AI-driven diet app reshapes microbiota and cuts belly fat

The researchers reported a statistically significant increase in microbial richness (Chao1 index) and phylogenetic diversity (Faith’s PD), key indicators of a more diverse and resilient gut microbiota. Notably, there was a measurable restructuring of microbial composition, with 18 amplicon sequence variants (ASVs) found to increase and 17 to decrease. Among the most prominent changes were increases in the butyrate-producing genera Faecalibacterium and Eubacterium coprostanoligenes group, and reductions in potentially harmful genera like Eubacterium ruminantium group and Gastranaerophilales.

A six-week dietary intervention using an artificial intelligence-based mobile application significantly reshaped the gut microbiome of healthy adults and was associated with measurable improvements in waist circumference and dietary intake, according to a new study published in Nutrients.

Titled "The Influence of an AI-Driven Personalized Nutrition Program on the Human Gut Microbiome and Its Health Implications," the research provides emerging evidence that AI-powered personalized nutrition can directly influence human microbiota with potential long-term health benefits.

How does AI-based personalization impact the gut microbiome?

The study tested the PROTEIN mobile app, developed under an EU Horizon 2020 initiative, among 29 healthy participants aged around 35. Over six weeks, users received customized dietary and physical activity plans designed to promote Mediterranean diet adherence. The intervention was monitored via food frequency questionnaires, anthropometric measurements, physical activity assessments, and biochemical and fecal sampling at baseline and post-intervention.

The researchers reported a statistically significant increase in microbial richness (Chao1 index) and phylogenetic diversity (Faith’s PD), key indicators of a more diverse and resilient gut microbiota. Notably, there was a measurable restructuring of microbial composition, with 18 amplicon sequence variants (ASVs) found to increase and 17 to decrease. Among the most prominent changes were increases in the butyrate-producing genera Faecalibacterium and Eubacterium coprostanoligenes group, and reductions in potentially harmful genera like Eubacterium ruminantium group and Gastranaerophilales.

Phylogenetic beta diversity also shifted significantly post-intervention, with unweighted and weighted UniFrac metrics indicating marked microbial community restructuring. While no major difference was detected in evenness (Pielou's index), the overall microbiota composition moved toward a profile associated with cardiometabolic benefits.

What measurable dietary and health changes accompanied microbiome shifts?

Participants’ diets saw significant reductions in carbohydrate (−18.1%), protein (−13.2%), and total energy (−12.7%) intake over the six-week period. These dietary shifts coincided with a 1.5% decrease in average waist circumference—a metric tied closely to metabolic risk—even though body weight and BMI reductions were not statistically significant.

A deeper analysis of food group changes revealed reductions in the intake of alcohol (−39%), sweets (−33%), and fast food (−14%), with a 62% increase in egg consumption. Despite these improvements, the Mediterranean Diet Score remained unchanged, suggesting that the AI app’s impact may have been more about food quality and microbiome-targeted optimization than strict adherence to traditional scoring models.

In terms of physical activity, although the overall IPAQ score remained static, the proportion of participants with low activity levels dropped, with several transitioning to moderate or high activity categories.

Biochemically, there was a noted increase in fasting glucose and liver transaminases (AST and ALT), and a reduction in insulin secretion proxy HOMA-B. While platelet counts rose, mean platelet volume and platelet distribution width both decreased. These findings were not directly attributed to the dietary intervention and may require further longitudinal investigation to assess clinical relevance.

Can microbial shifts be linked to specific health or dietary markers?

The researchers conducted correlation analyses between post-intervention microbial abundance and host health parameters. Changes in olive oil intake were significantly and positively associated with an increase in the genus Oscillospiraceae_UCG_002, which has been linked to reduced insulin resistance. Similarly, changes in triglyceride levels were positively associated with Lachnospiraceae_UCG_004 abundance.

Crucially, Eubacterium coprostanoligenes group and Oscillibacter—two of the genera that increased post-intervention—are known to convert cholesterol into less absorbable coprostanol, potentially aiding in lipid regulation. Their elevation supports prior evidence from large cohort studies suggesting their role in cholesterol homeostasis and cardiovascular protection.

On the inflammatory spectrum, the decline in Gastranaerophilales—a genus associated with elevated lipopolysaccharide (LPS) production, could signal reduced endotoxemia risk. Moreover, the reduction in Eubacterium ruminantium group, an opportunistic pathogen linked to multiple myeloma and respiratory inflammation, may further reflect the app’s protective modulation of the gut microbiota.

While correlation does not establish causation, these relationships suggest that AI-driven personalization can dynamically steer microbial responses aligned with favorable dietary and health metrics.

What are the broader implications and limitations?

The study marks a critical step toward integrating AI and microbiome science into preventive healthcare. By customizing meal plans based on individual characteristics and real-time tracking, the PROTEIN app demonstrated potential to shift users toward healthier gut profiles, subtly adjust eating behaviors, and improve select anthropometric markers.

However, the study's small sample size, short duration, and absence of a control group limit its generalizability. Moreover, functional microbiome activity was inferred rather than directly measured, and the self-reported dietary data may be subject to bias. Still, the statistically robust findings on microbial diversity and specific taxa changes provide strong justification for expanded trials.

Furthermore, the authors stress the need for longer-term studies incorporating multi-omics approaches, including metabolomics, transcriptomics, and genetic data, to fully map the health impacts of AI-guided diets. They also suggested exploring the application of such interventions in clinical populations with chronic diseases, such as diabetes or cardiovascular disorders.