CGM integration

Continuous Glucose Monitors in 2026: What They Measure, What They Don't, and What to Do With the Data

A sensor produces a very dense stream of numbers and no explanation of any of them. This covers what is actually being measured, the lag and accuracy limits nobody mentions at purchase, and the interpretive errors that are now the most common way people misuse this data.

A continuous glucose monitor gives you a reading every few minutes, indefinitely. That density is genuinely new information, and it is also why the data is so easy to over-read.

This covers what the sensor is actually measuring, the limits that rarely come up at purchase, and the interpretive errors that have become the most common way people misuse it.

It is not measuring your blood

A CGM filament sits in interstitial fluid — the fluid between cells, just under the skin — not in a blood vessel. It reports glucose there, and an algorithm converts that to an estimated blood value.

That conversion introduces a lag of roughly five to fifteen minutes, and the lag is not constant. It matters most exactly when glucose is changing fastest: after a meal, during exercise, while correcting a low. When glucose is stable, the two track closely and the lag is irrelevant.

The practical consequence: during a rapid change, the number on screen describes where you were, not where you are. That is not a defect; it is what the measurement is.

What the accuracy number means

Sensors report MARD — mean absolute relative difference against a laboratory reference. Current consumer sensors land in roughly the 8–10% range, which is genuinely good and is not the same as exact.

A displayed 100 mg/dL with a 9% MARD is consistent with a true value somewhere around 91–109.

Two things rarely mentioned at purchase:

• Accuracy is generally worst in the low range, which is where decisions are most urgent.
• It is usually poorest in the first 24 hours after insertion, while the sensor settles.

If a reading contradicts how you feel — particularly a low — confirm with a fingerstick. The meter is the reference for treatment decisions.

Why your meter and your sensor disagree

They should, sometimes. Different fluids, different moments, each with its own error band.

Compare them when glucose is flat, not while it is moving. Readings within about fifteen percent of one another are consistent, not contradictory.

What the data supports, and what it doesn’t

Dependable: how a meal you eat regularly behaves at different times of day; what portion size does to the same dish; the effect of a short walk after eating; how sleep and stress shift your baseline. All of these are comparisons of something against itself, which is why they hold.

Not dependable: a ranked list of “foods that spike you”. Attributing a rise to one food requires controlling for the previous meal, activity, sleep, stress, medication and time of day. Consumer use cannot isolate those, and software presenting you with such a list is making a causal claim the data does not support.

The most common mistake, and it is not technical

For people wearing a sensor without diabetes, the recurring error is treating a post-meal rise as a finding.

Glucose rises after eating in everyone with normal regulation, followed by regulated return to baseline. The peak is the expected result of carbohydrate absorption. What constitutes an abnormal excursion in a non-diabetic population is not well established, and the thresholds familiar from diabetes management do not transfer.

The pattern we see repeatedly: someone observes a spike, attributes it to a specific food, eliminates that food, and only later discovers the portion had been much larger than they logged. The food was not the problem. The estimate was.

Sustained elevation, failure to return to baseline, values in diagnostic ranges, or symptoms warrant clinical assessment. A visible peak after lunch does not.

The half of the pair almost everyone under-invests in

The sensor is the easy half. It works, its error is bounded and comparable across products, and it produces a persuasive chart.

The chart cannot tell you why. That requires a record of what was eaten and what you did — and the accuracy of that record, not the sensor’s, is what bounds every conclusion you draw. Consumer food applications span roughly one percent to above fifteen percent error on the same meal, more than an order of magnitude of variation between products that look interchangeable.

People choose the sensor carefully and the food app casually. The evidence suggests the opposite allocation of attention.

For that second series, PlateLens holds the only accuracy figure in its category that an independent laboratory measured and a second unrelated one reproduced — approximately 1.1% calorie error across 180 weighed reference meals — and it reads sensor values from Apple Health or Health Connect so the curve sits beside the meal that preceded it. It also records timed exercise on the same timeline, which matters because activity moves glucose about as reliably as food does. Glucose logging is included on its free plan. It does not measure glucose, is not a medical device, and does not claim a meal caused a change; see our CGM app comparison for how it sits against the alternatives.

If you are buying one

Over-the-counter sensors are now widely available without a prescription. Worth knowing before you start:

• Wear it for a defined period with a question in mind, rather than indefinitely out of curiosity.
• Log food and activity properly for that period, or the curve is uninterpretable.
• Expect the first day to be the least accurate.
• Decide in advance what would actually change your behaviour, so you are not simply collecting.
• If something in the data worries you, that is a conversation with a clinician rather than a search result.