Precision nutrition is NIH’s big bet, and the results aren’t in yet

Key takeaways

• The NIH’s Nutrition for Precision Health study plans to enroll 10,000 adults and use artificial intelligence to predict how each person responds to food.
• Up to about 3,000 of those volunteers will eat three researcher-chosen diets for about two weeks each, some while living at a research center.
• In an NIH study, the same meal served to the same person a week apart produced blood sugar responses too inconsistent to guide personal advice.
• An NIH funding page reviewed in October 2026 projects $122 million for All of Us, the program the study is built on, in fiscal 2027, nearly 80% below its 2023 level.
• Precision nutrition already works for a few single-gene conditions, such as PKU, which newborn screening catches early.

At the University of North Carolina, volunteers in a federal diet study go home with a stool-sample kit and a wearable device. Over 8 to 10 days at home, they log their food on some days, photograph their meals on others and collect one stool sample⁵.

Then comes a clinic visit that can last five to eight hours. Staff do a physical exam, draw blood, collect saliva and serve a standard test meal that resembles a smoothie⁵. Volunteers can earn up to $300⁵.

That routine is the front door of Nutrition for Precision Health, a study the National Institutes of Health (NIH) announced in 2022 with $170 million over five years². Its goal is to build computer models that predict how a specific person will respond to specific foods and eating patterns¹.

The idea will sound familiar to many natural-health readers: each body is different, so the best diet differs too. What’s new is a federal effort to test that idea at scale and find out which differences matter.

As of its May 2026 review, the program’s homepage was still inviting volunteers to enroll¹. Here is what the study measures, what earlier research suggests, and why answers will take time.

Precision nutrition means more than a DNA-based diet

“Precision nutrition” is the central theme of the NIH’s 2020–2030 strategic plan for nutrition research, which the agency updated in June 2025⁶. The plan describes it as advice that is comprehensive and changes over time.

That definition goes well beyond genes. It includes your eating patterns, your daily body clock, your health status, the food available where you live, your physical activity and your gut microbes⁶. The plan contrasts this with nutrigenomics, a field focused on genes and similar traits that mostly don’t change over time⁶.

The plan frames its four goals as plain questions⁶:

• What do we eat, and how does it affect us?
• What and when should we eat?
• How does what we eat promote health across a lifetime?
• How can we improve the use of food as medicine?

In other words, “precision” is meant to answer what, when and for whom, not just which gene variant you carry.

Inside the study: three levels of control

The study recruits adults 18 and older who are enrolled in All of Us, a large NIH program that gathers health records, genetic data and surveys from volunteers³. It is the first add-on study to use All of Us infrastructure¹. Six clinical centers run it at 14 sites across the country³.

Volunteers take part at one of three levels. Each level adds more control over what people eat.

Level	Who takes part	What happens	Length
Module 1	About 10,000 people	Surveys, daily diet reports, blood, urine and stool samples, and a test meal34	Up to 14 days, with two clinic visits4
Module 2	1,500 to 2,000 people from Module 1	Three researcher-chosen diets eaten at home, with a test meal after each4	About 14 days per diet4
Module 3	500 to 1,000 people from Module 1	The same three diets, eaten while living at a research center4	About 14 days per diet4

The three diets contain the same number of calories but differ in meaningful ways. They are not designed for weight loss, and there is no control diet⁴. That makes sense for the goal: researchers want to compare how each person responds to different diets, not test one diet against nothing.

Living at a research center, often called a domiciled study, matters because people misremember and misreport what they eat¹⁰. When the kitchen controls every bite, researchers know exactly what went in.

Volunteers get some of their own results back, including body fat percentage, gut microbe makeup, metabolism and diet composition³. The data will go to approved researchers through the All of Us Researcher Workbench²⁴.

“Nutrition is perhaps one of the most powerful medicines we have available, but is among the least understood,” said Geoffrey Ginsburg, chief medical and scientific officer of All of Us, when enrollment opened³.

Each research tool sees part of the picture

The study’s power comes from combining measurements. Each one has a blind spot.

Metabolomics. This means measuring thousands of small molecules in blood and urine, the chemical traces of food and of the body’s own processes. The study runs both targeted tests, which look for known molecules, and untargeted tests, which cast a wide net, on all samples⁴. The NIH plan calls the many unidentified molecules nutrition’s “dark matter”⁶.

Gut microbes. Stool samples get a basic genetic survey of their bacteria, and a subset gets deeper sequencing that can show what the microbes may be doing⁴. Our article on home gut tests explains why consumer versions of these readings can disagree.

Wearables and glucose. Volunteers wear a device during the at-home days⁵. The NIH plan calls for adapting continuous glucose monitors, small sensors that track blood sugar around the clock, for point-of-care testing, done on the spot rather than in a lab⁶.

Diet records. People report what they eat through questionnaires and photos⁵. Each method brings its own errors, including faulty memory, and different food databases can produce very different intake estimates¹⁰.

To fix that, a research group called the Dietary Biomarkers Development Consortium is hunting for chemical markers in blood and urine that reveal what someone ate¹¹. It feeds volunteers set amounts of test foods, looks for the molecular traces, then checks whether those markers work in ordinary people’s lives¹¹.

Artificial intelligence. Machine learning finds patterns across all these data. A 2026 review in Nature Communications warned that a model can predict well without showing cause, can latch onto quirks of one group, and must be tested on outside data before anyone trusts it¹⁰.

What earlier studies already suggest

Several large studies came before this one. Our article on personalized nutrition algorithms covers them in depth; here is the short version.

• People respond differently. In the 2020 PREDICT 1 study of 1,002 British adults, responses to identical meals varied widely. The meal’s mix of carbohydrate, fat and protein explained more of the blood sugar difference than genes did¹³.
• Gene-matched diets failed. In the DIETFITS trial of 609 adults, a gene pattern did not predict who lost more weight on low-fat or low-carb eating¹⁴.
• App advice helped a little. In a 2024 trial of the company ZOE’s app, 347 adults lost about 2.5 kilograms more over 18 weeks than a group given a leaflet, video and weekly check-ins, a comparison the authors said wasn’t matched for contact¹⁵.

A newer finding complicates the picture. NIH scientists served each meal twice, about a week apart, to adults living at the NIH Clinical Center, and tracked their blood sugar with two brands of sensors⁹.

The same person’s responses to the same meal agreed poorly. Researchers use a reliability score where 0.5 is a common minimum; the two sensors scored 0.31 and 0.14⁹. A person’s swings between identical meals were about as large as the swings between different meals⁹.

The authors concluded that advice built on just a couple of glucose readings is unlikely to be reliable, even under tightly controlled conditions⁹. The version we reviewed was posted on medRxiv, a server for preprints, which are studies shared before peer review⁹.

Precision nutrition already works when one cause dominates

Precision nutrition isn’t science fiction. It has saved lives for decades in a few conditions.

Phenylketonuria, or PKU, is an inherited condition in which the body can’t break down phenylalanine, a building block of protein. It occurs in about 1 in 25,000 U.S. newborns¹². Untreated, it can cause intellectual disability and seizures¹².

A carefully tailored low-phenylalanine diet prevents that harm, and the safe amount varies from person to person¹². Because newborn screening catches most cases right after birth, the severe form is now rarely seen¹².

PKU works because a single gene and a single nutrient explain most of the problem. Common conditions like heart disease, obesity and type 2 diabetes involve many small influences that shift over time. That is why the NIH needs thousands of volunteers, repeated test meals and machine learning to make progress.

Four obstacles stand between the study and your plate

Money. All of Us has relied on yearly federal budgets plus a special fund from the 21st Century Cures Act, a 2016 law, which covered fiscal years 2017 through 2026⁷. An NIH funding page, last reviewed October 7, 2026, projects $122 million for the program in fiscal 2027, nearly 80% below its 2023 level⁷. The same page says the cuts have reduced enrollment and scaled back data collection⁷.

The precision nutrition study has its own five-year award, pending available funds². But it relies on All of Us infrastructure¹, so the parent program’s finances matter.

Who gets studied. The study aims to reflect the diversity of the U.S. population⁴. It launched with English-only materials and planned Spanish ones³; the UNC site now enrolls English or Spanish speakers⁵.

Representation matters because models can mislead the people they skip. The 2026 Nature Communications review noted that training data often come from well-resourced groups. A model built on part of All of Us might work there yet fail for underrepresented groups¹⁰.

Privacy. All of Us strips names and other direct identifiers before researchers see data, and it holds federal Certificates of Confidentiality to resist legal demands for identifying information⁸. Researchers must promise not to try to identify participants, and the program’s principles bar selling data or using it for targeted ads⁸. Commercial diet apps follow their own rules, which our article on health app privacy explains.

Proof. Predicting a blood sugar curve is not the same as preventing a heart attack. The NIH plan makes rigor and reproducibility a cross-cutting priority⁶, and the duplicate-meal study shows why: a measurement that won’t repeat can’t anchor a personal plan⁹.

Your body’s individuality, now a federal research question

Natural-health traditions have long held that people differ and that food choices should fit the person. The NIH’s plan now rests on the same premise⁶.

That’s an opening, not a verdict. Research can confirm which differences matter, such as PKU, and which are noise, such as a single glucose spike⁹¹². That kind of testing protects you from paying for precision that isn’t real.

Your own experience counts, too. How full you feel, how you sleep and how your digestion responds are information worth recording. The trick is collecting it carefully.

A careful personal food test you can run now

You don’t need a lab to learn from your own meals. You need patience and repetition. This routine borrows the study’s logic.

1. Pick one question. For example: “Does an egg breakfast keep me fuller until lunch than oatmeal does?”
2. Hold the rest steady. Keep sleep, activity and meal timing as similar as you can.
3. Repeat each option several times on different days. NIH researchers found that two tries of the same meal gave unreliable results⁹.
4. Track what you care about. Note fullness, energy, digestion and sleep. If you check blood pressure or weight, look at weekly averages.
5. Decide in advance what would change your mind. Write it down before you start.
6. Judge the pattern, not one day. One great morning or one spike proves little.
7. Bring in help when health is on the line. If you take insulin or other diabetes medicines, have kidney disease, are pregnant, or have a history of disordered eating, plan changes with a clinician or registered dietitian.

What the evidence supports

Claim	Evidence level	What the research shows
People respond differently to the same foods	Strong evidence	Wide variation in responses to identical meals13
Genes explain most of your response to food	Evidence says no, or unsafe	Meal makeup outweighed genes; gene-matched diets failed1314
A single tailored diet prevents harm in PKU	Strong evidence	Newborn screening plus diet keeps severe PKU rare12
One or two glucose readings reveal your response to a meal	Not enough evidence	Poor repeatability in an NIH preprint study9
App-based personalization beats minimal advice short term	Promising but early	Modest gains against a leaflet-style control15
Precision-built diets improve long-term health	Not enough evidence	Trials so far are short; NIH models still in development115
Chemical markers can measure what people eat	Promising but early	Federal consortium is discovering and testing them11
AI diet models work equally well for everyone	Not enough evidence	Experts warn models may fail outside their training groups10

Questions readers ask

The bottom line

The NIH is running one of the most ambitious tests yet of an idea natural-health readers already hold: that food affects each body differently. It’s measuring blood, stool, molecules and meals in thousands of people, and it’s building the tools to make sense of them.

Those tools aren’t ready, and the program’s funding is under strain. Until results arrive, be wary of anyone selling finished precision, keep testing your own habits patiently, and lean on the basics that have held up across trials.

Related reading

• Your response to food is personal, but the payoff from diet algorithms is modest
• Glucose sensors catch every spike, and in healthy people most spikes are normal
• Same stool, different verdicts: what home gut microbiome tests can tell you
• Health apps share more than you think, and you can lock yours down
• DNA kits read a sliver of your genes, and the data can be sold

Sources

1. NIH Common Fund. “Nutrition for Precision Health, powered by the All of Us Research Program” (program homepage). NIH, last reviewed May 27, 2026. commonfund.nih.gov
2. National Institutes of Health. “NIH awards $170 million for precision nutrition study.” NIH news release, January 20, 2022. nih.gov
3. NIH Common Fund. “NIH launches largest precision nutrition research effort of its kind.” NIH, 2023, last reviewed November 14, 2025. commonfund.nih.gov
4. NIH Common Fund. “Nutrition for Precision Health: Frequently Asked Questions.” NIH, last reviewed December 5, 2024. commonfund.nih.gov
5. UNC Nutrition Research Institute. “Can a personalized diet help improve health and prevent chronic diseases?” University of North Carolina, 2024, updated 2026. uncnri.org
6. NIH Nutrition Research Task Force. “2020–2030 Strategic Plan for NIH Nutrition Research” (updated June 2025). NIH. dpcpsi.nih.gov
7. National Institutes of Health. “All of Us: Funding and Research Partnerships.” NIH, last reviewed October 7, 2026. nih.gov
8. National Institutes of Health. “All of Us: Data and Privacy.” NIH, last reviewed June 16, 2026. nih.gov
9. Hengist A, Ong JA, McNeel K, Guo J, Hall KD. “Imprecision nutrition? Duplicate meals result in unreliable individual glycemic responses measured by continuous glucose monitors across four dietary patterns in adults without diabetes.” medRxiv preprint, 2023. pubmed.ncbi.nlm.nih.gov
10. Massara P, Kirkland J, Pagani I, et al. “Applying Artificial Intelligence and machine learning in precision nutrition.” Nature Communications, 2026. nature.com
11. Chakraborty H, Sun Q, Bhupathiraju SN, et al. “The Dietary Biomarkers Development Consortium: An Initiative for Discovery and Validation of Dietary Biomarkers for Precision Nutrition.” Current Developments in Nutrition, 2025. pubmed.ncbi.nlm.nih.gov
12. MedlinePlus Genetics. “Phenylketonuria.” National Library of Medicine, updated April 25, 2023. medlineplus.gov
13. Berry SE, Valdes AM, Drew DA, et al. “Human postprandial responses to food and potential for precision nutrition.” Nature Medicine, 2020. nature.com
14. Gardner CD, Trepanowski JF, Del Gobbo LC, et al. “Effect of Low-Fat vs Low-Carbohydrate Diet on 12-Month Weight Loss in Overweight Adults and the Association With Genotype Pattern or Insulin Secretion: The DIETFITS Randomized Clinical Trial.” JAMA, 2018. pubmed.ncbi.nlm.nih.gov
15. Bermingham KM, Linenberg I, Polidori L, et al. “Effects of a personalized nutrition program on cardiometabolic health: a randomized controlled trial.” Nature Medicine, 2024. nature.com