“Is AI actually changing healthcare for Americans, or is it another hyped-up fad?”
Recently, my colleagues at iF and I set out to answer that question. To do so, we evaluated over 100 products, interviewed 30 companies and experts, and actually tested around 40 of these tools ourselves.
What did that look like?
- Wearing four different health trackers capturing hundreds of biomarkers, including a headband for brainwaves, a smartring, a smartwatch, and a chest sensor
- Giving around 25 vials of blood for three different companies
- Chatting with six different AI products with access to 49 pages of my aggregated personal health chart
I began this experiment expecting the ultimate answer to trend toward AI as a ‘fad.’ But I came out convinced that AI has already changed American healthcare indelibly, and will continue to do so. Why? Because testing these products changed me.
What’s new:
- AI is routing around the old care cycle…for now: The American healthcare system has long been a predictable loop, and that loop is a dead end for millions of people. None of the products we tested gave us access to some new, alternative system. Most aim to help people understand the existing health system, navigate it faster, or work around its many barriers. Incumbents like Epic (MyChart) are rolling out AI carefully within the system, while newer players like Doctronic, Superpower, and Curai are moving aggressively forward, taking advantage of regulatory sandboxes or legal grey areas. Meanwhile, California recently barred AI from independently performing licensed health tasks. Sandboxes in some states and restrictions in others further bifurcate an already confusing national health landscape.
- The frontier AI models are becoming the new front door: Hundreds of millions of people globally ask ChatGPT and Claude health and wellness questions every week. In my experience, I didn’t find these tools necessarily “better” than an alternative, partly because alternatives don’t exist. These tools are largely free (for now, and some ChaptGPT tiers already feature ads), always on, and they never stop responding. Increasingly, they can formally and securely connect to health records, wearables, and other players like Apple Health. When I plugged Claude into my records, I saw a longitudinal map of my health aggregated across every disconnected provider I’d visited in the past five years. If you’re struggling with a complex, exhausting system, why start anywhere else?
- Wearables were already booming, but AI makes their data meaningful: Step counters and sleep trackers have been around for years, generating “infotainment,” that few users really acted on. But companies integrating AI into products like Whoop and Oura are marketing on clinically relevant data, while their accuracy remains a big question. For example, the three sleep trackers I wore each night gave me three very different verdicts on the same nights of sleep repeatedly. Accuracy disparity may not matter a ton for sleep tracking or steps, but what about for heart, lung, or stroke detection?
What’s dangerous:
- Very different AI products are converging: The big AI companies position themselves as distinct products with unique features and values. I wanted to test that difference in healthcare. I fed four popular models (across paid and free subscription tiers) varying levels of my health data (some with the 49-page integrated report, some with no reports or test results) and asked the same six questions. The test was engineered to produce divergence. Instead, I saw convergence: across hundreds of input variations and hundreds of dollars in subscription differences, all the responses were 80% alike, landing on the same clinical picture and, notably, making big errors. So what is the consumer value when all the models might already be converging, and doing so while making big mistakes?
- Solving the system can mean conflicts of interest: Customizable diagnostics and insurance navigation were two areas where we saw growing AI product use. In diagnostics, many companies let consumers pay to choose their own lab tests and then use an AI chat to analyze the results. Some companies go even further and also have a marketplace to buy products tailored to consumer results. That convenience adds huge value to consumers used to going through all the hoops to get a prescription. However, we noticed some companies use proprietary reference ranges to recommend products. For example, one platform flagged my cholesterol as “high”, then offered me a $200/month supplement to fix it. But the company’s “high cholesterol” number they use to sell me products was noticeably stricter than the threshold my doctor uses. In insurance, consumers are using general AI tools and insurance-specific ones to appeal coverage decisions. Meanwhile, there are reports of insurers turning to the same AI tools to handle surging AI-generated appeals. If AI companies in the middle can profit from both sides of an escalating fight, is there a resolution incentive?
- These changes further stratify an already inequitable health system: Every product we tried is available directly to consumers. But at their price points, very few products are accessible to the millions who could benefit most. The stack of products I tested and would theoretically opt to keep would cost an extra $3,000 a year on top of my existing healthcare costs. Additionally, early user bases for wearables, longevity products, and AI tools skew toward wealthier, tech-savvy users, meaning these products could optimize for the populations that validated them. High costs and varying product performance across populations could widen existing care gaps for underserved communities.
What’s next:
- Consolidation: We discovered that most so-called “proprietary” AI products we tested are actually “wrappers,” built on top of OpenAI’s and Anthropic’s models. Ultimately, will the proliferation of specialty tools built on a few giants give way to consolidation, especially if the “giants” learn to do the same tasks as the specialty tools? Beyond the economic shockwaves, what happens to consumer data when specialty companies merge, collapse, or grow?
- Opting out will not be an option: Even if you don’t use AI tools personally, big companies keep agreeing to new deals with employers, schools, and the government, while also scraping the web for more training data. At the same time, AI is multiplying, and systemic healthcare challenges are metastasizing and worsening. Individually opting out of using AI is possible now, but will macro-level pull and push factors force people to adopt AI? It might become like the U.S. food system, where locally sourced organic food is an expensive luxury.
- Trust as a final frontier: While general AI tools seem limitless in application, we also found specialty products–with and without AI–thriving and growing. Meaning, consumers still opt for specific products beyond using big AI tools for everything. The pattern linking these specialty tools was often a company founder who personally lived the problem they’re trying to solve, and that problem was usually something the healthcare system has struggled with. For example, we saw this trend clearly in women’s health, where system failure is sharpest. Products are emerging that aim to help women access care, defend their rights, and reshape their experience, while also building datasets to improve understanding of perimenopause, IVF, endometriosis, or currently unknown but very real struggles. For the people with those struggles who have been ignored or misunderstood, can AI-fueled technologies deliver results with humanized empathy?
What’s left:
Personally, I’ve struggled with a confusing array of health issues for a decade that the formal system hasn’t been able to resolve. I didn’t expect AI to deliver answers the formal system hadn’t already thought of.
But I was wrong: after three months, AI-powered tools led me to an answer and a treatment. I still had to see doctors and work with insurance, and I couldn’t have afforded these tools without this project. Nevertheless, AI strapped a jet engine to a journey I’d been stuck in for too long, and I’m confident I wouldn’t have received the care I needed without using AI. It wasn’t a new system, but an accelerant for the slow system we all know.
These products did give me plenty of wrong turns that could have been dangerous. However, I’d take that risk again because navigating the U.S. healthcare system’s dead ends alone still feels more daunting. That desperation could become a dangerous push factor for millions when faced with the alternative: no care, or care through AI. And those people may not have the luxury I had to double-check dangerous advice.
For me, AI accelerated my care through a system that already serves me. But can AI reach people the system fails or leaves behind? And can it do so affordably and accurately?
During this project, I realized AI isn’t maximizing accuracy and adaptation. The real value is immediacy–answers to any question, any time. The products we studied mostly deliver on that promise. Right now, AI can deliver near-instantaneous expertise on pretty much anything, but not for everyone, not regardless of income, background, or struggle, and not consistently. So how can AI deliver quality and equality along with immediacy? And, potentially most powerfully, can AI grow beyond an advisor into an actor that can relieve the workload, stress, and desperation of millions of people stuck on their health journey?
These aren’t questions desperate patients should have to answer. The real leverage comes from the source: companies building these tools, health systems adopting them, regulators approving them, and funders fueling them. Those groups are tackling these questions now. Everyone deserves clarity and transparency about AI tools entrusted with sensitive information. That clarity is essential to ensuring the future of healthcare doesn’t inherit the past's failures.