Glycemic index and weight loss: what the diet trials show

Low-GI diets sometimes show a weight advantage, but results vary. Read the size of the difference, the long-term evidence and the blood sugar findings.

日本語で読む（Japanese version）

You reach for your usual lunch, then notice “low GI” on another package. The label seems reassuring: perhaps you can keep eating carbohydrates and still lose weight. A moment later, the reassurance becomes a question. If you choose your usual food instead, are you making weight management harder? Lunch has acquired a research problem before you have even opened the container.

The trials do not offer a single, confident answer that low-GI or low-GL diets produce substantially greater weight loss. A recent systematic review suggests little to no difference from comparison diets, while an older review and some narrower groups of participants show an advantage. Those findings deserve to sit alongside each other, rather than having the favorable ones printed on the package and the rest left behind.[1][5][7]

The central question concerns what changing carbohydrate quality changes in practice, and how to read the reported weight differences. Comparisons between carbohydrate restriction and fat restriction, and the detailed evidence about fiber and fullness, belong elsewhere in this feature. Here, the aim is to distinguish a useful way of thinking about meals from a promise about your next weigh-in. That distinction matters whether you buy lunch near Tachikawa Station, cook for a family, or fit dinner around a late return home.

GI and GL are not weight-loss scores

GI stands for glycemic index; GL stands for glycemic load. These are the terms used in the research on carbohydrate quality discussed here.[1][5] These trials compare dietary patterns; they do not turn a GI label into a rule about which foods you are allowed to eat.[1][5]

GL takes account of both GI and the amount of available carbohydrate.[16] That gives us a useful distinction to keep in mind: is a claim about carbohydrate quality, or does it also involve carbohydrate quantity? A label saying “low” does not tell you everything about the design of a diet. Recognizing the name of an index is only the start of deciding what a meal will contain.

This distinction is visible in a trial comparing a low-GL diet with a low-fat diet in young adults with obesity. The low-GL diet provided 40% of energy from carbohydrate and 35% from fat. The low-fat diet provided 55% from carbohydrate and 20% from fat.[4] GL was not the only difference between those diets. Describing the experiment as a simple exchange of carbohydrate quality would hide changes in the proportions of nutrients as well.

A longer trial in Brazilian women approached the question differently. Its design kept other dietary components equal, while carbohydrate remained approximately 60% of energy in both groups. Despite the difference in GI, the low-GI group did not have a long-term advantage in weight control.[19] These studies ask related questions, but they do not make identical comparisons. A diet that changes carbohydrate quality while retaining its share of energy is different from one that also changes the carbohydrate and fat shares.

At the supermarket, a short label rarely explains these distinctions. A practical question is what you intend to replace, and how much of it you intend to eat. If you change the staple food but also add another dish because the new staple seems reassuring, you have made more than a change in carbohydrate quality. That is a description of the change in your meal, not a tested recipe for losing weight.

The same applies when several things change at once. You might buy a different staple, cook more often and alter the other dishes you serve. Those choices can be considered together as a change in your eating pattern. They cannot, by themselves, tell you which part of that pattern explains a change in weight. The trials illustrate why naming a diet after a particular index does not mean that the index was the only condition that moved.[3][4]

A low-GI label can be an invitation to consider a meal more closely. It is not a forecast of how many kilograms you will lose. Across the studies, the result depends on the comparison diet, the participants and the time being assessed.[1][4][5][19] Keeping those conditions attached to the claim is more useful than treating a food score as approval for the whole meal.

Average weight differences are smaller than a label suggests

A useful starting point is the 2023 systematic review of randomized controlled trials in people with overweight or obesity. In these trials, participants were assigned at random to different diet groups. The review included 10 studies and 1,210 participants, with interventions lasting from 8 weeks to 18 months. Most participants were adults, although the included studies also covered children and older adults.[5]

For low-GI or low-GL diets compared with higher-GI or higher-GL diets, the pooled difference in weight change was −0.82 kg. Its 95% confidence interval ran from −1.92 to 0.28 kg. This particular weight analysis used 7 studies and 403 participants, and the certainty of the evidence was rated moderate.[5] The estimate leans toward the low-GI or low-GL diets, but the interval includes no difference between groups.

In this review, the interval from −1.92 to 0.28 kg shows the uncertainty around the estimated weight difference. Here, the figure does not mean that everyone who selects a low-GI diet will lose 0.82 kg. It also does not describe the total weight lost by participants in that group. It describes an average difference in weight change between the compared diets.[5] Those are distinct statements, even when an article places the same number next to all of them.

Imagine reading a headline that says a diet “lost more weight.” Before deciding how impressive it sounds, ask compared with whom, by how much, and with what uncertainty. Participants in both groups may change weight, while the difference between their changes remains small. A comparison can therefore be relevant without supplying the dramatic advantage suggested by the diet’s reputation. In this review, the authors’ conclusion was that there is probably little to no difference in weight change.[5]

The review also compared low-GI or low-GL diets with other diets. In that comparison, the mean weight difference was −1.24 kg, with a 95% confidence interval of −2.82 to 0.34 kg. Again, the authors concluded that weight change probably differed little or not at all.[5] Changing the comparator did not establish a clear weight-loss winner simply on the basis of the low-GI name.

The strength of a pooled result also depends on its ingredients. The trials in the review had unclear or high risk of bias across several areas. Many had small comparison groups, and the authors called for larger, better-designed studies.[5] Combining studies does not remove weaknesses in the studies themselves. The review is valuable partly because it makes those limitations visible rather than turning the combined estimate into a guarantee.

A separate 2022 meta-analysis likewise did not show that low-GI or low-GL diets reduced body weight more than higher-GI or higher-GL diets across adults with excess weight overall. Within adults whose body mass index, or BMI, was at least 30, however, there was a favorable mean difference of −0.93 kg, with a 95% confidence interval of −1.73 to −0.12 kg.[1] The overall finding and the finding in this narrower group need to remain separate.

Other body measurements do not automatically follow the weight result. The 2022 analysis found no differences between groups in fat mass, fat-free mass or waist circumference.[1] The 2023 review reported a fat-mass difference of −0.86 kg against higher-GI or higher-GL diets, with a 95% confidence interval of −1.52 to −0.20 kg. Yet the evidence was rated low certainty, and the authors still described the result as potentially little to no difference.[5]

Those fat-mass findings differ, and both belong in the account. It would be misleading to retain only the favorable number, but equally misleading to erase it because the conclusion is cautious. Weight, fat mass and waist circumference are separate outcomes. An advantage in one comparison cannot be carried across to all of them as though they were different names for the same result.[1][5]

Favorable studies still need their conditions attached

The hope that low-GI diets help with weight loss has a research history. A 2007 systematic review included 6 trials and 202 participants, and reported greater weight reduction with low-GI or low-GL diets than with comparison diets. The mean difference for body weight was −1.1 kg, with a 95% confidence interval of −2.0 to −0.2 kg.[7] That favorable result should not disappear from the discussion just because later evidence is more restrained.

Its time frame matters. Interventions lasted from 5 weeks to 6 months, and follow-up after the intervention ended extended to at most 6 months.[7] That is not the same evidence as observing a diet over a much longer period. When a short-term result is used to describe a lasting eating pattern, the unanswered question is how the difference behaves after the early phase.

The 2023 update was also not merely a new interpretation of an unchanged pile of trials. Following revised inclusion criteria, it retained only 1 study from the previous review and included 9 newly identified studies.[5] The older and newer reviews therefore differ in their material as well as their conclusions. It is more informative to explain that change than to say that science first proved the diet worked and then proved it did not.

A 2026 meta-analysis included 21 trials and 1,265 participants and reported a favorable association with body weight. The standardized mean difference was −1.09, while the indicator of variation between studies was 92%. The authors called for extremely cautious interpretation because of the substantial differences between studies and limitations in their methodological quality.[8] This result provides a favorable estimate alongside a strong warning about how to use it.

The unit is especially important. A standardized mean difference is not a number of kilograms. Reporting this result as “1.09 kg lost” would change what the analysis actually measured.[8] A reader should not have to become a statistician to recognize the problem: if the original result has no kg unit, an account of the result cannot supply one simply because kilograms make a more familiar headline.

The authors did not present the analysis as establishing definitive clinical efficacy. They emphasized the limitations of the literature.[8] Its recent publication date does not make the uncertainties disappear. The striking estimate and the warning are parts of the same finding, and the warning is essential to understanding why this study does not settle the question on its own.

Individual trials also show why a result can be more complicated than a diet’s name. One trial assigned 129 young adults with overweight or obesity to diets differing in GI or protein proportion for 12 weeks. Every group lost weight, and the difference between groups in mean percentage weight loss was not statistically significant. The proportion losing at least 5% of starting weight, and fat-mass changes among women, did differ across groups.[3]

Those findings are compatible because they concern different outcomes. The average percentage lost is not the same measure as the proportion of people passing a specified threshold. Likewise, a result among women for fat mass is not the overall group’s result for body weight. The useful lesson is to identify which outcome the claim refers to before deciding whether the study supports it.[3]

In that trial, the high-carbohydrate, low-GI group also ate more fiber, and the planned nutrient proportions were not achieved exactly.[3] A favorable outcome cannot therefore be attributed entirely to GI without acknowledging the other differences. For a household, changing meals in several ways may be a perfectly recognizable experience. For a scientific explanation, it leaves several possible components of the change intertwined.

There is another small but consequential reading habit: distinguish the number of participants in a review from the number contributing to a particular analysis. Not every included trial reports every outcome. The 2023 review’s total of 1,210 participants does not mean that its weight estimate against higher-GI diets used all of them; that analysis involved 403 participants.[5] Keeping the relevant denominator beside the result prevents the scale of the review from overstating the scale of one comparison.

Numbers: glycemic index, weight and blood sugar

−0.82 kg: The mean difference in weight change between low-GI or low-GL and higher-GI or higher-GL diets; the 95% confidence interval, −1.92 to 0.28 kg, includes no difference.[5]

1,210 participants: The total in the 2023 review, rather than the number used in every separate weight analysis.[5]

−0.93 kg: The mean weight difference in adults with BMI at least 30 in the 2022 analysis, distinct from its overall adult finding.[1]

18 months: The assessment period in the Brazilian women’s trial, which found no significant difference in weight change at that point.[19]

−0.31%: The HbA1c difference in the diabetes analysis, a blood sugar finding rather than a percentage of weight lost.[2]