CGM Exercise Accuracy and the Sensor-Outcome Paradox

Continuous glucose monitors can make a workout feel measurable down to the milligram. But the device that provides alerts, trend arrows, and a nonstop stream of numbers is not necessarily the one that measures glucose most accurately during exercise.

The only controlled head-to-head exercise comparison found lower raw sensor error with intermittently scanned CGM (isCGM) than with real-time CGM (rtCGM) in adults with type 1 diabetes. Yet a much larger real-world study found better overall glycemic outcomes among rtCGM users. The apparent contradiction matters because a CGM is both a measuring device and a behavior-changing device. Those are related jobs, but they are not the same job.

During cycling, the scanned sensor had lower error

In a secondary analysis of the ULTRAFLEXI-1 study, 22 adults with type 1 diabetes wore a FreeStyle Libre 1 isCGM and a Dexcom G6 rtCGM at the same time during a standardized 60-minute moderate-intensity cycling protocol. The researchers compared both sensors with capillary blood glucose measurements collected throughout the protocol.

That simultaneous-wear design is the study's main strength. Each person exercised under the same conditions while using both systems, which reduces the usual problems of comparing different people, workouts, fitness levels, glucose patterns, and sensor sessions.

Across the exercise period, the Libre 1 had a median absolute relative difference (MedARD) of 11.6% (interquartile range 5.6% to 19.6%), versus 14.6% (IQR 7.0% to 23.8%) for the Dexcom G6. The difference was statistically significant (p<0.0001). Lower MedARD means the sensor readings were, on average, closer to the capillary reference values.

The gap widened below 70 mg/dL, the range in which a wrong reading can lead to a poor treatment decision. The isCGM MedARD was 27.0% (IQR 17.0% to 34.6%), compared with 39.2% (IQR 31.8% to 46.8%) for rtCGM—an absolute difference of 12.2 percentage points (p=0.0001). In the 70 to 180 mg/dL range, the figures were 12.8% for isCGM and 16.1% for rtCGM (p<0.0001). Above 180 mg/dL, the difference narrowed to 8.0% versus 9.5% (p=0.0064).

Those results challenge a common assumption: more frequent display updates and predictive processing do not automatically produce more accurate sensor readings when glucose is moving during aerobic exercise.

Accuracy during exercise is only one part of device performance

The exercise finding does not mean isCGM produces better diabetes outcomes. A nationwide 2025 cohort study followed 7,786 people with type 1 diabetes for 24 months and found that rtCGM users had greater HbA1c reductions and lower glycemic variability than isCGM users. The result held in adults and children.

This was an observational cohort, not a randomized device trial. It therefore cannot prove that rtCGM itself caused the better outcomes; users may have differed in treatment access, education, insulin technology, or other factors. Still, its size and follow-up make it important counterevidence to the small exercise-accuracy study.

A plausible explanation is that alerts and visible data change what people do. In a 2023 randomized trial of adults with type 1 diabetes, 273 participants were assigned 2:1 to unblinded or blinded FreeStyle Libre monitoring. At 12 weeks, only the unblinded group showed a significant increase in time in range. The physical sensor class was the same; access to glucose readings differed. That does not isolate predictive alerts specifically during exercise, but it shows that seeing data can improve glucose management independently of a sensor's raw accuracy.

rtCGM can therefore win in daily life even if an isCGM reading sits closer to a capillary value during a particular workout. A useful alert that prompts carbohydrate intake before a low develops may matter more over 24 hours than a modest difference in point accuracy. The available studies support that explanation, but they do not prove it as the sole reason for the cohort result.

Newer devices show a similar resting pattern, with caveats

The exercise study used older hardware: Dexcom G6 and FreeStyle Libre 1. That limits any attempt to choose between current products based on its results. A separate parallel-wear comparison at rest, however, found a similar direction with newer systems. In 55 participants, the 2024 point-accuracy comparison reported a MARD of 8.9% for FreeStyle Libre 3 and 13.6% for Dexcom G7 (p<0.0001). Libre 3 also had 91.4% of paired readings within ±20 mg/dL of the reference, versus 78.6% for G7.

MARD and MedARD are related accuracy measures, but they are not interchangeable endpoints. More importantly, the newer-device study was performed at rest, not during exercise. It supports the possibility that the accuracy pattern extends across generations, but it does not answer whether G7 and Libre 3 behave the same way during cycling, interval training, resistance work, or sports with contact and rapid motion. The comparison also had manufacturer-adjacent context, so independent replication would add confidence.

Exercise makes every CGM harder to interpret

CGMs measure glucose in interstitial fluid rather than directly in blood. During exercise, interstitial readings can lag behind changing blood glucose. Changes in blood flow, tissue perfusion, hydration, and pressure on the sensor site can further affect accuracy. The ULTRAFLEXI-1 analysis documented the performance difference; it did not isolate why the rtCGM system had higher error.

That missing mechanism matters. It is tempting to blame algorithmic smoothing or prediction, but the study did not test that explanation. It also did not identify a rate of glucose change at which either system becomes misleading enough to prompt the wrong action. No published exercise comparison has supplied rate-of-change-stratified accuracy data for these CGM classes.

A statistically significant accuracy difference is not automatically a clinically important difference for every user or every workout.

What this means for athletes without diabetes

CGM readings can capture real physiological changes in people without diabetes. A 2026 systematic review and meta-analysis of 23 studies, including 18 randomized trials and 408 non-diabetic participants, found that exercise reduced 24-hour mean glucose by 0.17 mmol/L and mean amplitude of glycemic excursion by 0.27 mmol/L. A 2025 systematic review also found growing interest in CGM as lifestyle feedback for cardiovascular prevention, while concluding that cardiovascular-outcome evidence in healthy adults remains scarce.

Neither review establishes that one CGM class is more accurate during exercise in healthy athletes. No published head-to-head rtCGM-versus-isCGM exercise accuracy study has tested a non-diabetic population. The large error gap below 70 mg/dL in the type 1 diabetes cycling study may also have limited relevance for healthy exercisers, who do not routinely reach that range during typical aerobic activity.

For a non-diabetic user, the more relevant result is the 3.3-percentage-point MedARD difference in the 70 to 180 mg/dL range during the cycling protocol. It was statistically significant, but its practical importance for meal timing, training decisions, or long-term health remains uncertain.

The practical verdict

For athletes with type 1 diabetes, do not select a CGM solely on the exercise accuracy figures from one 22-person study. The isCGM system produced lower raw sensor error during standardized cycling, especially at low glucose values, but rtCGM users had better 24-month outcomes in a much larger real-world cohort. The evidence does not tell any one athlete which tradeoff will produce safer decisions across training and the rest of the day.

For biohackers without diabetes, claims that one current CGM is definitively more accurate during exercise run ahead of the data. CGM can show meaningful glucose dynamics, but the direct exercise comparison comes from type 1 diabetes, older devices, and one cycling protocol. The study is a credible signal—not a finished buying guide.