Educational

In 2015, researchers at the Weizmann Institute put continuous glucose monitors on 800 people and fed them standardised meals. Same bread. Same glucose solution. Same portion.

The responses were all over the place. Some people spiked hard on bread and barely moved on rice. For others it was reversed. A few had a bigger response to a banana than to a cookie.

I work on a tool built around glycemic load, so this is a finding I'm obliged to take seriously rather than skip past.

What the studies found

That first study, published in Cell, showed that individual responses to identical food varied far more than the published GI tables suggest. The team then built a prediction model using each person's microbiome, blood markers, and habits, and it outperformed carbohydrate counting at predicting who would spike on what.

The PREDICT studies followed, with over a thousand participants including hundreds of identical twins. Same picture. Genetics explained a surprisingly small share of the variation — under 50% for glucose responses, and less than that for fat and insulin responses. Twins with identical DNA still responded differently to the same muffin.

What did track: gut microbiome composition, sleep the night before, timing of the meal, physical activity around it, and the order in which food was eaten.

What that does and doesn't mean for GI tables

The unhelpful conclusion is that GI and GL are therefore useless. That doesn't follow, and the same researchers didn't draw it.

Population averages remain informative about food. Lentils really are slower than white rice for nearly everyone; the ranking holds even when the magnitudes differ. What the research undermines is the idea that a published number is a precise prediction for a specific person on a specific evening.

The honest framing is that a GI value is a property of the food measured across a group, and your response is that value filtered through your body, your gut bacteria, your sleep, and what else was on the plate. The food value is the starting estimate. It isn't the answer.

The variables you can actually see

Several of the factors driving the variation are things you're in a position to notice:

• Meal order. Eating vegetables and protein before the carbohydrate reliably reduces the peak, in study after study.
• Time of day. Most people are more insulin resistant in the evening. The identical meal at 8 p.m. often produces a higher rise than at 1 p.m.
• Sleep. One short night measurably reduces insulin sensitivity the following day.
• Movement. A ten-minute walk after eating flattens the curve for most people.
• What came before. Fibre at lunch affects the response to dinner.

None of those appear anywhere in a GI table, and together they can account for a large share of why a food that was fine on Tuesday isn't on Friday.

Why Glyc still shows the calculation

This is the reason the breakdown is visible rather than tucked away behind a single verdict.

If the tool handed you one number and nothing else, you'd have to treat it as authoritative, and it isn't. Showing which ingredients contributed what turns the number into something you can argue with. When your meter disagrees with the estimate, you can see exactly which assumption to distrust.

That's also why I'd rather flag an ingredient we can't match confidently than quietly substitute an average and present a clean total. A tidy number that hides its uncertainty is worse than a slightly messy one that shows it.

The limits of the personalisation pitch

A commercial industry grew out of this research, and it's worth being clear about where the science stops and the product begins.

The findings themselves are solid: individual variation is large, and it's partly predictable from microbiome and metabolic data. What's less established is that a several-hundred-pound testing package translates into better long-term outcomes than a couple of weeks of self-testing and paying attention. The trials measuring whether personalised nutrition programmes beat good general advice on hard endpoints are still few and mostly short.

Microbiome testing has a further problem: your gut bacteria shift with what you eat, so a snapshot describes a moving target. The measurement is real; its shelf life is shorter than the report implies.

My honest position is that the general principles from this research — meal order, movement after eating, sleep, fibre — are free, apply to nearly everyone, and capture most of the available benefit. The individual profiling is a refinement on top of that, not a replacement for it.

What to do with all this

Use the published values as a starting hypothesis, then test the foods you eat regularly. If you have a CGM, that's a few days of work. With finger sticks, a reading before and one at 60 to 90 minutes after a repeated meal gets you most of the way.

Test the same meal more than once, because a single reading contains too much noise to conclude anything from. Three consistent results is a pattern.

What usually comes out of that exercise is a short personal list: two or three foods that treat you worse than the tables predict, and one or two that treat you better. That list is worth more than any average, and it's yours.

It also tends to be shorter than people expect. Most foods behave roughly as the tables say. The value is in finding the handful that don't, because those are the ones quietly undoing an otherwise sensible week.