Which Foods May Be Addictive? The Roles of Processing, Fat Content, and Glycemic Load

Not all foods feel the same in our heads and bodies. Some trigger cravings, loss of control, and relapse-like patterns, while others barely register as tempting. That asymmetry drives the work by Erica Schulte, Nicole Avena, and Ashley Gearhardt on food addiction. They claim that certain highly processed foods may share meaningful properties with drugs of abuse. The key word is pharmacokinetic, which describes how a substance moves into and through the body. Schulte and colleagues argue that two features matter most: dose and speed. Processing concentrates rewarding ingredients into a compact, potent form. Refined carbohydrates get absorbed rapidly, producing a fast, large shift in blood sugar. Glycemic load, or GL, is the metric they use to index that rapid absorption. It captures both how much refined carbohydrate a food contains and how quickly those carbs enter the bloodstream. Concentrated dose plus rapid uptake: that's the profile shared by drugs of abuse and, potentially, some of what we eat. To study addictive-like eating in humans, the authors used the Yale Food Addiction Scale, or YFAS. This is a 25-item self-report measure that operationalizes disordered eating using DSM-IV substance-dependence criteria. Crucially, they removed the standard YFAS instruction to think about foods high in fat or refined carbohydrates. Instead, they replaced it with a neutral prompt to think about any food you've had a problem with in the past year. That small change kept the measurement clean. The paper ran two studies in sequence. Study One recruited one hundred and twenty undergraduates, with a mean age of nineteen and a mean body mass index of twenty-three. It put them through a forced-choice task using a bank of thirty-five foods. Each food was pitted against every other food, one pair at a time, and participants selected which of the two they were more likely to have problems with, such as trouble cutting down or loss of control. With thirty-five foods, each item could be chosen up to thirty-four times, giving the researchers a clean frequency ranking. Study Two scaled up to three hundred and eighty-four adults recruited via Amazon Mechanical Turk, with a mean age of thirty-one and a mean BMI of just under twenty-seven. Instead of forced choices, participants rated all thirty-five foods on a scale from one to seven for how likely each was to be associated with YFAS-style problems. That rating data fed into two-level hierarchical linear modeling, which is a statistical technique that accounts for the fact that one person's thirty-five ratings aren't independent of each other, while simultaneously testing food-level predictors and participant-level moderators. The thirty-five foods were selected deliberately to span a wide range of processing levels, fat content, and glycemic load. This approach allowed the researchers to test those attributes as predictors rather than just assume they mattered. Then came the reveal. In Study One, the most frequently nominated problematic food was chocolate, chosen an average of twenty-seven point six times out of a possible thirty-four. Ice cream came second at twenty-seven, and French fries were third at twenty-six point nine. Pizza, cookies, chips, and cake all clustered in the top tier. Every food at the top of the list was highly processed, with added fat and refined carbohydrates. At the bottom sat water, cucumber—specified “no dip”—and broccoli, each chosen fewer than seven times. Plain beans, brown rice, and carrots were similarly low. The contrast is almost cartoonish, but the pattern fits the hypothesis perfectly: processing status sorts foods cleanly into the high-problem and low-problem ends of the ranking. Study Two turned that observation into quantified predictors. Processing was the dominant food-level attribute. In the hierarchical model, the average rating for an unprocessed food was two point two four on the seven-point scale. Being highly processed added zero point six five points, an effect the authors classify as large, with a Cohen's d of one point four four. Individual differences moderated that effect, but more modestly. Each additional YFAS symptom a participant endorsed was associated with a zero point zero six three point increase in their processed-food ratings, a small to moderate effect with a d of around zero point three two. Body mass index added a similarly modest amplification: each BMI unit raised processed-food ratings by zero point zero one two points. Since processing correlates strongly with both fat content and glycemic load—the correlation between processing and GL was zero point seven six—the authors ran a separate model replacing processing with its nutritional components. Fat was a large predictor, with a d of one point five eight. Glycemic load was also large, with a d of zero point nine two. Sodium was significant too, but its effect size was smaller than fat's. The picture that emerges is coherent: it's not one magic ingredient. It's the combination of added fat and rapidly absorbed refined carbohydrates, which are the defining features of industrial processing, that drives the association with addictive-like eating. Now here's where the findings get personal. YFAS symptom count didn't just predict higher ratings overall; it specifically amplified sensitivity to glycemic load. People with more addiction-like eating symptoms showed a stronger relationship between a food's GL and how problematic they rated it. The GL interaction coefficient was small in absolute terms: zero point zero zero three per additional symptom, but statistically reliable, with a p-value of zero point zero zero four and a d of zero point three. Schulte and colleagues interpret this as individuals who already show compulsive eating patterns being more reactive to the specific property of rapid carbohydrate absorption. They’re not just eating more overall; they’re being disproportionately drawn to the foods that hit hardest and fastest. Body mass index told a different part of the story. It predicted greater problems with highly processed foods but did not drive the GL-sensitivity effect. Those are two separable vulnerability signals: one tied to body mass and processed food broadly, and another tied to addiction-like symptom load and glycemic speed specifically. Gender effects were narrow; men reported slightly more problems with unprocessed foods than women, but gender didn't meaningfully moderate the processing or GL effects. Rates of full food addiction—meeting the YFAS threshold—were six point seven percent in the undergraduate sample and ten point two percent in the Mechanical Turk sample. Those aren't huge numbers, but they're significant. They sit alongside a continuous distribution of symptoms, where even people below the threshold show graded vulnerability. The authors are clear about what this study can and can't establish. The design is cross-sectional—a snapshot, not a trajectory. The outcome measures are self-reported ratings and symptom counts, not direct observation of eating behavior. Height and weight were self-reported, introducing the usual measurement noise in BMI. Schulte, Avena, and Gearhardt explicitly call for future studies that measure biological responses and directly observe eating to test whether these foods produce the hallmark mechanisms of addiction—specifically withdrawal and tolerance. Those phenomena have been documented in animal models but haven't been cleanly established in humans eating processed foods in natural conditions. What this study contributes is a sharper target. Not "food addiction" as a vague metaphor applied to all overeating, but a specific profile: high processing, high fat, high glycemic load. If that profile drives addictive-like behavior, then the open question isn't whether some people eat compulsively; it’s whether the food itself is doing something pharmacologically distinct. The pharmacokinetic framing shifts attention from the person's willpower to the measurable properties of what's on their plate. That's a different conversation, and it has direct implications for dietary guidance, food labeling, and policy—if we can confirm what the mechanism actually is. The data say the foods at the top of the addiction chart are not random. They're the ones engineered to deliver fat and refined carbohydrates together, fast. Whether that's enough to call them drugs is a question for future work. But it's no longer reasonable to treat a cucumber and a chocolate bar as equivalent stimuli in the neuroscience of eating. This lecture was created by ennepō. Go to https://ennepo.ai to Discover, Create and Follow the latest research in your field. Read when you can. Listen when you want to.