14 days with a glucose sensor: your diet, with real data

You are sitting at the breakfast table. Your phone says 154. Yesterday, after the same bowl of muesli, it said 118. Same bowl, same spoon, same time of day. And now you sit there and wonder what you did wrong.

Probably nothing.

This is exactly the point where most people give up. Some put the sensor away because it is obviously measuring nonsense. Others cut out oats forever, because of one single number on one single morning. Both are understandable reactions to the same experience: the device delivers more data than it delivers context.

I would like to show you a third way. It takes 14 days, it is less comfortable than any app recommendation, and at the end there is something that genuinely belongs to you: your own pattern, in three to five sentences that you could defend against questions.

What is actually being measured under your skin

The device looks like a meter for blood. It is not one.

Under the round patch sits a wafer thin filament, a few millimetres deep in the fat tissue beneath your skin. There it does not swim in blood, it swims in the fluid that stands between your cells. An enzyme on the filament converts glucose, a tiny current is produced, and this current gets translated into a number. A measuring point every few minutes, and the software smooths what lies in between.

Picture a river, and beside it a quiet side arm. What happens in the river arrives in the side arm too. Just later, and a little blurred. Your sensor stands at the side arm.

The range you read in many guides is 5 to 15 minutes. That number comes from manufacturer information and secondary literature, not from this study. What is documented is 4 to 10 minutes. The difference sounds pedantic, but it is the core of this whole topic: many numbers circulate about sensors that nobody traces back to a source any more.

More important than the exact number of minutes is the second observation from the same work. Two sensors on the same abdomen, same person, same hour, and still a timing difference of almost seven minutes. If two parallel sensors do not run in sync, then your new sensor will not run in sync with the old one after a change either.

In clinical practice I observe that many people see restless values on the first day with a new sensor and draw hasty conclusions from them. I have deliberately not cited systematic data on this in this article, because I did not find any solid data. What is documented is only that two sensors on the same person can differ in time. The rest is experience, and that is what I call it.

So the first rule is set before you have even eaten anything: you are working with a signal, not with a laboratory value. And now you can follow why the lag figure in your app is an approximation and not a stopwatch.

How accurate this is, and why you can still learn something

"I was at 160." I hear this sentence often in my consultations, and it sounds like a finding. It is not one. It is a sensor statement.

The difference is not hairsplitting, not since there has been clean data on it. The most interesting work comes from Bath in England and is only a few months old.

The field describes sensor accuracy with the abbreviation MARD, the mean absolute relative difference. It says by how many percent a sensor value deviates from the reference value on average. Many specific percentages circulate, almost always from manufacturer documents, and they get passed on unchecked. So I am not naming any here, and prefer the number from the randomised study above. It describes exactly the situation you are in: healthy, at home, with a meal in front of you.

You could write the sensor off because of this. In my view that would be too early. Because a systematic upward offset is largely irrelevant when you compare yourself with yourself. If your device lifts every value by the same amount, the difference between Monday and Wednesday still stands. What gets lost is the absolute number. What remains is the pattern.

Three rules follow from this, and they apply to the whole rest of this article. First: you compare only within the same sensor cycle. Second: you do not adopt anyone else's thresholds as a target. Third: at the start of every new sensor you note your morning fasting range, so that you know the level at which this device speaks. And now you can explain why two people with an identical meal and an identical metabolism could still post different numbers.

Why this is interesting at all in people without diabetes

Your friend eats the same bread roll as you. Her curve stays almost flat. Yours climbs steeply. You are sitting at the same table, you are eating the same thing, and your bodies are telling two different stories.

That is the finding the whole idea of self measurement rests on. And it is well documented.

From the perspective of Clinical Psychoneuroimmunology this is no surprise. A meal never meets a neutral laboratory. It meets a system that is currently in a particular state. The microbiome can help decide how fast starch becomes sugar. The internal clock appears to help determine how willingly the beta cells answer. The sleep of the night before can change how sensitive muscle and liver are to insulin. And the stress axis can join in via cortisol, without asking you.

Nerves, immune system, metabolism and hormones are not separate chapters here. They are four voices in the same choir, and the curve on your phone is the sound that comes out.

One boundary, so you do not expect the wrong thing. If what mainly occupies you is the weight loss question, whether a sensor could change something about your weight, I have described that in detail elsewhere: in Blood sugar and weight loss: what a CGM sensor shows. This article is not about weight. It is about method, about how you arrive at a statement about yourself in the first place.

And now you can argue why you do not need a sensor to confirm tables, but to find out where you deviate from them.

What is normal in metabolically healthy people, in numbers

The most common expectation people have when they stick on a sensor is: flat. A line that barely moves, because surely that would be healthy.

The expectation is charming, and the reference data show something else.

And now the trap almost every app falls into. The metrics your program shows you, time in range, time above, time below and coefficient of variation, come from an international consensus paper from 2019. That paper was written for people with diabetes. The target ranges named there are treatment goals of a therapy, not normal values for healthy people. Anyone who applies them to themselves is borrowing a foreign yardstick.

Two curves, the same number of calories

So that you know what we are talking about, here are two schematic courses. They are not measured, they are drawn, and they show exactly the two shapes that the next two weeks are about.

And now you know why the expectation of "always flat" would lead to disappointment by the first breakfast at the latest, even though nothing has gone wrong.

Days 1 to 3: just look, change nothing

Now comes the part where most people lose patience. The sensor is on, the app shows numbers, and the reflex is overwhelming: drop the bread right now, cut the banana right now, do something differently right now.

Please do not. Not yet.

Before we go into the phases, here is the whole arc at a glance. Four phases, fourteen days, each phase with a single task.

Back to the first three days. They feel like lost time, and they are the opposite of that. Without a baseline you have nothing to compare against later. Every change you make on day 2 makes day 9 unreadable.

What you collect in this phase is four things. First, your morning range before breakfast, over three days, so that you know the level of this sensor. Second, your night course, along with a note on which side you slept. Third, your usual daily rhythm with the times of your meals. Fourth, your sleep duration, honestly noted, not estimated.

A practical note for the start, without turning it into a demand: applying a sensor for the first time is sensibly done with medical guidance. Not because it is complicated, but because the first questions are usually not technical ones, they are questions of content. What you are measuring, why you are measuring it, and where the limits lie, becomes clear faster in a conversation than in an instruction leaflet.

And now you can explain why a sensor worn for only three days often creates more confusion than clarity.

Days 4 to 7: the same food, different conditions

"Rice is bad for me. I have known that since Tuesday." I hear this sentence often, and it is usually premature.

Because before you can say anything about a food, you need to know how much your own response varies from day to day. There is a study on exactly that, and I consider it the most important one in this whole article.

This one number is the reason why this plan takes 14 days and not three. And it is the reason for the most important rule of the next phase: every test is done at least twice, better three times.

Why the same meal turns out differently on different days has several reasons. The best documented one is the time of day.

The second big influence is what lies in the stomach before the carbohydrates. The mechanism behind it is called gastric emptying and incretins, and it was made visible in a very small but very clean study: eight people with diet controlled type 2 diabetes drank a broth 30 minutes before a potato meal, once with 55 grams of whey protein, once without. With protein beforehand the stomach emptied most slowly, the glucose area was lower, and GLP-1 rose most strongly (Ma et al. 2009, [Case Series, n=8]).

A third variable worth trying: the degree of processing. Whole fruit versus puree versus smoothie was even part of the design in the accuracy study from section 2. Why texture and processing can make so much difference I have described more fully in Unprocessed food: why real food fills you up. For your phase 2 this is enough: the same fruit once as fruit, once as a smoothie, same amount, same time of day.

And now you can argue why a statement about rice takes you four days and not four minutes.

Days 8 to 11: testing the levers

Now it gets interesting. Because now you finally get to change something.

Four days, four levers, each lever tested against itself. The principle comes straight from the studies that examined these levers: same meal, same time of day, at least one day apart, and exactly one changed condition. That is a crossover design in miniature, and you can rebuild it at home.

Before we start, one piece of honesty up front: almost all of these studies were done in people with type 2 diabetes, prediabetes or obesity. The percentages from them cannot simply be transferred to healthy people. That is exactly why you test them on yourself.

Lever 1: the order on the plate

The same group later repeated the experiment in randomised form in 15 people with prediabetes, with three orders and blood samples over three hours. Peaks were more than 40 percent lower with "carbohydrates last", the area 38.8 percent lower (Shukla et al. 2019, [RCT, n=15]). So the effect also shows up before the diabetes threshold. What it does in a metabolically healthy person is not answered by this, and that is your test for day 8.

Lever 2: movement after eating

Two details are worth having in your test plan. First: the timing counts more than the total duration. In ten older people with elevated fasting values, three times 15 minutes of walking after meals beat 45 minutes of walking in one block in the hours after dinner (DiPietro et al. 2013, [RCT, n=10]). Second: in 41 people with type 2 diabetes, ten minutes of walking after each main meal lowered the area under the curve more than the same walking time at some point in the day, with the largest difference after dinner (Reynolds et al. 2016, [RCT, n=41]).

Lever 3: vinegar with the meal

The cheapest test candidate in the whole plan. A systematic review with meta-analysis found a standardised mean difference of minus 0.60 for the glucose area and minus 1.30 for the insulin area with vinegar taken with a meal (Shishehbor et al. 2017, [Systematic Review]). The confidence interval for glucose did however reach close to zero.

In twelve healthy volunteers a clear dose response relationship appeared: the more acetic acid added to the white bread, the lower glucose and insulin after 15 to 45 minutes, and the higher the reported satiety (Östman et al. 2005, [RCT, n=12]). They worked with 18 to 28 millimoles of acetic acid to 50 grams of carbohydrate. I deliberately do not convert that into tablespoons here, because an everyday dose does not follow directly from it. In any case I find the second finding more interesting: satiety rose along with it. Perhaps that is the more exciting part.

Lever 4: the night before

The classic on this is older: eleven young men spent six nights with only four hours in bed and afterwards showed poorer glucose tolerance, raised evening cortisol and increased sympathetic activity (Spiegel et al. 1999, [Case Series, n=11]). Small, male, young and under extreme conditions. Much closer to everyday life is an analysis from PREDICT: in 953 healthy adults with more than 8,000 standardised meals, sleep efficiency and sleep timing were linked to the glucose response to breakfast the next morning, also within the same person (Tsereteli et al. 2022, [Cohort, n=953]). Notable about it: it was not only less sleep that counted, but also sleep that was different from usual.

That leaves stress. And here I have to pull the brake. What is documented is that cortisol given as a night time infusion in eight healthy people lowered glucose effectiveness and raised glucose as well as free fatty acids, despite higher insulin (Nielsen et al. 2004, [Case Series, n=8]). That was an infusion, not a meeting. That your argument on the phone or your deadline lifts the curve is mechanistically plausible and I observe it often in clinical practice. As a human study with an everyday stressor I cannot document it for you here. That is exactly how you should read it.

One note on scope: the levers themselves, meaning order, combination and timing without a sensor, I have described in detail in Avoiding glucose spikes: order, combination, timing. That article says which levers exist. This one says how you check whether they show up in you at all. That is a difference that matters to me: not "eat vegetables first", but "this is how you check on yourself whether the effect from these studies becomes visible in you".

And now you can explain why a lever that works for your colleague might still change almost nothing in you, without either of you doing anything wrong.

Days 12 to 14: what the curve says, and what it does not say

In the end you want to know: was that good or bad? And that is exactly the question the sensor does not ask. It shows a shape. You have to add the meaning, and there are five places you can orient yourself by.

The third row is the one hardly anyone knows, and it is the most interesting.

Before you turn a night time drop into a diagnosis, check the technology. In a very small study of four healthy men, each wearing four sensors at once on the abdomen, individual sensors dropped by around 40 mg/dl over 30 to 60 minutes, precisely when the person was lying on that sensor (Mensh et al. 2013, [Case Series, n=4]). A modelling study in 72 sensor wearers puts such compression artefacts at a median of 45 minutes and 24 mg/dl, more frequent at night than during the day (Facchinetti et al. 2016, [Cohort, n=72]).

The counterposition, and why it belongs in this article

Part of this is also the most famous work behind the sentence "swings do more harm than height". In a French study, 21 people with type 2 diabetes were compared with 21 healthy controls. Only the amplitude of fluctuation and the area after meals correlated with a marker of oxidative stress in urine, but not HbA1c or fasting value (Monnier et al. 2006, [Cohort, n=21]). That is a small case control study in people with diabetes, measuring a laboratory value and not a heart attack. It cannot be transferred to metabolically healthy people, however often it is cited for that purpose.

A single metric rarely says as much as it appears to say. That is true for the calorie count on the package, as I described in The calorie myth: why "eat half of it" is rarely enough, and it is true for cholesterol, as you can read in Cholesterol and science. The glucose peak joins this family: an interesting signal that is easily over interpreted without context.

If you would rather not interpret your 14 days alone: below this article you will find the option to book an appointment.

And now you can say which place on your curve actually tells you something about your everyday life, and which one you are allowed to leave in peace.

When measuring becomes control, and who this is not for

There is a moment when the device takes the lead. It announces itself quietly. You are sitting at dinner with friends and looking at your phone. You no longer order what you feel like, you order what made a nice curve yesterday. You feel relieved when the number stays low, and guilty when it rises.

That is the point where a learning device tips over.

I want to be honest here about what is documented. There is no frequency figure for how often people without diabetes develop disordered eating through self measurement. What does exist is a professional warning: the review from 2024 mentioned earlier explicitly names the lack of data on unfavourable effects on eating behaviour as one of the biggest open gaps in use by people without diabetes. A research gap is not proof of danger. But it is not an all clear either.

Taking the sensor off in such a moment is not failure. It is a good decision, and one that your body is suggesting to you. In my consultations I hear this story regularly, and relief almost always follows once the device is gone.

Who these 14 days might be worthwhile for

And if your curve worries you? Then the next step is not a second sensor. The diagnosis of diabetes and prediabetes rests, according to the 2025 Standards of Care of the American Diabetes Association, on HbA1c, fasting plasma glucose and the oral glucose tolerance test. A tissue sensor is not a diagnostic criterion there (ADA 2025, [Guideline Document]). The sensor may ask you a question. It does not give the answer.

And now you know why the most honest answer to the question "should I stick one of those things on" begins with a counter question: what exactly do you want to find out?

Common questions about glucose sensors without diabetes

The glucose sensor in a wider context

A glucose curve is never only a glucose curve. It hangs on the insulin sensitivity of your muscles, on hormones, on your training history and on the question of how your brain reads fullness. Several paths lead onward from here.