Continuous glucose monitors show glucose trends reasonably but are imprecise during exercise, and no trial shows they improve fuelling or performance in athletes without diabetes
Interpretation
Accuracy evidence is consistent but small: in 13 recreational runners only about half of exercise readings were within 10% of finger-prick values and the authors judged CGMs valid for trends, not points (Mason 2024); error rose from 12% at rest to 34% during exercise in 13 para cyclists (Weijer 2024); CGMs overestimated glucose by about 0.9 mmol/L and time at high glucose about 4-fold in 15 healthy adults (Hutchins 2025); over-the-counter sensors in 39 young adults had 12.9% error and a 6-minute lag (Prakash 2026). Normative data show brief daily highs and lows in 153 healthy people (Shah 2019) and normal variability in 12 elite racewalkers (Bowler 2025), while athletes' values can look abnormal against narrow bands (Thomas 2016). Two expert reviews describe CGM use for fuelling or performance as untested, with no consensus on interpretation (Bowler 2023; Flockhart 2024). Graded limited: validation studies are small and short, few are in triathletes, and the performance question has no trial at all.
Practical implication
For healthy triathletes, a CGM is at most a way to view glucose patterns, not a fuelling gauge: fuel to evidence-based carbohydrate guidance and practice, don't act on single readings during exercise, don't compare yourself with diabetes thresholds, and never cut carbohydrate to flatten a trace. People with diabetes should follow their clinical team. Benefits for performance are unproven and the advice here is inference from small validation studies.
Sources
9 sources, independently verified.
Review of CGMs in athletes: athlete-targeted devices are marketed as a 'fuelling gauge', but whether interstitial glucose has meaning for fuelling or performance, and whether CGMs give reliable and accurate information in sport, is untested. Caveats include sensor dislodging, data loss, cost, possible bans and the need for professional interpretation.
Current opinion: no consensus on how to interpret CGM data in athletes with normal glucose control, and well-defined approaches to use glucose monitoring to improve endurance performance or health are lacking. Endurance athletes' glucose regulation can differ from non-athletes' and is affected by demanding training and by energy and carbohydrate availability.
During 30-90 min of steady treadmill running, CGM error versus finger-prick glucose (MARD) was 13.8% on the arm and 11.4% on the chest; only about 47% and 50% of readings were within 10%, variability was large and accuracy was worse at low glucose. Trends in mean glucose were tracked well: valid for trends, not point values.
High readings occurred almost only after meals and during exercise. CGM error versus capillary glucose (MARD) was 12% at rest but 34% during exercise, 41% in hand cyclists (sensor near working arm muscle) versus 24% in cyclists. Para athletes; small sample.
CGM read about 0.9 mmol/L higher than capillary samples fasting and after eating; bias varied by food and by individual; CGM overestimated time at high glucose about 4-fold, about 2-fold after adjusting for baseline differences. Healthy non-athletes at rest.
First independent test of over-the-counter CGMs: MARD 12.9%, 66.5% of readings within 15%/15 mg/dL; worse after sugar (15.3%) than placebo (10.6%) and during rapid change; exercise had no consistent effect; about 6-minute lag; 13.8% of trend arrows classed as moderate to extreme risk. Online ahead of print.
Mean glucose 98-99 mg/dL in all age groups except over-60s (104 mg/dL). Healthy participants spent a median of about 30 min a day above and 15 min a day below the thresholds used in diabetes care. Normative data; not athletes, not exercise-specific.
On a high-energy, high-carbohydrate diet with hard daily training, day-to-day glycaemic variability was similar to values reported in healthy non-athletes; 24-hour mean glucose was higher than overnight and tended to be lower in women than men. Small elite sample under controlled conditions.
Over about 6 days, 4 of 10 athletes spent more than 70% of monitoring time above a narrow researcher-chosen glucose band even excluding post-meal periods, and 3 of 10 had fasting values in a prediabetes range by a common definition; only 1 spent much time low, with low energy intake. Small, older sensor generation; shows how athletes' values can be read as abnormal.