Your Wrist Is Becoming a Hormone Lab
Clair Health is trying to make one of the body’s most elusive systems continuously visible. If it works, the result could be much bigger than another wearable.
A blood test is a photograph.
It captures what is happening inside the body at one particular moment: a concentration of estrogen, a level of progesterone, an LH reading. Useful, sometimes essential, but still a snapshot of a biological system that never stops moving.
Hormones are closer to a film.
They rise and fall over hours, days and years. They interact with sleep, temperature, cardiovascular function, fertility, metabolism and ageing. Yet for something capable of influencing so much of human physiology, the way most people encounter hormonal data remains surprisingly fragmented: a laboratory appointment, an ovulation strip, a period-tracking app, a symptom remembered after the fact.
A San Francisco startup called Clair Health wants to replace some of those snapshots with something closer to a continuous feed.
Its proposition sounds almost implausibly simple: wear a bracelet, go about your life, and allow artificial intelligence to interpret the physiological traces hormones leave throughout your body.
No needle. No blood draw. No cup. No strip.
And, importantly, no hormone sensor actually detecting estrogen molecules through your wrist.
That last part is where the story becomes interesting.
Because Clair is not really building a miniature laboratory.
It is attempting something potentially more ambitious: teaching a machine to read the body well enough to infer what the laboratory might have found.
The hormones are invisible. Their fingerprints aren’t.
Clair’s central insight begins with a relatively ordinary piece of physiology.
Hormones do not fluctuate silently.
After ovulation, rising progesterone affects thermoregulation. Research has repeatedly observed increases in body temperature during the luteal phase. Heart rate, heart-rate variability and respiratory patterns also shift across the menstrual cycle. Wearable studies have already shown that measurements such as wrist temperature and heart rate contain enough information to help algorithms identify fertile windows and distinguish parts of the cycle.
Clair’s bet is that looking at many of these changes simultaneously can reveal substantially more.
The company’s current system uses 10 sensor modalities and extracts more than 130 physiological biomarkers spanning cardiovascular, thermoregulatory, autonomic, electrodermal, respiratory, body-composition, sleep and activity signals. Skin temperature, heart rate, HRV, breathing and electrodermal activity are among the signals it has publicly discussed. Clair says its hardware also incorporates proprietary sensing technologies, including a biomagnetic sensor not typically found in mainstream consumer wearables.
Imagine trying to recognise a song after hearing a single note.
That is roughly the problem with relying on one physiological marker. Temperature can rise because progesterone rose, but also because you are ill. Heart rate can increase because of hormonal changes, or because you drank coffee, exercised, slept badly or became stressed.
Clair’s approach is closer to listening to the entire orchestra.
A temperature shift accompanied by particular changes in heart rate, HRV, sleep, fluid balance and other signals creates a more distinctive pattern. Machine-learning models can then search those patterns for the physiological signature associated with a particular hormonal state.
The underlying idea has scientific precedent. Studies involving hundreds of menstrual cycles have demonstrated that machine-learning systems using wearable-derived temperature and cardiovascular data can predict fertile windows and cycle events, although performance tends to deteriorate in irregular cycles.
Clair wants to take that principle considerably further.
Not merely:
When is your period coming?
But:
What is your hormonal system doing right now?
The important word is “infer”
This distinction matters enormously.
Clair does not directly measure estrogen, progesterone, LH or FSH molecules from the wrist.
Instead, its models infer hormonal activity from the physiological consequences those hormones produce.
The company itself now makes this distinction explicitly. A direct measurement might involve analysing estradiol with mass spectrometry and reporting a concentration in picograms per millilitre. Clair observes the body’s response and attempts to reconstruct the hormonal state responsible for it.
That doesn’t automatically make the concept invalid.
Medicine already relies heavily on inference.
A pulse oximeter shines light through tissue and derives blood-oxygen saturation from optical absorption. A wearable does not “see” sleep; algorithms interpret movement, heart signals and other measurements to classify it. Even familiar fertility products frequently measure metabolites or downstream signals rather than continuously observing the parent hormone itself.
But there is an important difference between saying inference can work and saying this particular inference has been proven accurate enough to guide health decisions.
That gap is where Clair’s future will be decided.
And it is why the company’s most important achievement may ultimately have little to do with the appearance of its bracelet.
It will depend on the data.
So, does it actually work?
For now, the answer is promising, but incomplete.
Clair says its prototype work involved more than 40 women, 127 complete menstrual cycles and over 5,000 days of continuous physiological data, with wearable predictions compared against hormone measurements rather than simply assuming cycle phase from calendar dates.
In that internal work, the company reports 94.1% overall accuracy when classifying the four major phases of the menstrual cycle using wearable signals alone. It says its models detected LH surges with 87% sensitivity and 93% specificity, with timing accuracy of around one day.
There is an especially important footnote.
Performance was lower for irregular cycles: 84.3%, compared with 95.2% for regular cycles in the company’s dataset. Clair itself acknowledges that physiological stressors and less predictable hormonal patterns make this population harder to model.
That matters because people with irregular cycles may be among those most interested in better hormonal information in the first place.
Clair’s initial study also remains company-generated evidence. As of September 2026, the performance figures widely reported for Clair have not yet been established through a published, peer-reviewed independent validation of the commercial product.
And that is a critical distinction.
A compelling algorithm trained on dozens of people is not the same thing as a medical technology demonstrated across thousands of different bodies, ages, health conditions, medications and lifestyles.
Clair appears to understand that problem. The company has laid out a three-stage clinical programme and says its final stage involves an independent 150-woman validation through the Stanford Gladstone BeeHive Initiative, comparing the technology against hormone ground truth under institutional oversight, with results intended for peer-reviewed publication.
That study may prove more consequential than any funding announcement.
Because a beautiful wearable can be marketed.
A biological inference has to be validated.
Clair 1.0 is not the product you might think it is
There is another subtle but extremely important detail.
The first commercial Clair will not give users clinical numerical hormone concentrations.
Clair describes its initial product as a general-wellness device. It is intended to provide cycle-phase classification, hormonal trends and events such as LH surges and ovulation confirmation.
The more radical product is what the company calls Clair 2.0.
Clair says its models are being developed to estimate quantitative concentrations of estradiol, progesterone, LH and FSH. But rather than immediately displaying numbers such as estradiol in pg/mL or progesterone in ng/mL, it plans to pursue FDA clearance before making those estimates available to consumers. According to the company, the underlying hardware would remain the same, potentially allowing existing devices to gain the feature through a software update if clearance is obtained.
That distinction separates an intriguing wellness product from something approaching an entirely new diagnostic interface.
And regulation draws almost exactly the same boundary.
Current FDA guidance gives substantially more latitude to low-risk products making general-wellness claims than to products intended to diagnose, treat or provide clinically actionable medical information.
In other words:
“Your physiology appears consistent with your luteal phase” and “your progesterone is X ng/mL” may appear adjacent on an app screen.
Regulatorily and medically, they belong to very different worlds.
The startup arrived at exactly the right moment
Clair is young in almost every sense of the word.
Its CEO and co-founder, Jenny Duan, was 21 when she graduated from Stanford in June 2026, the same week the company closed an $11.6 million seed round led by Khosla Ventures, with investors including a16z Speedrun and 23andMe co-founder Anne Wojcicki. Her co-founder, Abhinav Agarwal, comes from computer science and wearable technology.
The origin story stretches further back.
Duan has described working with Rose Haven, an organisation supporting women and children experiencing homelessness and domestic violence in Oregon, and seeing women struggle to have health concerns taken seriously without quantitative evidence. At Stanford, her interests in technology, startups and women’s health eventually converged with Agarwal’s experience in sensing and wearables.
Their timing is significant because the wearable industry is undergoing a transition.
The first era was about counting.
Steps. Calories. Distance.
The second was about monitoring.
Heart rate. Sleep. Blood oxygen. Temperature. Recovery.
The emerging third era is about interpretation.
Why did your temperature change?
Why did your recovery score collapse?
Why did your sleep shift?
What invisible biological process could be causing the signals your sensors are already collecting?
Clair belongs firmly to this third generation.
And hormonal biology may be one of the most valuable, and difficult, problems it could choose.
For decades, hormonal variation was treated as a complication
There is historical irony here.
For much of modern biomedical research, female hormonal variability was not treated as an exciting source of information.
It was often treated as a problem.
In 1977, US FDA guidance recommended excluding women of childbearing potential from early-stage drug trials. Over subsequent decades, research policy changed substantially, including the 1993 NIH requirement regarding women’s inclusion in clinical research, but sex bias persisted across areas of biomedical science. Researchers have specifically challenged the old assumption that cyclical hormonal variation necessarily makes females inherently more variable research subjects.
Clair essentially reverses that logic.
It treats the variation as the signal.
That is a fascinating conceptual shift.
Most general-purpose wearables were originally designed around relatively stable baselines. If your resting heart rate moves away from normal, something may be happening. If HRV declines, perhaps recovery is poor.
But what if part of the variation is normal?
A 2026 systematic review of wearable-derived HRV found measurable differences across the menstrual cycle and reproductive life stages and concluded that hormonal context should be considered when interpreting wearable HRV data.
A change that looks like “bad recovery” to a generic fitness algorithm may carry different meaning when interpreted alongside reproductive physiology.
Clair’s thesis is therefore bigger than cycle tracking.
Perhaps female physiology shouldn’t simply be accommodated by existing wearables. Perhaps the model itself should begin with female physiology.
The bracelet is also becoming an AI model of you
Here the story begins to move beyond hardware.
Clair describes its underlying system using an unusually ambitious term: a “Female Biology World Model.”
The idea is not simply to classify today’s data point as follicular or luteal. The company says it is building a model of the dynamics of the hypothalamic-pituitary-ovarian axis: the feedback network connecting the brain and ovaries that governs reproductive hormonal signalling.
Population data teaches the system general patterns; repeated measurements from one wearer could then teach it what is normal for that particular person.
Think of the difference between a map and a GPS.
A cycle calendar gives you the map: this usually happens around day 14.
A personalised physiological model would attempt to locate you continuously: this appears to be what your body is doing today.
That distinction becomes especially interesting when biology refuses to follow the calendar.
Stress.
Illness.
Perimenopause.
Anovulatory cycles.
Irregular cycles.
Travel.
Changes in training.
Hormonal therapy.
The promise of continuous sensing is not simply that it produces more data, but that it could capture change when the expected pattern breaks.
Whether Clair’s models can reliably do that remains to be demonstrated at scale. But the broader technological direction is difficult to ignore.
Even your voice could become a biomarker
And Clair isn’t stopping at the wrist.
TechCrunch reported that the company has also developed AI intended to analyse voice-derived biomarkers, with Clair claiming it can gather information relevant to cycle phase after several minutes of conversation. The same voice interface is designed to let users describe symptoms naturally rather than selecting from rigid menus.
It sounds almost sci-fi.
But it hints at what healthcare’s next interface might look like.
Your watch observes your cardiovascular system.
Your temperature becomes a signal.
Your sleep becomes a signal.
Your movement becomes a signal.
Perhaps even your voice becomes a signal.
Instead of asking a person to continuously document their body, the environment around them begins collecting enough physiological context to construct a model automatically.
The result is not just a wearable.
It is something closer to a digital representation of the body that updates while the body changes.
And that creates a second question almost as important as accuracy.
Who gets to see it?
When your wearable knows when you ovulate, privacy stops being abstract
A device that understands someone’s steps is useful.
A device that may infer menstruation, ovulation, fertility patterns, reproductive ageing, sleep, cardiovascular behaviour and other physiological states holds a different category of information.
Clair has clearly recognised the sensitivity.
Its privacy policy describes a deliberately local-first architecture. The company says biometric readings, cycle information, symptom logs and model predictions are stored in an encrypted database on the user’s phone rather than automatically sent to its servers. Its hormone models are intended to run on-device. Optional cloud backups use encryption for which the user controls the key, and research sharing is opt-in rather than the default. Clair also states that it does not sell personal health data.
It even proposes federated learning as an optional research mode: models can learn from data on a person’s phone while only model updates, rather than the raw health records themselves, leave the device.
That architecture deserves attention because reproductive data poses unusually intimate privacy questions.
If continuous hormone inference becomes mainstream, consumers will increasingly have to judge health products not only by what they can know, but by where that knowledge goes.
The more powerful personal health AI becomes, the less privacy can be treated as a paragraph hidden at the bottom of a terms-and-conditions page.
It becomes part of the product.
There is also a risk in knowing too much
More information is not automatically the same as better health.
A wearable capable of producing constant biological interpretation could create clarity.
It could also create anxiety.
We have already learned to wake up and ask a ring whether we slept well before deciding how tired we feel. We check recovery scores before exercising and heart-rate graphs after walking up stairs.
Hormones could become the next frontier of self-quantification.
Low energy? Check the hormones.
Bad workout? Check the hormones.
Irritable? Check the hormones.
Didn’t sleep? Check the hormones.
At its best, this could replace vague assumptions with useful context.
At its worst, it could encourage people to reduce complex experiences to algorithmic explanations that have not been clinically established.
That makes interface design almost as important as sensor accuracy. Clair will have to decide not merely what its models can infer, but what information users should see, how confidently it should be presented and when a pattern requires a clinician rather than another graph.
The company explicitly says Clair is not currently intended to diagnose conditions such as PCOS or endometriosis.
That line needs to remain visible.
Especially if the technology becomes very good.
Investors are betting that women want to know
The market appears ready to test the idea.
When Clair announced its seed financing in June, Fortune reported a waitlist of approximately 25,000 people. The company’s current website says its first 5,000 Founding Member places have sold out, with founding-member devices now scheduled to ship in December 2026. Earlier reporting put eventual retail pricing at around $369, with a subscription component for the app.
Clair is also entering a market that established wearable companies have begun exploring aggressively. Oura, Whoop and fertility platforms already use combinations of temperature, cardiovascular data, cycle information and algorithms to provide women’s-health insights. Research using Oura and other commercial wearables has demonstrated that menstrual-cycle changes can be visible in temperature and cardiovascular data.
So the question isn’t whether wearables can see anything about the menstrual cycle.
They can.
The interesting question is how far that visibility can go.
Can an algorithm move from recognising broad cycle patterns to estimating individual hormonal dynamics?
Can those estimates survive illness, travel, alcohol, poor sleep, irregular cycles and medication?
Can they generalise across age, skin tone, body composition and reproductive life stage?
Can they become clinically useful?
And ultimately:
Can a wristband produce information a doctor would trust?
The real product isn’t the bracelet
It is tempting to see Clair as a competitor to Oura or Whoop.
That may undersell what the company is attempting.
The more interesting comparison may eventually be with the continuous glucose monitor.
CGMs changed the relationship people with diabetes have with glucose partly because they replaced isolated measurements with a curve. Suddenly, it became possible to see not merely a number but a trajectory: what happened after a meal, overnight, during exercise.
Clair wants to introduce a similar temporal idea to hormones.
The analogy is imperfect. Importantly, a CGM chemically senses glucose in interstitial fluid, whereas Clair is attempting a more indirect physiological inference. The validation burden is therefore very different. But the conceptual transformation is similar:
from snapshots to streams.
And streams reveal things snapshots cannot.
A single progesterone result tells you what happened at one point.
A continuous model could potentially reveal the shape of change.
The timing.
The transition.
The deviation from your normal pattern.
The relationship with sleep, symptoms, training and recovery.
That possibility is what makes Clair worth watching even before its most ambitious claims have been independently validated.
The startup is proposing that some of the information medicine traditionally searches for inside a vial may already be echoing throughout the body: in temperature, heart rhythms, vascular changes, sleep, respiration and countless subtle physiological responses.
The data, in other words, may already exist.
The technological challenge is learning how to read it.
From tracking the body to understanding it
For the past decade, wearable technology has become extraordinarily good at observing us.
It knows when we move.
It knows when we sleep.
It knows how quickly our hearts beat.
Now the industry is confronting a far harder problem.
What does all of that information actually mean?
Clair represents one possible answer.
Not another sensor reporting another number, but a layer of intelligence attempting to reconstruct the invisible biological system beneath those numbers.
That ambition should be treated with both excitement and scrutiny. Clair’s early results are still early results. Independent peer-reviewed validation of its production technology will matter. FDA clearance will matter if it moves into numerical hormone reporting. Performance outside controlled development datasets will matter most of all.
But the direction is compelling.
Because the future of wearable technology may not be a wrist covered in ever more sensors.
It may be something subtler.
A device that disappears into everyday life while constructing an increasingly detailed understanding of the person wearing it.
One day, the remarkable thing about checking your hormones may not be that it requires a laboratory.
It may be that you didn’t have to do anything at all.
