Continuous Glucose Monitors for Healthy Adults

You can now buy a continuous glucose monitor without a prescription and watch your blood sugar all day. But for a healthy person the device reads a noisy, approximate signal, most of the "spikes" it flags are normal physiology, and no trial has shown that reacting to them makes an already-healthy person healthier — it is a curiosity-and-education tool with real failure modes, not a longevity instrument.

A continuous glucose monitor (CGM) is a coin-sized sensor worn on the arm that reads glucose in the fluid between your cells every few minutes and streams it to your phone. Built for people with diabetes — where it is genuinely life-changing — it entered the wellness market in 2024 as an over-the-counter gadget aimed at everyone else. This article is the device reality-check: what these sensors can and can't tell a healthy person, how accurate they actually are, whether watching your glucose changes any outcome, and the ways the data misleads. For the glucose physiology — why spikes might matter, glucotypes, and how to flatten the curve — see Glycemic index; for the fuel-switching side, Metabolic flexibility.

What the evidence says

Strong:

Healthy people routinely spike into "abnormal-looking" ranges, and it is normal. In the largest community study to date — 560 adults with a normal HbA1c (a marker of average blood sugar over about three months) and a normal fasting glucose, wearing a sensor whose readings they could not see — glucose sat in the 3.9–7.8 mmol/L (70–140 mg/dL) band about 87% of the day, which leaves roughly three hours a day above 7.8 mmol/L (140 mg/dL).^[1] A smaller earlier study that excluded people with obesity, enrolled younger participants, and let them see and calibrate their own sensors reported 96% and half an hour; those are the numbers still quoted everywhere, and they are the outlier.^[2] In a Stanford cohort, people whose blood sugar was normal on every standard test still reached prediabetic glucose levels about 15% of the time and diabetic levels about 2% of the time.^[3] A transient post-meal peak is expected biology, not a warning light.
Two different-brand sensors on the same person disagree enough to change the conclusion. Worn at the same time on 172 adults with type 2 diabetes, a Dexcom and an Abbott sensor disagreed by about 0.7 mmol/L (13 mg/dL) not on individual readings but on the person's two-week average — the summary number a user would actually act on. The same sensor on the same person, worn again three months later, shifted by a similar margin.^[4] In non-diabetic adults, the disagreement was large enough to flip which meals looked best — eating the "best" meals chosen by one sensor delivered roughly half the glucose benefit when scored by the other.^[5]

Moderate:

The devices are useful pattern-viewers, not lab instruments. Even the accurate flagship sensors are off by roughly 9% on a typical reading, and accuracy is worse in the low range where wellness users worry about "dips."^[6] The reading also lags real blood glucose, most when glucose is moving fast.^[7]

Weak / preliminary:

Personalized programs built partly on CGM data barely beat generic healthy-eating advice. The largest trial (ZOE) tested a multi-input program — CGM plus blood-fat and microbiome testing — against standard guidance and moved only one primary outcome, triglycerides, by a small amount; cholesterol did not budge, and the benefit clustered in the most adherent participants.^[8] No trial isolates the CGM as the useful ingredient.

Caution:

There is no evidence that watching or manipulating glucose improves any hard outcome in people without diabetes, and clinical guidance for interpreting a CGM trace exists only for people with diabetes.^[9] Self-reported distress and food fear are common among wellness users, so the honest risk-benefit for a healthy person is "unproven benefit, plausible harm."

The over-the-counter turn

For fifteen years a CGM needed a prescription and a diabetes diagnosis. That changed in 2024. The US Food and Drug Administration (FDA) cleared the first over-the-counter glucose biosensor, Dexcom's Stelo, in March 2024.^[10] Abbott followed within weeks with two deliberately different products: Lingo, aimed squarely at general-wellness consumers, and Libre Rio, which despite sharing the shelf is indicated for adults with type 2 diabetes who do not use insulin.^[11] Stelo lasts up to 15 days — Dexcom notes about one sensor in five falls short of that — and shows a glucose value and trend every 15 minutes; Lingo runs 14 days and pairs with a coaching app. Sensors cost roughly $40–55 each, or about $89 a month on subscription (mid-2026 prices).

Two facts about that clearance matter. First, these devices are cleared for people not taking insulin — Stelo's label explicitly excludes anyone with problematic hypoglycemia, and it carries no low-glucose alarm. Second, "cleared" here means the FDA judged Stelo substantially equivalent to a device already on the market — an ordinary class II clearance, not a verdict that the thing does you any good.^[12] It says nothing about whether tracking your glucose extends your health or lifespan. The marketing implies the second; the paperwork only supports the first. One wellness brand, Signos, has since obtained its own clearance — for the app rather than the sensor, which is Dexcom's — and it is the first whose approved wording mentions helping users "maintain a healthy weight", a notable drift in what "FDA-cleared" is being used to sell.^[13]

The category has since widened. In June 2026 the FDA cleared Stelo over the counter for anyone aged two and over who does not use insulin, making it the first over-the-counter monitor marketed for children.^[14] The same clearance also carries a caution the wellness marketing does not: alongside the existing warnings that the device is not for people with problematic low blood sugar or on dialysis, the FDA now advises that people with a history of disordered eating should talk to a clinician before using it.

What a healthy glucose curve actually looks like

The single most important thing to know before wearing one: a healthy metabolism produces a busy, spiky trace. In the largest blinded study of adults without diabetes, people with an entirely normal HbA1c and a normal fasting glucose still spent about 87% of the day between 3.9 and 7.8 mmol/L (70–140 mg/dL) — leaving roughly three hours a day above that line, and about a quarter of an hour above 10.0 mmol/L (180 mg/dL). The authors' own conclusion is that clinicians should expect many patients with normal blood work to produce sensor reports showing values above 10.0 mmol/L (180 mg/dL).^[15] Short excursions into the prediabetic range likewise happen in people whose fasting glucose, HbA1c and glucose-tolerance test are all perfectly normal.^[16] Nor does the sensor reliably turn up hidden trouble: among more than 1,100 adults past 75, a monitor found almost nothing a standard blood test had missed — not one participant with a normal or prediabetic HbA1c had an average glucose above 7.8 mmol/L (140 mg/dL), which led the authors to conclude that HbA1c already captures what matters at that age.^[17]

Because no one has defined what a "normal" wellness-CGM trace should look like, experts can't agree on how to read one. When 18 specialists were each shown the same 20 deliberately difficult reports from non-diabetic adults, they could not agree on who needed follow-up: on individual reports the share recommending it ranged from barely half to unanimous, and agreement across the panel was poor once chance is accounted for (κ ≈ 0.36, where 1.0 is perfect and 0 is a coin flip). More than half would refer someone who spent as little as 2% of the day above 10.0 mmol/L (180 mg/dL) even when both their HbA1c and fasting glucose were normal.^[18] If the specialists can't standardize the interpretation, a consumer app's "your glucose is elevated" verdict is on very soft ground. That interpretive layer is getting thicker, not thinner: a 2026 update to Stelo's app added a per-meal "Glucose Response Score" and AI-written summaries of what your readings mean, sitting on top of a signal that specialists cannot agree how to read.^[19]

How accurate are they, really?

CGMs are impressively convenient and only approximately accurate. The industry yardstick is MARD — mean absolute relative difference, the average percent gap between the sensor and a lab reference. The best current sensors land around 8–14% in head-to-head testing.^[20] A 9% MARD sounds tight until you translate it: at a true reading of 5.0 mmol/L (90 mg/dL), a typical sensor value could be roughly 0.4 mmol/L (8 mg/dL) off in either direction — and accuracy tends to be worse at low glucose — measured in adults with type 1 diabetes, since testing it means driving glucose down deliberately — exactly where a healthy user might panic about a "dip."^[21] (Note too that the published MARDs come from the prescription flagships; the over-the-counter versions ride the same platforms but their own accuracy hasn't been independently established.)

Three quirks matter for a healthy person reading meaning into wiggles:

The number lags your blood. A CGM samples interstitial fluid, not blood. The pure physiological delay is small, about 5–6 minutes,^[22] but once you add the device's own signal-smoothing the total lag averages around 12 minutes, stretching toward 17 when glucose is falling fast — figures measured on a 2009 sensor generation, and still the best available, since no one has published an equivalent measurement on current hardware.^[23] The curve you see is a slightly stale, smoothed version of reality.
Brand is not truth. Same-brand sensors agree reasonably well, but different brands diverge three to four times more — in one standardized-meal study the between-brand variability was about 13% versus under 4% within a brand.^[24] A "spike" your friend's Lingo shows might not appear on your Stelo.
Nighttime "lows" are often the mattress. A sharp overnight drop and rebound in someone who sleeps fine is usually a compression low — pressure on the sensor from lying on it — not real hypoglycemia. It is a well-recognized artifact that detection algorithms are specifically built to catch.^[25] Sensors can also be wrong in the low direction for far more mundane reasons: in late 2025 Abbott corrected around 3 million prescription FreeStyle Libre 3 sensors that could read falsely low because of a manufacturing fault, a recall the FDA rated its most serious class.^[26]

Does watching your glucose actually help?

This is the question the marketing skips, and the answer is: not in anyone like you. Randomized trials of glucose-guided behaviour change do exist — a 2026 review found seven of them in people without diabetes — but the benefit sits in one subgroup only. Pooled, glucose control improved in people with prediabetes, while no appreciable benefit appeared in adults with normal blood sugar, and body weight did not shift at all.^[27] The two newest trials show the pattern: both recruited people with overweight or recent weight loss rather than healthy adults, both were small — 35 and 49 participants — and the second calls itself a feasibility study and declares a senior author's consulting fees from a sensor manufacturer.^[28]^[29] That funding pattern is not incidental: across the 25 glucose-feedback trials pooled in a 2024 meta-analysis, 44% declared a conflict of interest tied to the monitoring industry — and not one of the 25 enrolled healthy adults.^[30] Reviewing the whole field in 2026, a JAMA Internal Medicine team concluded that only "limited and indirect" evidence supports monitoring even in prediabetes or obesity, and that it "should be deployed in response to a specific patient problem rather than as a default intervention."^[31]

The strongest-sounding evidence — commercial personalized-nutrition programs — doesn't isolate the monitor. The ZOE Method trial randomized 347 adults to a program combining CGM data with blood-fat and gut-microbiome profiling versus standard dietary advice; after 18 weeks only triglycerides moved on the primary analysis, by about 0.13 mmol/L, while the co-primary endpoint — LDL ("bad") cholesterol — did not shift at all, and most gains were concentrated in the most adherent users.^[32] That is a modest result from a bundle of interventions run by the company selling it — not proof that the glucose sensor was the active ingredient. The professional bodies reflect this, though the line has moved. The American Diabetes Association's 2026 standards recommend continuous monitoring at diagnosis and thereafter for people with diabetes — a wider tent than a year earlier, now reaching those on non-insulin treatments — and a separate graded recommendation extends to people with prediabetes, pairing the technology with licensed coaching.^[33] What the standards still do not contain is any recommendation to use one for screening or wellness in a person whose blood sugar is normal, and the guidance for interpreting a trace remains written for people with diabetes.

The honest summary is that everything a CGM can teach a healthy person about which foods spike them is a surrogate — the glucose curve — and no study has connected flattening that curve in an already-healthy adult to living longer or better. The physiological case that spikes matter is covered, with its caveats, under Glycemic index; it is a hypothesis with mechanistic support, not a settled outcome.

The failure modes

A noisy, lagged signal with no agreed-upon normal range, worn by someone primed to find problems, has predictable downsides.

It over-flags. Against finger-stick blood in 15 healthy adults, a sensor roughly quadrupled the time it judged them to be "above range" — still about double after correcting for each person's baseline offset — and rated a smoothie's glycemic impact at 69 where the blood test said 53. That last gap only just reached statistical significance, and the range of values each figure is consistent with overlapped heavily — 48 to 99 for the sensor against 40 to 69 for the blood test, on a glycemic-index scale where 100 is pure glucose. That range is where the true value most likely sits, so two of them overlapping this much means the comparison is suggestive rather than settled. The authors warned the pattern could push people toward "unnecessary food restrictions."^[34] Cutting fruit, legumes, or whole grains because they nudge a lagged sensor upward is the opposite of good nutrition advice.
Normal physiology reads as pathology. The dawn rise in glucose before you wake, the jump during hard exercise (adrenaline releases stored glucose — a workout can look "bad"), the bump after a poor night's sleep or a stressful meeting: all normal, all easy to misinterpret as damage.
It can breed food anxiety. Among 56 monitor users surveyed — a mix of people with and without diabetes — 89% reported positive changes to diet or activity, but 68% also reported fear of developing type 2 diabetes when they saw a high number, and eating-disorder symptoms tracked with distress about wearing the device.^[35] A scoping review that set 48 popular-media and sensor-company sources against 11 medical papers found the popular sources blaming glucose spikes for mood, sleep, brain fog and even skin ageing — effects the medical literature simply does not address, though both sets agreed on blood-vessel and inflammatory effects. Its authors are careful to say this does not prove the popular sources wrong, and raise the risk of disordered eating as a worry rather than a finding.^[36] The narrower charge — that commercial claims aimed at people without diabetes should be labelled misleading — is made directly by an independent review of the field.^[37] No trial has measured these harms, so they remain self-reported rather than quantified — but they are no longer merely the critics' worry. The FDA's own 2026 clearance language advises people with a history of disordered eating to speak to a clinician before wearing one, which is an unusual thing for a regulator to say about a wellness gadget with no demonstrated benefit.^[38]

In fairness to the other side: proponents argue that seeing your own data is a uniquely motivating nudge, and the "glucotype" research does show that some apparently healthy people carry genuinely dysregulated glucose that standard tests miss.^[39] That is a real, if small and mechanistic, finding — but it is an argument for a targeted check, not for continuous lifelong monitoring, and the "awareness improves health" claim remains untested.

Who might actually benefit — and how to run a useful two weeks

There are reasonable uses, though the evidence behind them is thinner than the marketing suggests. A borderline HbA1c has the best case: the pooled trials show glucose control improving in people with prediabetes, the one group where monitoring has moved anything.^[40] After gestational diabetes, a blinded sensor matched a formal glucose-tolerance test at picking up diabetes and prediabetes in the months after birth, where HbA1c missed cases — but that is a one-off diagnostic worn under supervision, not self-directed glucose-watching, and nothing shows the earlier detection changed anyone's outcome.^[41] Family history and unexplained symptoms have no trial support at all. In every case this is a test to run with a clinician, not an app. And the device can be a genuinely instructive short experiment for a curious healthy person, provided the goal is education, not surveillance.

If you wear one for a couple of weeks, get value without the anxiety:

Treat it as a teacher, not a scoreboard. Two weeks to learn your personal patterns, then take the sensor off. Continuous, indefinite monitoring has no demonstrated payoff and the most downside.
Run a few controlled comparisons rather than staring at every wiggle — the same portion of rice versus a slice of sourdough versus cold-then-reheated pasta tells you which staples to anchor on. The levers that flatten those curves — sequencing vegetables and protein before starch, a post-meal walk, vinegar, front-loading carbs to the morning — are laid out under Glycemic index.
Ignore the isolated scary reading. One overnight "low" is probably the mattress; one post-workout "spike" is adrenaline. Patterns over days mean something; single points rarely do.
Don't cut healthy whole foods because they raise the line. Fibre, fruit, and legumes are among the best-evidenced foods for longevity regardless of a transient glucose bump.

Bottom line

For a healthy adult, a continuous glucose monitor is a legitimately interesting window into your own metabolism and a legitimately weak longevity tool. It reports a lagged, roughly 9–12%-approximate, brand-dependent signal; the spikes it dramatizes are mostly normal physiology; no trial shows that reacting to them improves health in people without diabetes; and it carries a real risk of manufacturing food anxiety. Wear one for two weeks as an experiment if you're curious — then spend the subscription money on the things that are actually proven: fibre, fish, resistance training, sleep, and a Mediterranean-pattern plate. The lab markers worth tracking instead — HbA1c, fasting insulin, the ratio of triglycerides to HDL ("good") cholesterol — are covered under Glycemic index, Metabolic flexibility and Midlife labs.