Research & Evidence

Evidence snapshot

Counted 2026-10-22 using NCBI E-utilities. Counts include studies of extracts, blends and non-health topics, and a trial can find no effect. A higher count means more human research has been indexed, not that something works.

For a plainer walk through this table, see Research by food.

What the counts do and do not tell you

• Large counts can include work that is not about eating the food. The soy, legume, tomato and citrus searches are broad word searches (the tomato search also includes lycopene, for example), so they are not counts of food trials. [1]
• Fermented foods have far fewer trials than the probiotic capsule literature. In the snapshot, the probiotic search has 3,369 records in the randomized-trial column, yogurt 588, kefir 36, kimchi 9 and kombucha 6. Many probiotic trials test capsules, not food. [1]
• Words like "superfood" are search topics in their own right. The snapshot finds 6 randomized-trial records for the term. A review of human trials of foods described as superfoods found 113 trials across 17 foods, but "only limited evidence" for effects on metabolic syndrome measures, because results were inconsistent or trials were few. [2]
• A trial tag is not a quality mark. PubMed's own definition of a randomized controlled trial is a study in which treatments are "selected by a random process"; it does not say the trial was large, blinded or well run. [3]

Three examples of what food trials look like

A trial of an eating pattern, with a funder to note. In the CORDIOPREV randomized trial, 1,002 people with coronary heart disease ate either a Mediterranean or a low-fat diet. The main outcome occurred in 87 of 502 people on the Mediterranean diet and 111 of 500 on the low-fat diet, and the benefit was clearer in men than in women. The investigators who judged outcomes were masked, but the participants were not, and the funders included foundations focused on olive oil. [4] It tests a whole pattern, not one food.

A trial that had to be withdrawn and re-analyzed. The PREDIMED trial assigned 7,447 people at high cardiovascular risk to a Mediterranean diet with olive oil, a Mediterranean diet with nuts, or advice to eat less fat. The authors later found that some people had not been randomized as planned, withdrew the 2013 report, and published revised estimates. [5] A Cochrane review later described the result for deaths as showing little or no effect, with a lower number of strokes. [6]

A trial with a null result. In the MIND diet trial, 604 older adults with a family history of dementia followed the MIND diet or a control diet, both with mild calorie restriction, for 3 years. Thinking scores and brain scans did not differ significantly between the groups. [7] The authors noted that observational studies had suggested protective benefits while trial data were limited. [7]

How to judge a food claim

1. Randomized trial or observational? See Observational studies vs trials.
2. Marker or outcome? A change in a blood measure is not the same as fewer heart attacks or strokes.
3. Whole food, juice, powder or supplement? A result for an extract does not transfer to the food.
4. What was it compared with? "Instead of" another food is a different question from "in addition to" the usual diet.
5. How large, how long, how many people?
6. Who paid for it? We did not check funding for every study named on this site.
7. Who is reporting it? A study of press releases from 20 UK universities found 40% contained exaggerated advice, and that exaggeration in news was strongly linked to exaggeration in the press release. [8]

For headline examples, see How to read nutrition research headlines. For what meta-analyses can and cannot add, see Human trials and meta-analyses of foods. Claims are checked on Food claim checks.