Two plates with the same carb count can leave very different glucose curves, and so can two people eating the same plate. Here is what the research shows about why, and how strong that evidence is.
A bowl of plain pasta and two slices of pizza can carry about the same amount of carbohydrate. Logged side by side, they rarely leave the same glucose curve. The pasta often brings a quicker rise that settles within a few hours. The pizza may look gentler at first, then push glucose up late in the evening. Compare notes with a friend who also lives with diabetes, and their curve after the same meal may look nothing like yours.
Carbohydrate sets the size of the glucose load. How fast that load arrives, and how hard your body or your insulin has to work to clear it, depends on the fat, protein and fibre on the plate, the order you eat things in and your own biology. The carb count you start from also carries some error. Below is what the research shows on each, who was studied and how strong the evidence is.
Researchers at the Weizmann Institute of Science in Israel gave continuous glucose monitors (CGMs) for a week to 800 adults aged 18 to 70 who had not been diagnosed with type 2 diabetes, and analyzed responses to 46,898 meals 1. Responses to identical meals varied widely from person to person, and the standardized meal that caused the biggest rise differed from one person to the next 1. When the same person ate a standardized meal twice, the two responses agreed fairly well (correlations of about 0.7 to 0.8), which suggests much of the variation reflects stable differences between people 1. A model built from blood tests, eating habits, body measurements, activity and gut bacteria predicted individual responses, held up in 100 new people and, in a small blinded randomized trial of 26 people, chose diets that lowered post-meal glucose 1.
The PREDICT 1 study, led by researchers at King’s College London with co-authors at Massachusetts General Hospital, Harvard and Stanford, gave standardized meals to 1,002 healthy adults in the United Kingdom, many of them twins 2. Glucose responses to identical meals had a coefficient of variation (a measure of spread) of 68% 2. The meals’ carbohydrate, fat and protein content explained about 15% of the variation in glucose responses, and the gut microbiome about 6% 2. Comparing identical and non-identical twins, the team estimated that genetic factors accounted for 48% of the variation in the two-hour glucose response, yet the specific genetic variants they measured explained only about 9% 2. Variation was large even between identical twins. The study was funded in part by Zoe Global, and several authors were employees of or consultants to the company 2.
At Stanford, researchers gave CGMs to 57 adults for two to four weeks. Of these, 38 had normal results on standard glucose tests, 14 had prediabetes and 5 had type 2 diabetes 3. Thirty of them ate three standardized breakfasts. Cornflakes with milk pushed glucose above 7.8 mmol/L (140 mg/dL) in 80% of this group, which means at least 14 of the 20 people with normal standard tests crossed that line 3.
These studies mostly enrolled people without diabetes 1–3. If you take mealtime insulin, the dose and its timing shape your curve as well. Published food values describe averages, and your own data show how your body responds.
The glycemic index (GI) comes from a 1981 study led by David Jenkins at the University of Toronto 4. Healthy volunteers ate 62 common foods and sugars after an overnight fast, 5 to 10 people per food. The team compared each food’s two-hour glucose response with the response to the same amount of carbohydrate taken as glucose 4. On average, the vegetables tested scored 70%, breakfast cereals 65%, fruit 50%, dairy products 35% and dried legumes 31% 4. Foods with more fat and protein tended to produce smaller rises 4.
Today GI is measured with a standardized method, usually on a portion containing 50 g of available carbohydrate, and scored against glucose on a 100-point scale, with 55 or less classed as low 5. The 2021 international tables, compiled at the University of Sydney, list more than 4,000 items 5. Dairy foods, legumes, pasta and fruit usually test low and give consistent values worldwide. Breads, breakfast cereals, rice and snack foods come in high-, medium- and low-GI versions, and most potatoes test high 5.
GI describes the type of carbohydrate and ignores how much of it you eat. Glycemic load (GL) adds portion size by multiplying a food’s GI by the grams of carbohydrate in a serving and dividing by 100 5. A food with a GI of 70 has a GL of 21 in a portion with 30 g of carbohydrate, and 14 in a portion with 20 g.
In plain terms: GI tells you how quickly a food’s carbohydrate tends to raise glucose when it is eaten on its own. GL adds how much of it you eat. Neither accounts for the rest of your plate or for your own physiology.
GI has limits for everyday decisions. It is measured on single foods eaten alone after a fast, and real meals rarely look like that 4, 5. The value itself also moves. At Tufts University, researchers measured the GI of white bread three times in each of 63 healthy adults 6. The average was 62, with a standard deviation of 15. The coefficient of variation was about 20% between repeat tests in the same person and 25% between people, and adding more tests did not narrow the spread 6. The authors concluded that GI is unlikely to be a good way to guide food choices 6.
Diabetes Canada’s nutrition guideline still recommends choosing lower-GI carbohydrate foods to help manage glucose 7. The long-term evidence for GI is weaker than for fibre, though. A set of systematic reviews partly funded by the World Health Organization found smaller or no reductions in disease risk with low-GI or low-GL diets 8.
Those reviews, from the University of Otago in New Zealand and published in The Lancet in 2019, pooled 185 prospective studies and 58 clinical trials, covering about 135 million person-years 8. People who ate the most fibre had 15-30% lower rates of death from any cause and from cardiovascular disease, and of coronary heart disease, stroke, type 2 diabetes and colorectal cancer, than those who ate the least 8. The biggest reductions came at about 25-29 g of fibre a day 8. These comparisons were mostly observational, so they show associations, and they cover the general population, not people already living with chronic disease 8.
The same group then looked at people with diabetes. Their 2020 meta-analysis included 42 trials with 1,789 adults with prediabetes, type 1 or type 2 diabetes, plus two cohorts of 8,300 adults with diabetes followed for an average of 8.8 years 9. In the trials, higher-fibre diets lowered A1c by 2 mmol/mol (about 0.2 percentage points) and fasting glucose by 0.56 mmol/L (about 10 mg/dL), with small improvements in LDL cholesterol and weight 9. In the cohorts, eating 35 g of fibre a day instead of the average 19 g was linked with 14 fewer deaths per 1,000 people over the study period 9. The benefits held across fibre types and types of diabetes, although the trials differed a lot from one another and trials lasting a year or more were lacking 9.
Diabetes Canada suggests that adults with diabetes aim for 30-50 g of fibre a day, with a third or more (10-20 g) from viscous soluble fibre 7. The Otago group proposed increasing intake by 15 g a day, or to 35 g, as a reasonable target 9.
Carb counting treats every gram of carbohydrate alike, whatever else is on the plate. Studies in type 1 diabetes show where that breaks down.
At the Joslin Diabetes Center in Boston, Howard Wolpert and colleagues studied seven adults with type 1 diabetes (average age 55, average A1c 7.2%) on two evenings 10. The dinners had identical carbohydrate and protein, but one contained 60 g of fat and the other 10 g. A closed-loop system adjusted insulin automatically for 18 hours after each meal 10. The high-fat dinner needed more insulin (12.6 vs 9.0 units on average, roughly 40% more) and still produced more hyperglycemia, with about twice the glucose exposure above 6.7 mmol/L (120 mg/dL) 10. That works out to a unit for every 9 g of carbohydrate instead of every 13 g, and the extra need varied widely between people 10. The authors pointed to free fatty acids, which impair insulin sensitivity, as a likely contributor 10.
In Australia, Carmel Smart and colleagues gave 33 children and teens aged 8 to 17, all using intensive insulin therapy, four breakfasts with the same carbohydrate 11. Only fat (4 g or 35 g) and protein (5 g or 40 g) changed, and each person received the same individually calculated insulin dose for every breakfast. Over five hours of CGM, the high-protein breakfast raised glucose more from 3 hours onward, and the high-fat breakfast from 3.5 hours onward 11. With both together, glucose was highest at 5 hours, so the effects added up 11. Protein also lowered the risk of hypoglycemia after the meal 11.
A 2015 systematic review by researchers in Australia and at Joslin pulled these findings together 12. Every included study, seven each on fat, protein and GI, found that these factors changed post-meal glucose. Fat’s main effect was a late rise. In some studies glucose was lower than expected for the first 2-3 hours, possibly because fat slows stomach emptying 12. Meals high in fat and protein needed more insulin than lower-fat, lower-protein meals with the same carbohydrate 12.
Mealtime doses are usually calculated from carbohydrate alone. After a rich meal, that can leave glucose lower than usual for the first couple of hours and higher three to five hours later, which after dinner can mean overnight 11, 12. In Wolpert’s study, the high-fat dinner caused more hyperglycemia even though a closed-loop system was adding insulin as glucose rose 10.
Diabetes Canada’s guideline notes that meals high in fat and protein may need extra insulin and that, for pump users, that insulin may be best delivered over several hours 7. Studies have tested different delivery patterns, such as split or extended boluses, and the 2015 review found they disagreed on which works best, so there is no single formula 12. Any change like this is a decision to make with your care team and to test carefully against your own data.
Researchers at Weill Cornell Medical College in New York tested meal order in 16 adults with type 2 diabetes 13. On three days, in random order, participants ate the same meal in three ways: carbohydrate first with protein and vegetables 10 minutes later, the reverse, or everything together 13. With carbohydrate last, the glucose rise over three hours was 53% smaller and the peak rise 54% smaller than with carbohydrate first, about 1.9 mmol/L (35 mg/dL) above baseline compared with 4.2 mmol/L (75 mg/dL) 13. Compared with eating everything together, carbohydrate last cut the three-hour rise by 44% and the peak by 40% 13. Insulin levels were lower and levels of the gut hormone GLP-1 were higher when carbohydrate came last instead of first 13.
This was a small study of single meals in type 2 diabetes. It shows that order can change the shape of the curve, but it can’t show long-term effects on A1c, and it tells us little about people who dose insulin for meals. If you do, eating carbohydrate later also changes how well your insulin timing matches the meal, so discuss it with your care team first.
Average peak rise above baseline in 16 adults with type 2 diabetes who ate the same meal three ways (Weill Cornell crossover study).
Stanford researchers led by Christopher Gardner compared two lower-carbohydrate diets in 40 adults with prediabetes or type 2 diabetes, in a randomized crossover trial in which each person followed each diet for 12 weeks 14. Both diets emphasized non-starchy vegetables and cut out added sugars and refined grains. The well-formulated ketogenic diet also excluded legumes, fruit and whole grains, which the Mediterranean-Plus diet kept 14.
Among the 33 people with complete data, A1c fell 9% from baseline on the ketogenic diet and 7% on the Mediterranean-Plus diet, a difference that was not statistically significant 14. Triglycerides fell more on the ketogenic diet (16% vs 5%), but LDL cholesterol rose 10% on it while falling 5% on the Mediterranean-Plus diet 14. Weight fell 8% and 7%. The ketogenic phase was lower in fibre and three other nutrients, and 12 weeks after the study, participants’ eating was closer to the Mediterranean-Plus diet, which the authors read as a sign it was easier to sustain 14.
Diabetes Canada’s guideline allows wide ranges for carbohydrate (45-60% of energy), protein (15-20%) and fat (20-35%) to fit individual needs 7. It lists several eating patterns with evidence of benefit, including Mediterranean-style, vegetarian and DASH diets for type 2 diabetes, and encourages people to choose the pattern that best fits their values, preferences and treatment goals 7.
Long experience with type 1 diabetes doesn’t make carb counts exact. Researchers at the Montreal Clinical Research Institute (IRCM) and Université de Montréal compared the estimates of 50 adults who had lived with type 1 diabetes for 21 years on average with a dietitian’s analysis of three days of food records, while participants wore CGMs 15. Across 448 meals averaging 72 g of carbohydrate, estimates were off by 15.4 g per meal on average, about 21% of the meal 15. Most misses (63% of meals) were underestimates. People with larger errors had more glucose variability and spent less time between 4.0 and 10.0 mmol/L (72-180 mg/dL) 15.
How much error matters was tested directly in 34 children and teens using intensive insulin therapy in Australia 16. Each took the insulin dose they would normally use for 60 g of carbohydrate, then ate breakfasts containing 40, 50, 60, 70 or 80 g. Eating 20 g less than the dose covered led to significantly more hypoglycemia, with about a one-in-three chance of a low between 2 and 3 hours 16. Eating 20 g more led to higher glucose over the same window and a greater chance of readings of 12.0 mmol/L (216 mg/dL) or above 16. The authors concluded that estimates should be within about 10 g of the true amount 16. These were single breakfasts in young people, so the margin may differ for adults or for bigger meals.
At Bern University Hospital in Switzerland, 53 people with type 1 diabetes estimated the carbohydrate in 26 hospital-cooked meals served in three sizes, then used two commercial food-recognition apps 17. The apps were about as accurate as people’s own estimates, with no significant differences, and the authors suggested they may help most where a person’s errors are large or inconsistent. Their effect on glucose was not tested 17.
At university hospitals in Caen and Strasbourg, France, trained medical students used four carb-counting apps and the general-purpose AI chatbot ChatGPT-5 to estimate 246 hospital meals, checked against dietitians’ values from weighed food 18. Average errors ranged from 13 to 21 g per meal, and 50% to 80% of estimates fell within 20 g, depending on the tool. The chatbot’s average error was 18 g 18. Results at home, with less controlled meals, may differ.
A photo can’t show the oil in the pan, the sugar in a sauce or how deep a bowl is, and a carb estimate says nothing about the fat and protein that shape the late part of the curve. Treat any estimate, from an app or your own eye, as a draft to check against labels, recipes or a kitchen scale.
Carbohydrate sets the size of the glucose load. How fast that load arrives, and how hard your body or your insulin has to work to clear it, depends on the fat, protein and fibre on the plate, the order you eat things in and your own biology.
Yes. In a Stanford study, cornflakes with milk pushed glucose above 7.8 mmol/L (140 mg/dL) in 80% of those who ate it, including at least 14 of the 20 with normal results on standard glucose tests 3.
In studies that mostly enrolled people without diabetes, glucose responses to identical meals varied widely from person to person 1, 2. When the same person ate a standardized meal twice, the responses agreed fairly well, which suggests much of the variation reflects stable differences between people 1.
Studies in type 1 diabetes show that fat and protein can blunt the early rise and push glucose up 3 to 5 hours later 11, 12. In some studies glucose was lower than expected for the first 2-3 hours, possibly because fat slows stomach emptying 12.
In a small study of single meals in 16 adults with type 2 diabetes, eating carbohydrate last made the peak glucose rise 54% smaller than eating it first 13. It can't show long-term effects on A1c and tells us little about people who dose insulin for meals.
Pick two or three meals you eat often that have similar carbohydrate but different amounts of fat, protein or fibre, and log them carefully for a few weeks. In Glyve, each logged meal sits on the day’s glucose line next to your shaded target range, so you can look past the two-hour mark into the three-to-five-hour window where fat and protein show up. The week view makes it easier to compare similar meals. Meal-photo, barcode and nutrition-label scans give a starting carb estimate that you review and edit before saving, which is a good moment to check the portion. Over time you may see which meals leave a late rise and how far your estimates drift. Glyve shows what happened together; it doesn’t tell you what caused it. Bring the patterns, or a PDF report, to your care team.
How we write about diabetes. Spotted a mistake? Email info@glyve.ca and we’ll check it against the source.
Glyve keeps glucose, food, insulin, exercise, sleep and stress on one timeline.
For iPhone, iPad and Apple Watch.