Glycemic Index: What It Measures and Its Limits

The glycemic index was developed in 1981 by David Jenkins and colleagues at the University of Toronto. Jenkins was studying blood glucose responses in diabetic patients and needed a way to compare different carbohydrate foods’ effects on blood sugar. The framework he developed has since become one of the most cited and most misapplied concepts in nutrition science.

What It Measures

The glycemic index (GI) measures how quickly a food raises blood glucose, relative to a reference food. The reference is either pure glucose (scored at 100) or white bread (also used as a reference in some older studies).

To measure a food’s GI, test subjects eat a fixed portion of the food containing 50 grams of available carbohydrate. Blood glucose is measured over two hours. The area under the blood glucose curve for the test food is divided by the area under the curve for the reference food, multiplied by 100. The result is the food’s glycemic index.

The scale:

Low GI: 55 or below
Medium GI: 56-69
High GI: 70 or above

Selected values: white bread ~75, glucose ~100, sucrose (table sugar) ~65, watermelon ~76, apple ~36, black beans ~30, boiled white rice ~72, boiled sweet potato ~63, cornflakes ~81, oat porridge ~55.

What Affects the GI of a Food

GI is not a fixed property of a food’s chemistry alone. Several factors influence the measured GI:

Fiber content. Soluble fiber (in oats, legumes, fruits) forms a gel in the intestine that slows glucose absorption. High-fiber foods tend to have lower GI regardless of sugar content.

Physical structure. Intact grain cells are digested more slowly than ground flour. Pasta made from the same wheat as bread has a lower GI than bread, because the dense pasta structure resists enzymatic access. Al dente pasta has a lower GI than overcooked pasta.

Amylose-to-amylopectin ratio. Starch is a mixture of amylose (linear chains) and amylopectin (branched chains). Amylose digests more slowly than amylopectin. Basmati rice has more amylose than jasmine rice and a lower GI.

Fat and protein content. Fat and protein slow gastric emptying, the rate at which food leaves the stomach, and therefore slow glucose absorption. Whole milk has a lower GI than skim milk partly because of its fat content. Adding butter to white bread lowers the blood glucose response.

Ripeness. As fruit ripens, starch converts to simple sugars. Ripe bananas have a higher GI than unripe bananas; the same starch-to-sugar conversion occurs in other fruit.

Cooking and processing. Cooking gelatinizes starch, increasing its digestibility. Cooling cooked starch (resistant starch formation) partially reverses this. Cold, cooked potatoes have a meaningfully lower GI than hot potatoes.

Acidity. Vinegar and other acids slow gastric emptying. Sourdough bread, which is acidic, has a lower GI than regular bread made from identical flour.

Glycemic Load: What GI Misses

The GI is calculated based on a fixed 50g of available carbohydrate. But actual portion sizes vary enormously, and the glucose response depends on both how fast carbohydrate is digested (GI) and how much carbohydrate is eaten (portion size). Glycemic load (GL) combines both:

GL = (GI × grams of carbohydrate) ÷ 100

Watermelon has a GI of 76, high. But a typical serving of watermelon (150g) contains only 11g of available carbohydrate. GL = (76 × 11) ÷ 100 = 8.4, low.

Carrots had a GI measured at 92 in early studies (the “carrot problem” was widely discussed). But a serving of carrots contains about 6g of carbohydrate. GL = (92 × 6) ÷ 100 = 5.5 — trivially low. The fear that carrots spike blood sugar from GI alone was not reflected in actual glycemic load.

This is why GL is generally more useful than GI for evaluating real-world food choices.

Practical Limitations

Testing variability. GI is measured on fasting subjects eating a single food in isolation. Real meals contain combinations of foods with fats, proteins, and fibers that collectively determine the actual blood glucose response, typically much lower than the GI of any individual component eaten alone.

Individual variation. Blood glucose responses to the same food vary substantially between people. A 2015 study by Zeevi et al. in Cell measured glucose responses to identical meals in 800 people and found enormous individual variation, driven by microbiome composition, sleep, physical activity, and genetics. GI measured on a population average poorly predicts individual response.

The GI-calorie tradeoff. Low-GI foods are not automatically healthy. Fructose has a low GI (around 19) because it doesn’t directly raise blood glucose. But its liver metabolism has distinct negative effects not captured by GI. Ice cream has a lower GI than boiled potatoes because of its fat content.

Short-term metric for a long-term problem. GI measures a two-hour blood glucose curve. Chronic health outcomes develop over years and depend on overall dietary pattern, caloric balance, and many factors unrelated to blood glucose speed.

Research Evidence on Low-GI Diets

Despite the tool’s limitations, diets organized around lower-GI foods (not low-GI products specifically, but whole foods that happen to have low GI) do show consistent benefits in research.

A 2012 Cochrane review found that low-GI diets were associated with modest reductions in body weight, total cholesterol, LDL cholesterol, and fasting blood glucose compared to control diets.

A 2014 meta-analysis in British Journal of Nutrition of 25 randomized trials found that low-GI/GL diets reduced HbA1c and fasting blood glucose in type 2 diabetic patients.

The research generally suggests that the principle underlying low-GI eating — favoring whole foods with fiber and intact structure over refined and processed carbohydrates, is valid, even if the specific GI score is a blunt tool for implementing that principle.

Glycemic Index: Key Numbers

GI scale: low ≤55, medium 56-69, high ≥70
White bread: GI ~75
Table sugar (sucrose): GI ~65
Watermelon: GI ~76, but GL per typical serving ~8.4 (low)
Fructose: GI ~19 (but low GI doesn’t mean metabolically harmless)
Individual glucose response variation: same meal can produce 2-10x different blood glucose responses in different people (Zeevi et al., 2015)