Why Your Food Preferences Make New Diet Rules Harder to Learn

Swapping a beloved snack for a healthier option may be harder than you think, and not just because of willpower. Researchers at Yale University have found that established food preferences actively interfere with the brain's learning machinery, distorting the process by which people learn new value associations for foods they already love or loathe. The study, published in Communications Psychology in October 2026, also shows that the very first experiences in a learning sequence leave a lasting imprint on how flexible that learning becomes, according to the paper.

The experiment was personalized in a way that most value-learning studies are not. The team, led by Alexandra Rich and Ifat Levy, recruited 415 participants online through the Prolific platform across two independent cohorts, with the final analysis covering 279 people. Each volunteer first rated 65 snack images from the Food Folio database, ranging from granola bars to croissants, on a four-point scale from strongly dislike to strongly like. From those ratings, the researchers built a personal set of four foods for every participant: one strongly liked, one liked, one disliked, and one strongly disliked. The team verified that these ratings did not track the foods' caloric or nutritional content, confirming that the experiment captured genuinely subjective values rather than objective properties of the snacks.

Participants then played a reversal-learning game. One of their four personal foods appeared on screen, and they had three seconds to predict whether it would be followed by a positive outcome of plus five points or a negative outcome of minus five points, across six blocks of 32 to 36 trials each. In Congruent blocks, the outcomes matched participants' existing preferences nine times in ten. In Incongruent blocks, the relationship flipped, forcing people to learn associations that ran directly against their priors. Participants knew the rules would switch, but they were never told when. This is a sharp contrast with earlier value-learning studies that used neutral shapes such as triangles or rectangles: in those experiments, people could switch between assigned values with ease. Here, as Levy noted, everyone struggled to go against their feelings.

Reinforcement first makes learning rigid

The most striking result was an order effect. Participants who began with a Congruent block — meaning their preferences were reinforced right away — achieved exceptionally high accuracy on later Congruent blocks, averaging 0.73, while dropping to 0.59 on Incongruent blocks. Those who started with a block that challenged their priors performed similarly across both block types, at around 0.62. The interaction replicated across both independent cohorts, which makes the finding unusually robust for a behavioral study. The researchers propose that an initial confirmation of the brain's existing associations acts like an anchor: once validated, those links become harder to overturn later, even though participants explicitly knew the rules would change. Possible mechanisms include confirmation bias, increased confidence, and egocentric discounting of later evidence.

"Our pre-existing preferences for food are very strong and hard to change," Levy said in a statement about the findings. "But there might be slight manipulations that could make a big difference."

Adding beats subtracting

The study also uncovered an asymmetry between liked and disliked foods. For liked items, prediction accuracy was consistently higher in Congruent blocks than in Incongruent blocks, regardless of which block type participants started with — positive priors made these associations stubbornly inflexible. Disliked foods behaved differently: people performed better in whichever condition they started in, suggesting that associations with non-preferred foods were far more suggestible to the surrounding environment. In plain terms, it may be considerably easier to add a disliked food to your diet than to remove a liked one. "Say, broccoli isn't your favorite thing. Maybe we can integrate it more easily than to stop eating ice cream," Rich said.

A separate positivity bias ran through all the data. Across every condition, participants predicted reward-associated outcomes more accurately than punishment-associated ones, 0.69 versus 0.61, even though the point outcomes were arbitrary and carried no real-world value beyond a small bonus tied to overall accuracy. Translated to the dinner table, the finding hints that emphasizing the rewarding aspects of healthy foods may shape associative learning more effectively than warnings about negative health consequences. And among participants who began with a challenging block, willingness to pay, enjoyment, and expected satisfaction for disliked foods measurably increased after the task — a cumulative shift in valuation that could be harnessed by interventions aimed at reshaping eating habits.

For anyone trying to change how they eat, the practical takeaway is about sequencing. Dietary interventions routinely ask people to overturn deeply held food values, yet the study suggests that leading with experiences that reinforce old preferences builds rigidity into the very learning process those interventions depend on. Programs that instead start by gently challenging existing preferences could produce more flexible food-value learning. That matters in everyday life too: trying a new dish before reaching for the usual favorite may do more than satisfy curiosity — it may keep your learning machinery open to change.

There are limits to what the study can claim. The outcomes were arbitrary points rather than motivationally salient rewards, and the researchers cannot yet say whether the findings extend to other strongly valued things or to real dietary change over weeks and months. An exploratory link between higher binge-eating scores and poorer learning on some trial types was weak and needs replication. Still, the central message is clear: we rarely learn about value in a naive environment, and our prior preferences confound the process from the very first trial.