encourage behaviour that looks successful on a dashboard while gradually shrinking exploration and weakening the user’s relationship with the service.
That makes the design problem less about eliminating personalisation than about constraining it. The evidence on choice overload matters here as much as the evidence on prediction. Classic behavioural research by Sheena Iyengar and Mark Lepper showed that too many options can suppress decision-making altogether, so the answer is not to force people through endless menus. The more durable approach is calibrated agency: enough structure to make choosing manageable, but enough control for the user to feel that the system is helping rather than steering. In that sense, the real design task is to preserve meaningful decisions while automating only the parts of the journey that are genuinely tedious.
For companies building recommender systems, shopping agents or adaptive feeds, the implication is straightforward. Personalisation should be treated as a support layer, not a substitute for authorship. Users generally want assistance, but they also want visibility into how that assistance works and the ability to override it. Products that leave room for editing, comparison and deliberate choice are more likely to sustain trust because they preserve the psychological value of deciding, not just the efficiency of arriving at a result. That balance is increasingly important as AI systems become more agentic and begin acting further ahead of the user’s own judgement.
Disclaimer: This content is intended for informational purposes only. Readers are advised to exercise their own judgement, conduct due diligence, or consult a qualified expert before acting on any information provided.
