Personalised Health Optimisation: What's Real and What's Hype

This article assesses the evidence behind personalised health tools; it does not endorse, rank, or recommend any specific product or service.

Personalised health optimisation is the idea that combining artificial intelligence, wearables, continuous glucose monitors, genome sequencing and microbiome tests can tailor diet, exercise and lifestyle advice to one individual rather than to a population average. The evidence is genuinely mixed. Some forms of personalisation rest on solid, replicated research; others are sold well ahead of the data that would justify them.

The short version: there is good evidence that people respond differently to the same meal, and that some of this difference is measurable and reasonably stable within a person [Berry 2020; Zeevi 2015]. There is much weaker evidence that consumer genomics, microbiome diet kits, or continuous glucose monitoring in metabolically healthy people change health outcomes in ways that matter. The honest position in 2026 is that personalisation is a promising research direction with a handful of validated applications and a long tail of premature commercial claims. This article sets out where the line currently sits.

What "personalised health" actually means

"Personalisation" is doing a lot of work in marketing copy, so it helps to separate the layers.

• Data collection. Wearables (heart rate, sleep stages, step counts), continuous glucose monitors (CGMs), at-home blood panels, genome or polygenic scores, and stool-based microbiome sequencing.
• Modelling. Algorithms, increasingly described as AI, that turn that data into a prediction or a score.
• Recommendation. Diet, training, supplement or lifestyle advice said to be tailored to you specifically.

A product can be impressive at one layer and weak at another. A sensor can be accurate while the advice built on top of it is unsupported. When you read a claim, it is worth asking which layer the evidence actually addresses. For context on the wider market, see our overview of the consumer health AI landscape and the broader work on AI in health.

Where the evidence genuinely supports personalisation

The strongest case for personalisation comes from postprandial (after-meal) metabolism. Two large datasets have shaped the field.

In the PREDICT 1 study, researchers measured responses to standardised meals in roughly 1,000 UK and US adults, including identical and non-identical twins. They found large between-person variability in blood glucose (around 68% coefficient of variation), triglyceride (around 103%) and insulin (around 59%) responses to the same food [Berry 2020]. Crucially, genetics explained only part of this. Additive genetic factors accounted for about 30% of the variance in the glucose response and far less for triglyceride and insulin, which implies that much of the variation is driven by modifiable and non-genetic factors such as the microbiome, sleep, and meal timing [Berry 2020].

Earlier, an Israeli cohort of 800 people logged nearly 47,000 real-world glycaemic responses and showed that an algorithm combining clinical data, blood markers and gut microbiome features could predict an individual's post-meal glucose better than counting carbohydrates alone [Zeevi 2015]. A follow-up arm reported that personalised diets based on these predictions lowered post-meal glucose in a small validation group [Zeevi 2015].

Two things make this body of work credible. First, the individual response appears to be reasonably reproducible: when the same meal is given to the same person twice, the responses correlate moderately well (around R = 0.77 in the modelling work) [Berry 2020]. Second, the glucose response seems to connect to something people care about: in a PREDICT analysis, larger post-meal glucose "dips" were associated with greater hunger and higher subsequent energy intake [Wyatt 2021].

This is real personalisation: a measurable, partly stable, individually varying trait linked to a plausible outcome. It is also worth being precise about what it is not. It is not yet proof that eating to your glucose curve produces long-term improvements in weight, cardiovascular risk or mortality. Those trials are largely still to come.

Where the picture is genuinely contested

Even within glucose research, reproducibility is debated. A controlled study that gave adults without diabetes duplicate meals across four dietary patterns found that individual glycaemic responses measured by CGM were often unreliable on repeat testing, leading the authors to question how "precise" precision-nutrition claims can be from a single set of readings [Howard 2023]. This does not overturn the PREDICT and Zeevi findings, but it shows the science is still arguing about how stable a person's response really is, and how many measurements you would need to characterise it. That kind of disagreement is normal at this stage and is one reason nutrition research is hard to translate into individual advice.

Where claims are overstated or premature

Continuous glucose monitoring in healthy people

CGMs are well validated for managing diabetes. Their use in metabolically healthy people is a different question. A 2026 systematic review and meta-analysis of CGM in non-diabetic populations found that monitoring modestly improved mean glucose overall, but that the benefit was concentrated in people with prediabetes; in healthy, normoglycaemic individuals there was no clear glycaemic benefit, and effects on actual cardiovascular outcomes were unclear because the data are scarce [Tartof 2026]. A separate systematic review reached a similar conclusion: CGM can increase engagement and prompt behaviour change, but direct evidence that it improves established cardiovascular risk factors in people without diabetes remains limited [Lazar 2025].

In plain terms: a CGM will show a healthy person interesting-looking spikes, but those spikes are largely a normal physiological response to eating, and there is not yet good evidence that chasing flatter curves improves long-term health in this group.

Consumer genomics and polygenic risk scores

Polygenic risk scores (PRS) aggregate the small effects of many gene variants into a single number. They are advancing, and for some conditions such as coronary artery disease they add information when combined with conventional risk factors [Klimentidis 2024]. But two careful appraisals caution against over-reading a consumer report. The predictive performance of a PRS used on its own is consistently modest; combining it with a clinical score typically produces only a moderate improvement over either alone [Lewis 2023]. Scores also perform unevenly across ancestries, because the underlying studies were dominated by people of European descent, and they capture only the heritable portion of risk, ignoring environment and behaviour [Lewis 2023; Klimentidis 2024]. A genome result can be a useful input for a clinician; it is rarely a verdict.

Microbiome diet kits

Stool-based microbiome testing sold with bespoke food lists is among the least settled areas. An evaluation of direct-to-consumer gut microbiome services found that the variation between companies analysing comparable samples was as large as the biological variation between different people, driven by differences in DNA extraction, sequencing and bioinformatics pipelines [Caspi 2025]. If the same sample yields materially different profiles depending on the lab, dietary advice built on that profile is hard to trust. Much of the underlying science is also correlational rather than causal, which our piece on gut microbiome research explores in more detail.

Wearable sensors

Wearables are the most mature layer, but accuracy is conditional, not absolute. Wrist-worn optical heart rate sensors are generally good at rest and during steady activity, yet their accuracy degrades during vigorous or intermittent movement and can be affected by skin tone and sensor contact [Bent 2020]. Sleep-stage estimates and energy-expenditure figures carry larger uncertainties still. The signal is often directionally useful for tracking your own trends over time; treating a single number as clinically precise is where people overreach. Our guide to what wearables can and can't measure covers this in depth.

A measured comparison

A framework for what's worth doing now

You do not need to wait for every trial to act sensibly. A reasonable way to weigh any personalised tool is to ask three questions.

• Is the measurement valid? Does the sensor or test reliably measure what it claims, and is that consistent across labs or repeat readings?
• Is the link to outcomes established? Is there evidence that acting on the result changes something that matters, not just a surrogate number?
• Is the cost, including anxiety, justified? Some tools generate worry or restrictive behaviour out of proportion to their proven benefit.

On that basis, several things are worth doing now because they are low-risk and well-supported, even if not glamorous: using a wearable to track your own activity and sleep trends rather than absolute values; paying attention to the broad finding that meal composition and timing affect how you feel after eating; and discussing any genetic or metabolic result with a clinician before changing your regimen.

Several things are better treated as experiments or as "wait for evidence": routine CGM if you are metabolically healthy, microbiome-guided diet kits, and acting on a polygenic score without professional interpretation. None of these is harmful in itself, but the case that they improve outcomes is not yet made.

The discipline that holds all of this together is reading the underlying claim carefully. Before you accept that a product is "clinically proven", it is worth knowing how to read a clinical trial: what was measured, in whom, against what comparison, and whether the result was a real-world outcome or a short-term surrogate. Personalised health is a field where that scrutiny matters most, because the marketing is far ahead of the data, and the gap is precisely where the hype lives.

Related Proco pages

• The consumer health AI landscape
• What wearables can and can't measure
• Why nutrition research is hard
• How to read a clinical trial

Sources

Before acting on any personalised test or device result, discuss it with a qualified healthcare professional who knows your history.