Microbiome tests with a diet plan

What microbiome tests are and how they work

You send in a stool sample, the lab determines the composition of your gut bacteria, and a model derives recommendations from this about which foods are supposed to suit you. Some offerings also include genetic information.

The idea behind it: how strongly blood sugar and blood fats rise after a meal differs from person to person, and part of these differences is linked to the microbiome. If you can predict your own response, the idea goes, you can align your diet with it. In PREDICT 1, such a model predicted the blood sugar response with a correlation of 0.77, the blood fat response with 0.47.

What the studies show

Berry 2020 (PREDICT 1) — the same meal, very different responses

1,002 healthy adults in the United Kingdom, twins and unrelated people, received identical test meals. The variation in response was 103 percent for blood fats, 68 percent for blood sugar and 59 percent for insulin. For blood fats, the microbiome contributed 7.1 percent of the variance, the macronutrients of the meal 3.6; for blood sugar it was the other way round, 6.0 versus 15.4 percent. Genetic variants improved the prediction by 9.5 percent for blood sugar, 0.8 for blood fats and 0.2 for C-peptide. The prediction model reached a correlation of 0.77 for blood sugar and 0.47 for blood fats. This was confirmed in 100 people in the US.

Servetas 2026 — seven providers, the same material

Researchers at the US standards institute NIST sent a standardized human stool material they had developed to seven providers of direct-to-consumer microbiome tests. The results differed markedly from one another, between providers and also within individual providers. The differences between the companies were of the same order of magnitude as the biological differences between different donors. The authors attribute this to differing methods and a lack of quality control.

Where the data stop

PREDICT 1 shows that the individual response is real and can be partly predicted. The study does not show that a diet plan from a purchased test improves health. It is precisely this leap from research to product that is the reason for the deduction in the hype gap.

In addition, there are three limitations that matter before paying. First, the contribution of genes is small: 9.5 percent for blood sugar, 0.8 for blood fats, 0.2 for C-peptide. Second, for blood sugar the meal clearly beats the microbiome, 15.4 versus 6.0 percent. Third, the measurement is not reliably reproducible between providers. If the same sample yields different findings at two companies, every recommendation built on it stands on uncertain ground. The NIST authors state that reliable measurement is the prerequisite for well-founded recommendations.

Status, approval and legal

In the assessment of the NIST authors, direct-to-consumer microbiome tests sit between strictly regulated medical devices and barely regulated wellness products, and this difference is often not apparent to buyers. The authors see a need for standards so that the measurements become reliable.

Three questions are worth asking before buying: Which studies does the model rely on, and were they tested on the product itself? How does the lab assure the quality of its measurement, and does it name its reference group? And does anything follow from the result that you would not also do without a test?

Safety

The test itself carries no physical risk; a stool sample is non-invasive. The question of safety shifts to what follows from the result. Besides analytical and clinical validity, the NIST authors explicitly name consumer protection as an open question. A test report is not a diagnosis. Anyone with persistent digestive complaints should have them checked by a doctor instead of relying on a diet plan from a consumer test.

Sources

• Berry et al., Nat Med 2020 (PREDICT 1, individual responses to meals)
• Servetas et al., Commun Biol 2026 (NIST evaluation of seven test providers)