We've spent decades giving everyone the same dietary advice: eat less, move more, five-plus a day. It's not wrong. It's just incomplete. Two people can eat the same meal and have wildly different glycaemic responses, different nutrient absorption, different metabolic outcomes. The science now explains why - and AI is making that science practical.
What You Need to Know
- Individual responses to the same food vary dramatically based on genetics, gut microbiome composition, metabolic status, and lifestyle factors
- AI models can now integrate genetic, microbiome, and biomarker data to generate personalised nutrition recommendations that outperform generic dietary guidelines
- The most clinically validated applications today are glycaemic response prediction and nutrigenomics (how genetic variants affect nutrient metabolism)
- Personalised nutrition is moving from research into clinical practice, but the gap between legitimate science and consumer marketing remains wide
Why One-Size Nutrition Fails
The premise of personalised nutrition rests on a well-established scientific observation: people respond differently to the same food.
A landmark study from the Weizmann Institute tracked continuous glucose monitors on 800 participants eating identical meals. The variation in glycaemic response between individuals was enormous. A food that barely raised one person's blood sugar caused a significant spike in another.
20x
variation in postprandial glycaemic response between individuals eating identical meals, observed in the landmark Weizmann Institute study
Source: Zeevi et al., Cell, 2015
This isn't surprising when you consider the variables. Your genetic variants affect how you metabolise carbohydrates, fats, and specific nutrients. Your gut microbiome composition affects how you break down fibre and produce short-chain fatty acids. Your metabolic status - insulin sensitivity, liver function, hormonal balance - affects how your body processes what you eat. Your sleep, stress, exercise timing, and meal timing all modulate the response further.
Generic dietary guidelines can't account for any of this. They're population averages applied to individuals. Useful as a starting point. Inadequate as a health strategy.
The Three Pillars of Personalised Nutrition
Personalised nutrition draws on three data sources, each contributing a different layer of insight.
Nutrigenomics
Your DNA affects how you metabolise specific nutrients. Variants in the MTHFR gene affect folate metabolism. Variants in CYP1A2 affect caffeine metabolism. Variants in FTO influence appetite regulation and fat storage. Variants in LCT determine lactose tolerance.
These are well-characterised genetic effects with clear nutritional implications. Someone who is a slow caffeine metaboliser based on their CYP1A2 genotype has a measurably different health response to coffee than a fast metaboliser. The dietary recommendation should differ accordingly.
Nutrigenomic testing is the most established pillar of personalised nutrition. The genetic variants are well-studied, the testing is reliable, and the dietary implications are clinically meaningful.
Microbiome Analysis
Your gut microbiome - the trillions of microorganisms in your digestive system - plays a substantial role in how you process food. Different microbiome compositions produce different metabolic outputs from the same dietary inputs.
Microbiome analysis is less mature than nutrigenomics. The science is strong on broad patterns but still evolving on specific recommendations. We know that microbiome diversity correlates with better health outcomes. We know that specific bacterial populations affect glycaemic response. Translating this into precise dietary recommendations is improving rapidly.
Metabolic Monitoring
Continuous glucose monitors (CGMs), blood biomarker panels, and other metabolic tracking tools provide real-time data on how your body is actually responding to what you eat.
This is the feedback loop that makes personalised nutrition practical. A genetic profile tells you about predispositions. A microbiome analysis tells you about current capacity. Metabolic monitoring tells you what's actually happening in response to specific foods.
Personalised nutrition isn't about eliminating food groups because of a genetic variant. It's about understanding your biology well enough to make food choices that work specifically for you.
Jay Harrison
Health Technology Advisory
Where AI Enters
The challenge with personalised nutrition is complexity. A single person's nutritional profile involves thousands of genetic variants, hundreds of microbiome species, dozens of biomarkers, and continuous metabolic data. No human clinician can integrate all of this in real time.
AI can.
Machine learning models trained on population-level datasets can identify patterns between genetic profiles, microbiome compositions, metabolic responses, and dietary inputs. They can predict, with increasing accuracy, how a specific individual will respond to a specific food.
79%
accuracy of machine learning models in predicting individual postprandial glycaemic response from genetic, microbiome, and lifestyle data
Source: Berry et al., Nature Medicine, 2020
The practical application looks like this: you provide your genetic data, microbiome sample, and a period of metabolic monitoring. An AI model integrates these inputs and generates dietary recommendations tailored to your biology. Not "eat more vegetables" but "your glycaemic response to white rice is significantly higher than average - here are specific alternatives that your biology handles better."
The PREDICT Studies
The largest body of evidence for AI-powered personalised nutrition comes from the PREDICT studies, conducted by researchers at King's College London with the company ZOE. These studies tracked thousands of participants with detailed genetic, microbiome, and metabolic data, building predictive models for individual food responses.
The results demonstrated that personalised AI-generated dietary recommendations improved metabolic markers more effectively than generic guidelines. Participants following personalised plans showed better glycaemic control, improved blood lipid profiles, and reduced markers of inflammation compared to those following standard nutritional advice.
What's Practical Today
Not all personalised nutrition applications are equally ready for clinical use. Here's where the science stands.
Ready now: Nutrigenomic testing for well-characterised variants (caffeine metabolism, lactose tolerance, folate metabolism, vitamin D requirements). CGM-guided glycaemic optimisation. Pharmacogenomic-guided supplement interactions.
Emerging: Microbiome-based dietary recommendations. Multi-omic integration (combining genetic, microbiome, and metabolic data). AI-predicted meal responses for metabolic optimisation.
Not yet ready: Fully automated AI nutritional planning without clinical oversight. Microbiome-based precision supplementation. Long-term health outcome prediction from nutritional genomics alone.
The Clinical Guardrails
Personalised nutrition, like all health AI, needs clinical governance. The risk of an AI model recommending a diet that's optimal for metabolic markers but deficient in essential nutrients is real. The risk of consumers over-interpreting genetic variants and unnecessarily restricting food groups is documented.
The model that works: AI generates recommendations, a qualified nutritionist or dietitian reviews them in context of the individual's complete health picture, and the person receives guidance that's both scientifically informed and clinically safe.
Remove the guesswork, but keep the clinician.
Where This Goes
Within five years, I expect personalised nutrition to be standard practice for anyone managing a chronic metabolic condition. Type 2 diabetes management guided by individual glycaemic response data. Cardiovascular risk reduction with genetically informed dietary plans. Weight management based on metabolic profiling rather than calorie counting.
The science is there. The AI tools are maturing. The remaining challenge is clinical integration - getting personalised nutrition recommendations into the workflow of GPs, dietitians, and wellness professionals who see patients daily.
That integration challenge is solvable. We've seen it solved in pharmacogenomics, where genetic data now routinely informs prescribing decisions. The same pathway - science, then tools, then clinical integration - will play out in nutrition.
- Do I need genetic testing to benefit from personalised nutrition?
- Genetic testing provides the deepest layer of personalisation, but continuous glucose monitoring alone can provide meaningful dietary insights by showing how your body responds to specific foods. Starting with CGM data and adding genetic information later is a practical approach.
- Are personalised nutrition apps scientifically valid?
- Some are, many aren't. The key differentiator is whether the app uses validated biological data (genetics, microbiome, metabolic monitoring) or just self-reported questionnaire data. If the personalisation comes from a quiz about your preferences and goals, it's a lifestyle app, not personalised nutrition.
- How does personalised nutrition differ from seeing a dietitian?
- It complements rather than replaces a dietitian. Personalised nutrition provides biological data that a dietitian can use to make more targeted recommendations. The best model is AI-generated insights reviewed and contextualised by a qualified professional.