A patient without diabetes came in recently with her phone open to a chart. For three weeks, she had worn an over-the-counter glucose sensor and logged every meal in an app that promised a diet tuned to her biology. The app had flagged her morning oatmeal and her lentil soup as “spike foods,” so she had stopped eating both. Her labs were normal. What she wanted from me was permission to keep avoiding them.
Many of you have had a version of this visit. In 2024, the FDA cleared the first over-the-counter continuous glucose monitor for adults who don’t use insulin, and a fast-growing industry has built AI-driven apps on top of it. Patients now arrive with readings, food scores, and an algorithm’s verdict.
I’m an endocrinologist who runs clinical trials on diet and metabolic disease, and I recognize the pitch. We’ve seen it before: the blood-type diet in the 1990s, mail-order DNA diet tests in the 2000s, gut microbiome kits in the 2010s. The technology improves each time. The promise stays the same: Your body is one of a kind, population-level advice doesn’t apply to you, and for a monthly fee, here is your true diet.
The real finding, and the stretch
There is real science underneath. A widely cited 2015 study in Cell showed that different people’s glucose responses to the same meal can vary widely, and that a model using gut microbiome and lifestyle data could predict some of that variation. The finding has held up reasonably well. People do differ.
Real physician voices, twice a week
Free, and one click to unsubscribe.
The problem is what’s been built on top of it. Three ideas nutrition science spent decades testing and setting aside have come back as app features.
The “personal” plan is mostly the standard plan. Look at where these apps end up: more fiber, more whole plant foods, less ultra-processed food. The Dietary Guidelines have said that for years, for free. If an app helps a patient stick with those habits, that has value, because adherence is the hard part. But it’s a much smaller product than the one being sold.
A postprandial rise isn’t pathology. In someone with normal glucose tolerance, blood sugar rising after a meal is normal physiology. When every bump on a screen is flagged as a problem, patients start cutting beans, oats, and whole grains, foods with some of the strongest evidence in all of nutrition science, because those foods move a line on a graph. That is exactly what happened to my patient.
The signal is also noisier than the app suggests. When researchers gave adults without diabetes the same meal on two separate occasions, their glucose responses often differed. An accompanying commentary in the American Journal of Clinical Nutrition put the problem in its title: personalized nutrition by prediction of glycemic responses, “garbage in, garbage out.” A diet built to chase an unstable input rests on shaky ground.
Sophisticated isn’t the same as proven. Diets built around glucose predictions have not been shown to produce better long-term health than sound general dietary advice. There’s also a simpler explanation for why many users feel better: People eat more carefully when they know their food is being tracked. The sensor and the food log may be doing much of the work, not the algorithm.
What I tell patients
When a patient brings in app data, I don’t wave it away. I do three things.
First, I reframe the spike. A rise after a bowl of oatmeal that comes back down within a couple of hours is the system working, not failing. What tells us about long-term risk is fasting glucose, A1C, weight, blood pressure, and the overall pattern of the diet, not a single curve.
Second, I ask what the app told them to stop eating. That’s usually where the harm is. If the answer is legumes, intact whole grains, or fruit, I explain that these foods are consistently linked with lower risk of type 2 diabetes and heart disease, in large cohort studies and in randomized trials, including the trials of plant-based diets my colleagues and I run.
Third, I keep what’s useful. If logging meals has made a patient more aware of what she eats, I encourage her to keep the habit, with or without the sensor.
Where personalization earns its place
None of this is an argument against the technology. Continuous glucose monitors have transformed care for people on insulin, and tailoring diet to the individual matters in diabetes, food allergies, and inflammatory bowel disease. I expect AI to do remarkable things in medicine.
My concern is narrower. In the consumer market, “personalized nutrition” is becoming a way to sell certainty and to recast a population problem as a private one. Most of our patients aren’t sick because no one decoded their individual biology. They’re sick in large part because the food around them is engineered to be overeaten. A sensor and a subscription won’t change that.
The most striking thing about nutrition research isn’t how much it varies from person to person. It’s how consistent it has been. Across countries, decades, and study designs, diets built mostly around whole, minimally processed plant foods are linked to better metabolic health.
I told my patient to keep the app if it helps her pay attention, and to put the oatmeal and lentils back. Her biology is unique. The basics of a healthy diet mostly aren’t.
Hana Kahleova is an endocrinologist.
