Glycemic load vs glycemic index — the one that matters

Glycaemic index gets the attention. Glycaemic load does the actual work. The distinction is not academic — it changes which foods you eat, how much of them, and whether the number on the label is any use at all.

GI measures how fast a fixed 50 g carbohydrate dose raises blood glucose relative to pure glucose. The problem: almost nobody eats 50 g of carbohydrate from a single food in one sitting. A carrot has a high GI (71) but a typical portion delivers about 5 g of carb, giving it a glycaemic load of 4 — clinically low. White rice carries a GI around 72 and a 200 g cooked serving gives a GL of roughly 22. Two foods with similar GI values produce entirely different blood glucose responses because GL is portion-weighted: it multiplies the GI by the actual carbohydrate grams you eat, then divides by 100. That quotient is what actually matters at the table. The underlying biochemistry of why glucose and fructose follow different metabolic pathways adds an important layer to this picture for anyone managing liver health or uric acid.

For daily meal decisions, the ‘so what’ is direct. A diet targeting low GL doesn’t prohibit high-GI foods categorically — it calibrates portion size instead. You can eat watermelon (GI 76). You eat 120 g of it, not 400 g. The GL stays under 10. That kind of nuance is invisible if you’re managing only by GI. It also explains why two people on nominally ‘identical’ diets can show different post-meal glucose curves: portion variation drives GL variation, and GL variation drives the physiological response.

The definition, properly

GI is defined as the incremental area under the blood glucose response curve produced by a 50 g available-carbohydrate portion of a food, expressed as a percentage of the response to the same amount of carbohydrate from a reference food — either pure glucose or white bread — tested in the same subject.¹ The scale runs from 0 to 100. A food at GI 55 or below is classified low. From 56 to 69 is medium. At 70 and above it is high.

GL takes that number and corrects it for the real-world portion:

GL = (GI × grams of available carbohydrate in the serving) / 100

That’s the whole math. Anything else you’ve read that makes it sound mystical is selling something.

Available carbohydrate means total carbohydrate minus dietary fibre — because fibre largely bypasses rapid intestinal absorption and does not contribute meaningfully to the acute glucose spike. Most food composition databases already report this number directly, so you are not doing additional subtraction in your head.

Work through a concrete case. Jasmine white rice has a GI of 73.² A standard 200 g cooked serving of jasmine rice contains approximately 52 g of available carbohydrate. Plug in: (73 × 52) / 100 = GL 38. That is high by any clinical threshold. Now take brown rice, GI 50. The same 200 g cooked serving contains around 42 g available carbohydrate. GL = (50 × 42) / 100 = 21. Still medium-to-high, but measurably lower — and the difference was produced by both a lower GI and a slightly different carbohydrate yield from the same weight of food.

The formula itself has been stable since Jenkins et al. first introduced the concept in 1997 and Salmeron et al. validated it prospectively against disease risk in the Nurses’ Health Study and Health Professionals Follow-up Study.³ It has not been revised because it doesn’t need to be. What has been revised repeatedly is the underlying GI database — the empirical inputs — as more foods have been formally tested.

Why GI alone misleads

The watermelon case is the clearest demonstration of the problem with GI-only thinking, and it is worth running the full numbers rather than hand-waving at it.

Watermelon has a GI of 76, placing it firmly in the high category.² If you stopped reading there, you would conclude that watermelon is a food to avoid. But 100 g of watermelon contains approximately 7.6 g of total carbohydrate, of which roughly 0.4 g is fibre, leaving 7.2 g of available carbohydrate. GL = (76 × 7.2) / 100 = 5.5. That is low — clinically low, below the 10-unit threshold that most guidelines use. You would need to eat approximately 600 g of watermelon in one sitting to produce a medium GL response. At 600 g, you are consuming a large bowl of fruit that happens to be 92% water. The glycaemic effect is still only borderline medium.

Compare that directly to white bread, GI 75 — almost identical to watermelon’s GI of 76.² Two slices of standard white bread weigh around 60 g and contain approximately 30 g of available carbohydrate. GL = (75 × 30) / 100 = 22.5. High. Same GI tier, entirely different physiological outcome, because the carbohydrate density per gram of food is radically different.

White rice (jasmine) at GI 73, 200 g cooked serving: GL 38, as calculated above. That is nearly seven times the glycaemic effect of 100 g of watermelon from two foods sitting in the same GI band.

The practical import is not that white rice is poison or that watermelon is a health food. It is that GI tells you the speed of glucose delivery per carbohydrate gram, and GL tells you how much glucose you actually delivered. Managing only by GI is like rating a car’s fuel consumption per litre while ignoring how big the fuel tank is. The tank size — the carbohydrate content of the actual portion — is what determines the total fuel load.

This is the central failure mode of popular low-GI dietary advice from the late 1990s and early 2000s: it correctly identified that not all carbohydrates behave identically, then drew the wrong operational conclusion by treating GI as a binary filter rather than an input to a calculation.

Where GI values come from

The canonical source for GI values is the University of Sydney Glycemic Index Database, maintained at sydney.glycemicindex.com and published in its comprehensive tabular form as the International Tables of Glycemic Index and Glycemic Load Values by Atkinson, Foster-Powell, and Brand-Miller in Diabetes Care in 2008.² That paper catalogued GI and GL values for more than 2,400 individual food items. It remains the standard reference.

The test methodology behind each entry is standardised. A group of at least ten healthy human subjects consume a test meal containing exactly 50 g of available carbohydrate from the food being evaluated. Blood glucose is measured at 0, 15, 30, 45, 60, 90, and 120 minutes. The area under the two-hour glucose response curve (AUC) is calculated. The same subjects then consume the reference food — pure glucose or white bread — in a separate session, and their AUC is measured again. The test food’s GI is expressed as the mean ratio of the two AUC values across the group, multiplied by 100.¹

The requirement for ten subjects per food item is a minimum. Larger subject pools reduce variability and tighten confidence intervals. The database reports values from individual published studies rather than aggregating unpublished industry submissions, which gives it an academic audit trail.

Several important caveats live inside this methodology. First, values are derived from healthy subjects without insulin resistance. GI in individuals with type 2 diabetes can differ — typically glucose responses are proportionally larger, though the rank ordering of foods tends to hold. Second, the test isolates a single food eaten in isolation, which is not how people eat. Mixed meals attenuate glycaemic response through macronutrient interaction, a limitation addressed in the next section. Third, inter-individual variability within a ten-person sample produces confidence intervals wide enough that a food’s true GI might span 10 to 15 units around the reported mean.⁴

None of this makes the database useless. It makes it a tool to be applied with appropriate calibration rather than treated as biochemical law.

GL thresholds and what they mean clinically

The American Diabetes Association and the published research literature converge on three GL bands for individual foods per serving:⁵

• Low GL: 10 or below
• Medium GL: 11 to 19
• High GL: 20 and above

For a full meal, the relevant threshold is the sum of GL across all components. A meal with a combined GL under 20 is generally considered low-impact. Between 20 and 30 is moderate. Above 30 begins to produce the sustained postprandial glucose elevation that drives HbA1c upward over months.

In practical terms, what does this look like across a day? A low-GL breakfast might be two eggs with a slice of whole-grain toast (GI 51, approximately 12 g carb per slice, GL 6) plus a small bowl of full-fat yoghurt — combined GL around 10 to 12. A medium-GL lunch is a bowl of lentil soup (lentils GI 28, 180 g serving at 20 g carb, GL 6) with one piece of sourdough bread (GI 54, GL 7) — combined GL 13. A high-GL dinner is a plate of white pasta (GI 50 for al dente, but 200 g dry weight cooked produces 70 g carb, GL 35) with tomato sauce — combined GL above 35 before accounting for any accompaniments.

The implication for meal planning is cumulative. If breakfast is already at GL 12 and lunch at GL 13, the remaining budget for dinner before crossing a daily GL of 80 — a rough threshold associated with improved metabolic outcomes in prospective studies³ — is approximately 55. That is meaningful headroom, but it disappears quickly if dinner is a large plate of rice or pasta without compensating factors.

A worked example: Indian thali versus Western lunch

An Indian thali is one of the more glycaemically complex meals in common use, because it combines multiple carbohydrate sources with legumes, vegetables, and dairy simultaneously. Work through a standard composition.

Two medium wheat chapati — whole wheat chapati GI 52,² each weighing approximately 40 g with 20 g available carbohydrate. Total available carbohydrate: 40 g. GL = (52 × 40) / 100 = 20.8.

Dal (cooked yellow lentils, one katori / 150 ml) — boiled lentils GI 28,² approximately 15 g available carbohydrate per serving. GL = (28 × 15) / 100 = 4.2.

Brown rice (one medium serving, 150 g cooked) — GI 50,² approximately 31 g available carbohydrate. GL = (50 × 31) / 100 = 15.5.

Mixed vegetable sabzi (150 g) — predominantly non-starchy vegetables; combined GI approximately 45, available carbohydrate approximately 8 g. GL = (45 × 8) / 100 = 3.6.

Raita (100 g plain yoghurt with cucumber) — yoghurt GI 36, approximately 5 g carbohydrate. GL = (36 × 5) / 100 = 1.8.

Pickle (20 g) — negligible carbohydrate, GL approximately 0.2.

Total thali GL: approximately 46.1

That is high by any threshold — above 30 for a single meal. However, the composition is instructive. Dal and sabzi together contribute only 7.8 GL units despite being substantial in volume. Chapati accounts for 20.8 units — nearly half the total — and brown rice adds another 15.5. A straightforward reduction: swap two chapati for one, reduce brown rice to 100 g cooked (GL drops to 10.3), and the total meal GL falls to approximately 28. Still medium-to-high, but within a range that most people with managed type 2 diabetes can accommodate at one meal without spiking into sustained hyperglycaemia.

Now compare the Western lunch: one standard sandwich (two slices white bread, GL 22.5, plus 30 g processed cheese, GL negligible) plus one standard bag of salted crisps (25 g, predominantly starch, GI approximately 57, available carbohydrate 13 g, GL 7.4) plus a 330 ml can of regular cola (GI 63, available carbohydrate 35 g, GL 22).

Total Western lunch GL: approximately 51.9

Higher than the full thali, and with far less nutritional density. The thali’s proteins, fibre from dal, and fat from raita collectively slow gastric emptying and blunt the glucose curve relative to what the GL arithmetic predicts in isolation. The cola-and-crisps combination has no such moderating factors. The absolute GL numbers are comparable; the physiological reality is not.

Limitations of GL

GL is the right metric for meal planning, but it is not a complete description of postprandial glucose response. Several factors modulate the actual outcome relative to the calculated number.

Cooking method and starch structure. GI values in the database are testing-condition values, and cooking changes starch gelatinisation. Overcooked pasta has a higher GI than al dente pasta — the Sydney database entry for al dente spaghetti sits at GI 46 to 50, while overcooked pasta rises toward 65.² Long-grain rice has a lower GI than short-grain rice because of its higher amylose content, which resists digestion more effectively than amylopectin. The same food category can span a 20-unit GI range depending on preparation.

Fibre content. Soluble fibre — the kind found in oats, legumes, and certain fruits — forms a viscous gel in the gut that slows glucose absorption and attenuates the peak. This effect is partially captured by the GI testing process when fibre is an intrinsic part of the food. It is not captured when fibre is removed during processing and then added back as a label claim.

Fat and protein co-ingestion. Fat slows gastric emptying, reducing the rate of carbohydrate delivery to the small intestine. Protein stimulates insulin secretion directly, independent of glucose. Both effects lower the glycaemic response to a mixed meal relative to what pure GL arithmetic would predict. This is part of why raita alongside a high-GL thali produces a blunted response compared to the same GL eaten without the yoghurt. The GL calculation does not model these interactions — it is a food-level metric applied additively, not a pharmacokinetic model of digestion.

Individual insulin sensitivity. A calculated GL of 25 will produce different absolute glucose excursions in a lean insulin-sensitive individual, a person with pre-diabetes, and a person with established type 2 diabetes on metformin. GL predicts relative glycaemic impact within an individual over time better than it predicts absolute values across individuals at a single point. This is a feature of the metric’s design — it was validated against longitudinal health outcomes, not against individual meal CGM traces.

Ripeness and processing. A ripe banana has a higher GI than an underripe one. Finely ground flour raises GI relative to coarsely milled grain. Canned beans have a slightly higher GI than home-cooked beans from dry. The database captures these distinctions imperfectly — a single entry for “banana” or “lentils” obscures real variation.

None of this makes GL useless. It makes it the best practical metric available for food selection and portion sizing, while acknowledging that continuous glucose monitoring remains the only way to measure an individual’s actual postprandial response. Our guide to reading the glucose curve after meals shows how to use CGM data alongside GL calculations to calibrate your personal response.

Why CalEye reports GL

CalEye’s meal analysis pipeline produces GL because it is the clinically actionable number. GI alone tells you nothing useful without knowing the portion. Total carbohydrates alone tell you nothing useful without knowing the absorption speed. GL combines both into a single number per logged meal that a person managing blood sugar can actually act on. For a practical starting point, our carb counting 101 guide explains how to count available carbohydrate — the key input to every GL calculation. Our diabetes nutrition tracker shows how CalEye surfaces GL data alongside other metrics for clinical management.

The technical implementation works as follows. When you photograph a meal, CalEye’s vision model identifies each food component and estimates its weight. The carbohydrate content per gram is resolved against a food composition database. The GI value for each identified food is pulled from the Sydney University Glycemic Index Database reference tables.² GL is then computed inline — (GI × estimated carbohydrate grams) / 100 — for each component, summed to a meal total, and displayed alongside the citation so the underlying reference is traceable.

The citation matters. GI values vary across sources because different databases use different testing conditions and different reference foods. When CalEye shows you a GL number, the GI input comes from a specific published entry in a peer-reviewed database, not from a proprietary estimate. The number is auditable.

Conclusion

The practical implication of tracking GL rather than GI is that it shifts the question you ask at every meal from “is this food allowed?” to “how much of this food fits?” That is a harder question to answer without quantitative support — which is exactly why most dietary advice defaults to GI and why most dietary advice fails people who try to use it for actual meal management. GL requires knowing the portion, and knowing the portion requires measuring it. Once you have the measurement, the calculation is trivial. The hard part is building the habit of measuring portions consistently enough that the data means something.

References

1. Jenkins DJ, Wolever TM, Taylor RH, et al. Glycemic index of foods: a physiological basis for carbohydrate exchange. Am J Clin Nutr. 1981;34(3):362–366.

2. Atkinson FS, Foster-Powell K, Brand-Miller JC. International tables of glycemic index and glycemic load values: 2008. Diabetes Care. 2008;31(12):2281–2283. Sydney University Glycemic Index Database: sydney.glycemicindex.com.

3. Salmeron J, Manson JE, Stampfer MJ, Colditz GA, Wing AL, Willett WC. Dietary fiber, glycemic load, and risk of non-insulin-dependent diabetes mellitus in women. JAMA. 1997;277(6):472–477.

4. Venn BJ, Green TJ. Glycemic index and glycemic load: measurement issues and their effect on diet–disease relationships. Eur J Clin Nutr. 2007;61 Suppl 1:S122–S131.

5. American Diabetes Association. Standards of Medical Care in Diabetes — 2024. Diabetes Care. 2024;47(Suppl 1):S1–S321.