The short version
- Average glucose 110 mg/dL across the wear period, giving an estimated A1c of 5.9. By every summary measure, a good result.
- Grouped by meal, my breakfasts raised my blood sugar three times as much as my dinners — and snacks were worse than either.
- Between 3am and 8am, asleep and eating nothing, my glucose rose about 8 points on its own. 9am was my highest hour, every day.
- The food I would have called the healthiest in the cupboard produced my worst reading of the whole period. Half a pizza produced a small fall.
- Eating protein and vegetables before carbohydrate lowered my average rise from 23.8 to 19.9 mg/dL across 40 meals.
I was diagnosed with type 2 diabetes in January 2025 and I am still on metformin. Before that I tested the way most people test: a finger-prick, occasionally, usually fasting, usually when I remembered. That is roughly one reading out of the several thousand your body actually produces in two weeks.
So I wore a sensor and found out what the other several thousand were doing.
What I did
A continuous glucose monitor sits on the back of the arm and samples glucose in interstitial fluid every few minutes. Over sixteen days of wear that produced 7,665 readings. Alongside it I kept a written log: what I ate in enough detail to recognize it a month later, when I started eating, and what I had changed compared with the previous day.
The honest accounting: sixteen days of sensor wear, fourteen days of complete meal logs, and 41 meals with a recorded peak. The first five meals were deliberate baseline days where I changed nothing, so that the test days had something to be compared against.
Three rules made the data worth keeping:
- One change at a time. Change five things in a week and you learn nothing about any of them.
- Wash your hands before testing. Fruit residue on a fingertip can produce a dramatically false high — high enough to make you abandon a food that was never the problem.
- Read the peak, not the recovery. I recorded the rise at roughly one hour. Two hours tells you whether you came back down; one hour is nearer the peak, and the peak is the information.
This is a record of one person. The limitations section below is not a formality.
Finding 1: the summary number missed almost everything
My average glucose across the wear period was 110 mg/dL. Converted using the Glucose Management Indicator — the standard formula for estimating A1c from CGM data — that gives an estimated A1c of 5.9. [1] Under the diabetic threshold, and the sort of result that ends a conversation with a clinician happily.
It was also close to useless as a guide to what to eat.
Split the same data by day type and the reason appears. Across my five baseline meals, where I changed nothing, the average rise after eating was 17.2 mg/dL. Across the 36 test meals, where I was deliberately provoking responses, it was 21.7 mg/dL. The average absorbed both without complaint.
I then went back through the file and removed every spike I had caused on purpose — close to a tenth of all readings, including the morning I hit 220 — and recalculated. The estimated A1c moved by about five hundredths of a percentage point.
That is not a flaw in the measure. An average is built to be insensitive to individual events, and HbA1c integrates roughly three months of glucose exposure, so no single meal moves it. [2] It is simply worth knowing that everything useful I learned lived in the part of the data the average is designed to ignore.
Finding 2: mornings were the problem, and breakfast was not the cause
| Meal | Meals logged | Average pre-meal | Average peak | Average rise |
|---|---|---|---|---|
| Breakfast | 13 | 105.9 | 139.6 | +32.8 |
| Lunch | 13 | 113.9 | 126.5 | +11.1 |
| Dinner | 10 | 103.7 | 114.7 | +11.0 |
| Snack | 5 | 109.6 | 147.2 | +37.6 |
All figures in mg/dL. Rise = peak minus pre-meal.
Three times as much in the morning as in the evening. The obvious explanation is that I ate worse at breakfast, and that is not what the data showed — because of what was happening before I ate anything at all.
Between 3am and 8am, asleep and fasted, my glucose rose about 8 points on its own. Nine in the morning was my highest hour on every day of the experiment.
This is the dawn phenomenon, and it is well described: in the hours before waking, cortisol, growth hormone and other counter-regulatory hormones prompt the liver to release glucose whether or not you eat. [3] In people without diabetes the rise is buffered by insulin. In type 2 it often is not.
So breakfast does not land on a flat line. It lands on one already climbing. The practical consequence is the most useful thing I took from the whole exercise: before you give up a breakfast food, try it at dinner. If it only spikes you in the morning, the problem was never the food.
The snack column deserves a note, because it surprised me more than the breakfast one. Five snacks, and the highest average rise of any category — driven by an apple eaten alone in the afternoon and a beer. Food eaten on its own, away from a meal, behaved very differently from the same sort of food inside one.
Finding 3: I was confidently wrong about specific foods
These are my readings, in my body, on metformin. They are not a ranking of foods and they will not transfer to you. That is the finding, not a caveat on it.
| What I ate | When | Pre-meal | Peak | Change |
|---|---|---|---|---|
| All Bran flakes | Breakfast | 110 | 220 | +110 |
| Draft beer | Afternoon | 103 | 166 | +63 |
| Green apple, on its own | Afternoon snack | 118 | 180 | +62 |
| Toasted chicken and cheese sandwich with mayo | Lunch | 117 | 172 | +55 |
| Wholewheat pasta with beef and veg | Dinner | 101 | 132 | +31 |
| Brown toast and avo | Lunch | 131 | 159 | +28 |
| Roast baby potatoes, eaten at the end of the meal | Lunch | 106 | 106 | 0 |
| Half a pizza, eaten after a salad | Lunch | 130 | 126 | −4 |
All figures in mg/dL, peak measured at roughly one hour.
The All Bran was eaten on the assumption that it was the safest thing in the cupboard. It produced the worst reading of the entire period, despite the fibre.
I am not claiming pizza is health food, and I would not eat it as a strategy. I am saying I held confident opinions about both of those meals and was wrong about both, and I only found out because I measured instead of guessing.
Finding 4: the one thing that worked was the order, not the food
The two flattest results in the table have something in common that took me a week to notice. The potatoes were eaten at the end of the meal. The pizza was eaten after a salad. Neither moved my blood sugar at all.
Across every meal where I recorded it, the pattern holds:
| Habit | Meals | Average rise |
|---|---|---|
| Protein and vegetables first | 24 | +19.9 |
| Carbohydrate not held back | 16 | +23.8 |
That is a modest difference in a small sample and I would not lean on it alone. It is worth reporting because it points the same way as the controlled evidence: eating vegetables and protein before carbohydrate substantially lowers the glucose and insulin response to an identical meal. [4] In my data the effect shows up three separate ways — in the averages, in the potatoes, and in the pizza.
It is also the only change I made that required giving nothing up.
The result that went the wrong way
I walked after three meals to see whether it would blunt the rise. Those three meals averaged a rise of +34.3, against +20.4 for the 37 meals I did not walk after.
Read naively, that says walking made things worse. It does not, and the reason is worth more than the number: I only walked after meals I expected to spike. Every one of the three was a breakfast containing fruit, eaten in the morning window when my glucose was already climbing. The comparison is confounded by the fact that I chose when to do it.
Light walking that interrupts prolonged sitting does improve post-meal glucose handling in controlled studies. [5] My three data points cannot test that, and I am including them because leaving out the result that disagreed with me would make everything else here less trustworthy.
What this does not show
This section matters more than the findings, so it is not at the bottom in small type.
- One person, sixteen days, 41 meals. No control, no blinding, a sample size of me. Nothing here establishes what a food does in general.
- I am on metformin. Every reading was taken by someone on a glucose-lowering medication.
- A CGM is not a blood test. It measures interstitial fluid, runs roughly five to fifteen minutes behind the bloodstream, and individual readings carry real error. Good for patterns; not good for arguing over single digits.
- GMI is not a lab A1c. The estimate above is derived from CGM data and commonly differs from a laboratory result by up to about half a percentage point. [1]
- I chose what to test. The foods above were picked because I was curious, which is not a representative diet — and is exactly why removing them barely moved the average.
- Several comparisons are confounded by choice, as the walking result shows plainly.
Your numbers will not be mine. That is not modesty. Individual responses to identical meals vary widely between people, which is why generic food lists disappoint so many people who follow them carefully.
What the research supports, as distinct from what I found
- Meal order. Vegetables and protein before carbohydrate markedly reduces post-meal glucose and insulin versus the same meal in reverse order. [4]
- Walking after meals. Short bouts of light walking that break up sitting improve post-meal glucose handling. [5]
- Lifestyle change outperformed medication in prevention. In the Diabetes Prevention Program, an intensive lifestyle program cut progression to type 2 diabetes by 58%, against 31% for metformin. [6]
- Remission is possible for some people, and is not a cure. In the DiRECT trial, structured weight management produced remission in 46% of participants at one year. [7] I use remission rather than reversal deliberately, and I have not achieved it.
If I were starting again tomorrow
Measure first, for a week, changing nothing. It is the only week you cannot reconstruct later, and without it every change afterwards is compared against a guess.
Then one change, one week, one test. Wash your hands. Read the peak. Put the carbohydrate last. Take the log to your doctor rather than a decision — particularly before changing anything while on insulin or a sulfonylurea, where eating differently can cause low blood sugar.
And expect to be wrong about at least one food you are certain about. I was wrong about the one I thought was healthiest, and finding that out was worth the whole sixteen days.
The printable log I used is free. Four pages — a daily blood sugar log with space for before and one hour after each meal, a weekly meal planner, a movement log and a weekly review. It comes with the Blood Sugar Wellness Bundle, alongside a seven-day meal structure and a movement framework. No cost, no purchase.
Disclosure: I am the founder of PureNovus, which publishes paid guides. This article is free, nothing in it is gated, and the method is described in full whether or not you ever buy anything. It is educational and is not medical advice.
References
- Bergenstal RM, Beck RW, Close KL, et al. Glucose Management Indicator (GMI): A New Term for Estimating A1C from Continuous Glucose Monitoring. Diabetes Care, 2018; 41(11): 2275–2280.
- American Diabetes Association. Standards of Care in Diabetes — 2024. Diabetes Care, 2024; 47 (Supplement 1).
- Porcellati F, Lucidi P, Bolli GB, Fanelli CG. Thirty Years of Research on the Dawn Phenomenon: Lessons to Optimize Blood Glucose Control in Diabetes. Diabetes Care, 2013; 36(12): 3860–3862.
- Shukla AP, Iliescu RG, Thomas CE, Aronne LJ. Food Order Has a Significant Impact on Postprandial Glucose and Insulin Levels. Diabetes Care, 2015; 38(7): e98–e99.
- Buffey AJ, Herring MP, Langley CK, Donnelly AE, Carson BP. The Acute Effects of Interrupting Prolonged Sitting Time in Adults with Standing and Light-Intensity Walking on Biomarkers of Cardiometabolic Health in Adults: A Systematic Review and Meta-analysis. Sports Medicine, 2022; 52(8): 1765–1787.
- Knowler WC, Barrett-Connor E, Fowler SE, et al. Reduction in the Incidence of Type 2 Diabetes with Lifestyle Intervention or Metformin. New England Journal of Medicine, 2002; 346(6): 393–403.
- Lean MEJ, Leslie WS, Barnes AC, et al. Primary Care-Led Weight Management for Remission of Type 2 Diabetes (DiRECT): An Open-Label, Cluster-Randomised Trial. The Lancet, 2018; 391(10120): 541–551.