Consumers Are Becoming the Front Door to Healthcare

A watch can now tell someone something no doctor has told them: they may have high blood pressure.

When Apple announced hypertension notifications in September 2025, it projected that the feature would alert more than a million people with undiagnosed hypertension in its first year. That was a forecast, not a count of confirmed diagnoses. The FDA cleared the feature that month.

The watch does not measure blood pressure. It analyzes optical heart-sensor data over 30-day periods for patterns associated with hypertension. An alert prompts the wearer to record cuff readings for seven days and share them with a clinician. (Apple's feature description)

The need is substantial. Nearly half of U.S. adults, approximately 120 million people, have high blood pressure. Only about one in four has it under control, and roughly one in six does not know they have it. Hypertension often produces no obvious symptoms. Waiting to feel ill can mean missing the opportunity to intervene. (CDC hypertension data)

An alert creates that opportunity. Everything that happens afterward determines whether it helps.

That is the consumer health opportunity: making prevention part of daily life, with a reliable route into care when it is needed.

Health happens between appointments

There are 8,760 hours in a year. Even someone who sees a doctor several times a year spends only a handful of them in an exam room.

Health is shaped in the rest: by sleep, movement, nutrition, medication, and decisions about when to ask for help. Yet an appointment captures only a brief snapshot. What happened beforehand is reconstructed from memory and a few measurements.

Access makes that problem harder. AMN Healthcare's 2025 survey found an average wait of 31 days for a new-patient appointment across six specialties in 15 major U.S. metropolitan areas. For family medicine, the average was 23.5 days. Those are surveyed markets, not a national average, but they illustrate the gap between having a question and getting an appointment. (AMN appointment survey)

Consumer health technology is beginning to fill some of that space. Wearables observe patterns between visits. Home tests bring measurement closer to daily life. AI offers an accessible place to ask a question at 2 a.m.

But filling the space between appointments only helps if what happens there can reach the people who provide care. That was the tension running through Rock Health Summit on October 14, 2026. On the consumer health panel, the AMA, Google, and WHOOP kept returning to it: consumers can see months of data on their phones, while clinicians often lack a routine way to receive, interpret, and act on it. The equity discussion raised the other half of the problem: the people who could benefit most may be the least likely to have the technology.

The future of healthcare will be consumer-driven because consumers increasingly start the journey. The question is how to make that journey useful, safe, and accessible.

What does active prevention require?

The first generation of consumer health devices counted steps, charted heart rate, and scored sleep. The next generation is beginning to influence what happens next.

Apple's sleep apnea notifications, authorized in 2024, use wrist movement to identify breathing disturbances associated with possible moderate to severe sleep apnea. Hypertension notifications extend the approach to another condition that can go unnoticed. Both prompt evaluation rather than establish a diagnosis. (Apple's sleep apnea validation)

At DeepStart, I think about active prevention as a five-stage loop:

1. Sense: Collect a reliable signal.
2. Understand: Explain what it may mean for this person.
3. Prevent: Recommend a proportionate action, from a behavior change to confirmatory testing.
4. Connect: Give the person a usable route to care.
5. Learn: Establish whether the action happened and whether it helped.

This is a company-building framework, not a claim that today's products complete every stage. Detecting an existing condition earlier creates an opportunity to prevent complications. It does not prove that a stroke, hospitalization, or other event was prevented.

The measurement that matters is whether the loop closes.

How are consumers using AI for health?

Consumers are not waiting for the loop to be built. They have already started.

Rock Health's 2025 Consumer Adoption Survey of 8,000 Census-matched U.S. adults found that 32% had used an AI chatbot for health information, double the previous year's 16%. Among those users, 64% engaged with AI about health at least weekly.

The survey was fielded in December 2025. Even then, consumers were using these tools for symptoms, treatment options, medications, and navigation. Eighty-one percent of health-AI users reported taking some follow-on action: 40% consulted a healthcare professional and 32% tried a new health behavior.

Eighteen percent reported adjusting medication.

The survey does not establish whether those changes were appropriate or supervised. It does establish that AI health information is influencing consequential decisions. It can lead people into formal care, but the data does not show that every interaction improves care or that no one delays seeking help. (Rock Health's health AI survey analysis)

As the distance between a question and an action shrinks, the quality of the information underneath matters more. I've seen why firsthand.

When my watch gets it wrong

I had the original Apple Watch Ultra and recently upgraded to the latest version. I rely on it every day.

I have also seen my original Ultra add sleep when I was fully awake. On drives of 30 minutes or more with my Tesla's driver-assistance system engaged, my recorded sleep time sometimes increased. I have seen the same thing while watching a movie.

I know I was awake. The record said otherwise.

That is my experience, not a controlled accuracy test. I do not know whether the error came from the algorithm, a setting, or another part of the data pipeline. But it illustrates the risk: an AI can build a persuasive explanation around an incorrect measurement.

An error in sleep logging does not invalidate a separately tested heart feature. Each feature needs evidence appropriate to its intended use.

Apple's hypertension validation makes the distinction concrete. In the primary analysis, the feature detected 41.2% of participants with hypertension and correctly avoided an alert in 92.3% of those without it. It missed more than half of cases in that analysis. (Apple's hypertension validation)

An alert is a reason to check. The absence of an alert is not an all-clear.

That difference has to be understandable to the person wearing the device. When it is, imperfect signals can still do real good, both in everyday behavior and in catching disease.

Can wearables change behavior and catch disease early?

The WHOOP discussion offered an interesting view of how personal feedback might influence behavior.

An observational study of 11,914 new members associated daily wear with approximately 20 minutes longer sleep per night and roughly 91 more weekly activity minutes than wearing the device fewer than five days per week, after adjustment for baseline measurements and other characteristics. The activity metric weighted higher-intensity minutes. Frequent wearers were already healthier on several baseline measures, so the study could not establish that the device caused the differences. (WHOOP wear-frequency study)

A separate study of 30,000 new members found that the modeled probability of drinking on a logged day declined from 23.0% to 17.2% over 72 weeks. Among consistent drink-count loggers, consumption declined by approximately 1.1 drinks per week. Alcohol use was self-reported, there was no control group, and the authors disclosed WHOOP employment and financial interests. (WHOOP alcohol study)

These are associations, not proof that a wearable caused behavior change. But the mechanism is worth taking seriously:

General advice says alcohol can impair sleep.

Personal data shows what it did to your sleep.

A good product creates the moment you choose differently tonight.

Other devices initiate clinical action. The Apple Heart Study enrolled more than 419,000 participants; 0.52% received an irregular-pulse notification. Among 450 participants who returned analyzable follow-up ECG patches, atrial fibrillation was detected in 34%. (Apple Heart Study)

The Fitbit Heart Study enrolled 455,699 people. Among 225 participants who had a further irregular-rhythm detection while wearing an ECG patch, 221 had atrial fibrillation confirmed concurrently. That 98.2% positive predictive value applies to this monitored group, not every alert across all users. Neither study established that screening improved long-term clinical outcomes. (Fitbit Heart Study)

The pathway is already visible: a consumer device notices a signal, the person responds, and clinical evaluation follows. The next challenge is making that pathway dependable, and that comes down to trust.

How do consumer health companies earn trust?

Trust develops when a product repeatedly does what it claims and its evidence survives scrutiny. The gap is clear in the numbers: a 2026 AMA survey of 2,222 physicians across six countries found that 97% review wearable data in some capacity, while no country exceeded 6% integration into clinical workflows. Interest is not the problem. Confidence and workflow are. (AMA wearable survey)

These are the seven moves I would prioritize to close that gap.

1. Validate the claim you actually make. FDA authorization supports a defined use. It does not validate every score or recommendation elsewhere in the app. Clinical claims need the appropriate regulatory pathway and evidence for the intended population. Companies should explain which features are medical, which are wellness estimates, and what each can support.

2. Publish the evidence and its limitations. Validation should explain who was studied, how performance was measured, what was missed, and where the findings may not generalize. Apple's hypertension results are useful precisely because they expose the trade-off. A credible company should be able to discuss false alerts and missed cases as clearly as its strongest result.

3. Design the handoff as part of the product. An alert that says "speak to your doctor" leaves the hardest work unfinished. Apple's seven-day cuff-reading recommendation gives the user a concrete next step. The broader product question is whether someone completes testing, reaches a clinician, and receives an appropriate response.

4. Match confidence to consequence. A suggestion to take a walk and advice about medication should not arrive with the same authority. Greater risk or uncertainty should trigger clearer limits, more questions, or human escalation. Communicate uncertainty before someone acts, not in a disclaimer afterward.

5. Let clinicians experience the product themselves. WHOOP's speaker described getting healthcare professionals onto the platform as users. A clinician who has seen their own data respond to sleep, exercise, and recovery develops practical familiarity before a patient brings it into the exam room. That supports informed recommendations; it does not replace clinical validation.

6. Make sharing useful to the receiving clinician. Google's speaker discussed SMART Health Links, a standard for sharing selected health information through a link or QR code, with options such as passcodes and time-limited access. Google publicly described plans for this sharing capability in May 2026. That sets a direction, not widespread clinical adoption. Someone still has to open the information, assess it, and decide what to do. (Google's sharing plans and the HL7 standard)

7. Earn trust in the community. Colby Takeda of Pear Suite described why "AI" can sound threatening to community health workers whose value rests on relationships. The conversation with investors may focus on technology; the conversation with community organizations needs to address practical outcomes, administrative relief, and getting workers paid. Product design and distribution must begin with how people actually work and whom they already trust.

Across all seven, responsible data use matters. Companies should explain who can access sensitive information, why it is needed, how sharing can be controlled, and who is accountable when something goes wrong.

Who owns the next step after an alert?

An alert creates work before it creates value.

Who reviews the signal? Who arranges confirmatory testing? Who notices when the patient never books? Who pays for the clinician's time?

These are product and business-model questions. A device can detect risk accurately and still fail if the consumer cannot access care or the receiving team cannot absorb the work. Prevention requires a distribution strategy, a payment model, and clear responsibility for follow-through.

The equity problem makes this more urgent. Rock Health found that device owners tended to be healthier and wealthier than non-owners. The population with the most room for improvement may be the one a direct-to-consumer subscription reaches least effectively. (Rock Health's wearable survey analysis)

At the summit, Commonwealth Fund president Joseph Betancourt argued for focusing AI on people with poorly controlled conditions such as diabetes and hypertension. For founders, that means treating primary care, community health workers, employers, pharmacies, and public programs as partners in access and delivery.

Affluent early adopters can validate demand. They cannot establish that a product will work for everyone else.

Five questions we ask founders at DeepStart

For founders building consumer and preventive health companies, our underwriting comes down to five questions:

1. Can the signal support the claim? Show accuracy, limitations, and performance in the intended population.
2. Does it produce an appropriate action? Measure what people do after the insight, including mistakes and drop-off.
3. Does the loop close? Establish who owns clinical follow-up, who pays, and when a human takes over.
4. Does the outcome improve? Distinguish engagement, behavior, and risk-factor changes from demonstrated reductions in clinical events.
5. Who can benefit in practice? Test beyond motivated early adopters and make access, privacy, and accountability part of the operating model.

What happens after the alert

Consumer devices are becoming a new entry point into healthcare. Their promise extends from everyday behavior to earlier recognition of conditions that might otherwise remain undetected.

The evidence is stronger for detecting certain signals than for proving complications were prevented. Closing that gap is the work ahead.

An alert on a wrist can start a medical conversation that might never otherwise happen. What matters is whether the person reaches care, whether the clinician trusts the information, and whether anyone measures the result.

Build for that entire journey, and you are building healthcare infrastructure. Stop at the notification, and you are building another dashboard.