Is ChatGPT Now The World's Largest Health App? | OpenAI VP of Health, Nate Gross
Forty million people use ChatGPT for health-related questions every day, making it one of the most widely used tools for health information in the world. So what is their team doing to maximize impact and minimize harm? For one, they've brought in hundreds of physicians globally to continuously review outputs and shape how the models respond across different scenarios, literacy levels, and edge cases. Second, they've hired my Rock Health co-founder, Dr. Nate Gross, as their VP of Health.
In this full-circle episode, I sit down with Nate, who also co-founded Doximity (DOCS) and knows a thing or two about building in digital health. We discuss the astonishing speed of AI progress, how models are trained for safety and accuracy, and what this technological evolution means for every part of the healthcare system.
Key topics:
How ChatGPT is becoming a 24/7 front door for health questions, and whether it is replacing Dr. Google or starting to compete with the healthcare system itself
How OpenAI is trying to reduce hallucinations, avoid sycophantic behavior, and build guardrails for sensitive use cases like mental health
OpenAI’s goals to “raise the floor, sweep the floor, and raise the ceiling” with new product launches like ChatGPT for Clinicians and GPT-Rosalind
How Nate thinks about the AI race and what winning in healthcare actually requires
Where startups should focus their efforts now that specialized products are launching for clinicians and life sciences
The single hardest problem in healthcare that AI, according to Nate, probably won't fix anytime soon
About our guest:
Dr. Nate Gross is the VP of Health at OpenAI. He previously co-founded Doximity and Rock Health. He graduated from the Emory University School of Medicine with an MD, Harvard Business School with an MBA, and Claremont McKenna College with a BA in Government. He serves as affiliated faculty for the Clinical Informatics Fellowship at Stanford.
—
Listen:
Full Transcript (AI edited)
Halle: Nate, welcome to the show.
For listeners, a quick background: Nate and I met at HBS many years ago. You were also working on your MD at the time. We worked on Rock Health together as an independent study with Professor Bob Higgins, while you were starting Doximity with Jeff Tangney. After school, you spent 15 years as co-founder and Chief Strategy Officer of Doximity and helped take it public. You’ve also stayed involved with Rock Health as a board member.
Then last summer you called me—I remember I was walking around a lake in the Berkshires—and told me you were leaving to join OpenAI. For those who know you, it wasn’t a surprise. You’ve always been on the forefront of innovation. But most people after a career like yours either retire or go into VC.
So tell us about your new role at OpenAI and how you ended up there.
Nate Gross: Thanks, Halle. It’s great to be here.
If I zoom out, a lot of my career has really been the same question from different angles: how do we make healthcare work better for the people inside it?
Back when we were in school, healthcare was still largely paper-based. Meanwhile, the consumer internet was exploding—iPhones, apps, social networks. But healthcare didn’t fully take advantage of that shift.
Doximity gave me a practical view into how clinicians actually use tools—what works, what creates friction. And Rock Health gave a broader view across patients, systems, and entrepreneurs.
That’s what made the OpenAI mission resonate. The idea of making sure AI benefits everyone, especially in healthcare, where the impact becomes very tangible.
Halle: When I think about the major AI players—OpenAI, Anthropic, Google, and we’ll give a shout-out to your brother at Microsoft—I picture a horse race. One company takes the lead, then another leapfrogs with a new release.
What ultimately determines who wins, if anyone does?
Nate Gross: I’d reframe that a bit. I don’t think this is a zero-sum race.
Healthcare is too large and too fragmented for one company to dominate. Success will come from helping different parts of the system succeed—patients, clinicians, health systems, and scientists.
Each of those groups has different needs:
Patients need better understanding and guidance
Clinicians need more capacity
Systems need less friction
Scientists need to move faster
The companies that matter will:
Solve real workflows
Earn trust
Build tools that are broad enough to have real impact
Halle: But OpenAI isn’t just a platform—you’re also building products directly. How do you think about that?
Nate Gross: We’ve been fortunate to see massive adoption. Around 900 million people use ChatGPT weekly, and about 230 million are asking health-related questions.
At that scale, you have a responsibility to build directly for healthcare use cases. People are going to use it anyway—both patients and clinicians—so we need to make sure it works well and safely.
At the same time, we’re empowering an ecosystem of partners building on top of the platform.
Halle: Let’s talk about how each part of healthcare is using AI—patients, providers, and pharma—and also how you’re thinking about guardrails. Starting with patients:
You now have 230 million people asking health questions every week. What are you doing to ensure those answers are safe and accurate?
Nate Gross: First, we have to meet users where they are.
People are using ChatGPT for:
Understanding symptoms
Interpreting lab results
Preparing for visits
Following up after care
So a big focus is:
Triage and translation → helping people understand what something means
Preparation for limited clinician time → making visits more effective
Escalation for high-risk situations → ensuring urgent cases are handled correctly
We’re also working on improving adherence—helping people actually follow through on care plans by making them more understandable and personalized.
Halle: One stat that stood out to me is that about 70% of these conversations happen outside normal clinic hours. And the product is free or low-cost.
So is this competing with Dr. Google, or with the healthcare system itself?
Nate Gross: It’s definitely evolving from search.
Unlike search, the model understands context—age, history, symptoms—and can adapt responses. It’s less about replacing care and more about helping people prepare for and navigate care.
Halle: Let’s talk about the biggest critiques: that AI is sycophantic and that it hallucinates.
Nate Gross: A big part of addressing that is redefining how we measure quality.
Traditional benchmarks don’t reflect real healthcare conversations. So we built HealthBench, which evaluates models on realistic scenarios.
Instead of just asking “did it get the diagnosis right,” we measure:
Did it ask for missing context?
Did it express uncertainty appropriately?
Did it escalate when needed?
Did it adapt to the user’s level of understanding?
We’ve worked with hundreds of physicians to define what “good” looks like in real-world settings.
Halle: How has the experience changed compared to earlier models?
Nate Gross: Earlier models were more generic, less context-aware, and less cautious.
Now they:
Ask better follow-up questions
Express uncertainty appropriately
Integrate real-world constraints
Use tools like retrieval for up-to-date information
Halle: Let’s talk about mental health.
Nate Gross: This is an area where we need to be especially careful.
The goal is support, not substitution.
People come to ChatGPT in vulnerable moments—from everyday stress to serious emotional challenges. So we focus on:
Detecting distress
De-escalating appropriately
Encouraging real-world support
Avoiding unhealthy dependence
We’ve built advisory boards and safety mechanisms to guide this.
Halle: What about loneliness and companionship?
Nate Gross: AI can help reduce friction in human relationships—for example, helping people reconnect or communicate more easily.
But it shouldn’t replace real human connection.
Halle: Let’s shift to providers.
You’re working with major health systems and just launched ChatGPT for Clinicians. What does that unlock?
Nate Gross: Clinicians were already using ChatGPT informally. So we built secure, integrated tools that allow them to use it safely.
Key capabilities include:
Secure access and data controls
Integration with clinical systems
Evidence-based retrieval
Repeatable workflows and “skills”
It supports both clinical and non-clinical tasks—from documentation to decision support.
Halle: What does this mean for startups?
Nate Gross: There’s still a lot of opportunity.
We work closely with startups, many of whom build on our models. The biggest shift is that generic “chat” products are harder to differentiate. Value comes from:
Deep integrations
Workflow-specific solutions
Trust and domain expertise
Halle: Finally, let’s talk about pharma and research.
Nate Gross: We think about three categories:
Raise the floor → expand access
Sweep the floor → eliminate repetitive work
Raise the ceiling → accelerate discovery
GPT-Rosalind fits into that last category—helping scientists reason, analyze, and move faster.
Drug discovery takes over a decade. AI won’t solve everything, but it can accelerate many steps along the way.
Halle: To close—what does healthcare look like in 15 years?
Nate Gross: One big shift will be generational. People growing up with AI will interact with it very differently.
We’ll also see more automated loops—systems that can run and improve processes over time.
But some things won’t change easily—like incentives and accountability. Those still require human decisions and policy changes.
Halle: Nate, thank you so much for your time.
Nate Gross: Thanks, Halle.