168 mg/dL. That was the reading from my smart ring at 3:15 PM on a Tuesday in July. I had just eaten a turkey sandwich and an apple. The Dexcom G7 on my arm — the actual continuous glucose monitor, the one with a needle, a transmitter, and FDA approval — said 103 mg/dL. The ring was off by 65 mg/dL. That is not a minor discrepancy. That is the difference between normal and diabetic. That is the difference between peace of mind and a panicked call to my doctor. And it is the central problem with non-invasive glucose monitoring in 2026.
I have been testing three devices that claim to track glucose trends without breaking skin: the Circular Ring 2, a Samsung Galaxy Watch with experimental PPG glucose algorithms, and a fitness band from a startup that shall remain nameless because I do not want to get sued. I wore them simultaneously with a Dexcom G7 for 28 days. I logged every meal, every reading, every discrepancy. I have 5,712 data points. The conclusion is not flattering. But it is important.
Here is what the data says. Non-invasive glucose monitoring is the hottest feature in wearable technology for 2026. The Circular Ring 2 announced blood glucose trend analysis using PPG sensors and machine learning. Samsung has been developing cuffless blood pressure and glucose estimation since 2020. Multiple startups are pursuing similar approaches. The promise is transformative: continuous metabolic monitoring for hundreds of millions of people who would never wear a CGM patch. The reality is that the technology is not ready. And the marketing is dangerously ahead of the science.
Calculate your BMI instantly with visual category charts and history storage.
All data stays in your browser — we never see it.Day 1 of the experiment was calibration day. I fasted for 12 hours, took baseline readings from all devices, then ate a standardized meal: 60g carbohydrates, 25g protein, 15g fat. The Dexcom showed a predictable curve: baseline 89, peak at 45 minutes at 142, return to 94 by 90 minutes. The ring showed: baseline 95, peak at 30 minutes at 178, return to 88 by 60 minutes. The timing was wrong. The magnitude was wrong. The trend direction was wrong. The ring thought I was spiking earlier and higher than I actually was. If I had been diabetic and adjusting insulin based on the ring, I would have overdosed.
By day 7, I had identified the primary failure modes. Motion artifact is the biggest problem. The ring's PPG sensor sits on the finger. Fingers move constantly. Typing, gesturing, holding coffee cups, gripping bike handlebars. Each motion introduces noise that the algorithm tries to correct, often over-correcting. During a 30-minute bike ride on the Butler Trail, the ring showed a glucose "crash" to 58 mg/dL. The Dexcom showed 108. The motion of gripping handlebars had confused the optical sensor into thinking blood flow had changed dramatically.
Skin temperature interference is the second problem. The ring measures skin temperature to adjust its optical readings. But skin temperature changes with environment, not just metabolism. On day 12, I walked from my air-conditioned apartment to a coffee shop on East 6th Street. The outdoor temperature was 104°F. My skin temperature jumped 4.1°F. The ring's glucose algorithm interpreted this as metabolic stress and showed a glucose spike to 171 mg/dL. The Dexcom showed 96. The ring had confused heat stress with carbohydrate absorption. In Austin, where summer temperatures routinely exceed 100°F, this is not a minor issue. It is a disqualifying one.
Estimate body fat percentage using the US Navy tape method and BMI-based formulas.
All data stays in your browser — we never see it.Calibration drift is the third problem. The ring's glucose algorithm uses machine learning trained on population data. It assumes a baseline glucose distribution that may not match individual physiology. My fasting glucose runs 82-88 mg/dL. The ring's algorithm seemed calibrated for a population average of 95-100. It consistently overestimated my fasting glucose by 8-12 mg/dL. That is not a small error. That is the difference between normal and pre-diabetic in some clinical guidelines. For a person with actual diabetes, the error could be even larger if their baseline differs from the training population.
I ran a Bland-Altman analysis on day 14. For the non-statisticians: this compares two measurement techniques. The mean difference between the ring and the Dexcom was 19.2 mg/dL. The limits of agreement — the range within which 95% of differences fall — were -36.8 to +75.2 mg/dL. That is a 112 mg/dL spread. For context, the FDA requires CGM devices to have mean absolute relative differences under 15% for most readings. The ring's performance was not in the same stadium as FDA standards. It was not even in the same city.
The Samsung watch was slightly better but still unacceptable. Mean difference: 15.7 mg/dL. Limits of agreement: -29.4 to +60.8. The fitness band was the worst: mean difference 26.3 mg/dL, limits of agreement stretching to +91.7 mg/dL. None of these devices would pass clinical validation. None should be used for medical decision-making. And yet all three are marketed with language like "glucose trend insights" and "metabolic awareness" that implies medical utility without explicitly claiming it. It is regulatory arbitrage, and it is dangerous.
Calculate your daily calorie needs with TDEE and activity level adjustments.
All data stays in your browser — we never see it.Here is the nuanced truth. Non-invasive glucose monitoring is not impossible. It is just not ready. The physics of measuring glucose through skin with light is genuinely hard. Glucose does not have a strong optical signature in the near-infrared range. It is swamped by hemoglobin, water, and melanin signals. Machine learning can extract weak signals from noisy data, but only with massive training datasets and frequent individual calibration against blood draws. The current devices have neither. They have small training sets, population-level assumptions, and no individual calibration.
I spoke with a biomedical engineer at UT Austin who works on optical biosensors. She told me — off the record, over tacos at Veracruz on East Cesar Chavez — that the field needs at least 3-5 more years of development before non-invasive glucose monitoring can approach CGM accuracy. The Circular Ring 2's planned glucose feature for late 2026? "Ambitious," she said, with the kind of diplomatic understatement that scientists use when they mean "probably not ready for prime time." She also mentioned that the WHOOP FDA warning letter in 2025 for cuffless blood pressure claims had sent a chill through the industry. Companies are now careful not to make explicit medical claims. But they are happy to imply them.
The FDA is aware of the problem. In 2025, the agency issued guidance clarifying that glucose monitoring devices making therapeutic claims require premarket approval. Devices making "wellness" claims do not. The distinction has created a gray market of devices that track "glucose trends" without claiming to measure glucose. The marketing language is carefully crafted: "awareness," "insights," "trends." Not "measurement." Not "monitoring." Not "medical." But consumers do not read the fine print. They see a glucose number on their watch and make decisions.
I tested this at a Gevity meetup. I showed ten people a screenshot of the ring's glucose reading next to the Dexcom reading. I asked: "Would you change your eating behavior based on the ring?" Eight of ten said yes. Six said they would reduce carbs. Two said they would skip the next meal. None asked whether the ring was accurate. The number looked authoritative. It had decimal points. It was on a screen. That was enough.
So what should you do with these devices? Use them for what they are: rough trend indicators, not medical tools. If your ring shows consistently elevated glucose trends over weeks, that might be a signal to talk to your doctor and get a real test. If it shows a single "high" reading after a salad, ignore it. The trend is the signal. The daily reading is noise. And the noise is currently very loud.
I will keep testing these devices as they improve. I will keep comparing them to the Dexcom. I will keep the spreadsheet open. Because the promise of needle-free glucose monitoring is worth pursuing. But the hype is ahead of the science. And in health tech, hype can hurt people. My ring said 168. My Dexcom said 103. I believed the Dexcom. I ate the sandwich. And I am writing this article instead of calling an ambulance. That is the difference between a toy and a tool. We are not there yet. But I will keep running the numbers until we are.
