Author(s): Todd Shryock
Fact checked by: Chris Mazzolini
A new global survey finds up to a third of patients never start prescribed treatment, and the gaps aren’t about access — they’re about a healthcare model still designed around the doctor’s office as the front door.
Key Takeaways
- Patients increasingly act on AI-derived health information before clinical contact, narrowing perceived value differentials with physicians and increasing presentation with self-diagnoses and named treatment preferences.
- Consistent cross-country attrition spans missed routine checks, delayed presentation, prolonged diagnostic timelines, unfilled prescriptions, and premature discontinuation, suggesting structural journey gaps rather than payer-specific anomalies.
- Discordance in perceived treatment-initiation barriers is pronounced, with U.S. clinicians over-weighting cost while patients prioritize side effects and burden, underscoring the need for direct barrier elicitation.
- Trust remains clinician-centered, yet patients resist overt AI involvement in clinical judgment and rarely surface online findings, creating parallel decision pathways that weaken shared understanding and adherence.
- Highest-leverage AI applications emphasize pre-/post-visit triage, referral handoffs, and continuity, with navigation platforms demonstrating shorter time-to-resolution and substantial ED diversion.
For decades, healthcare has operated on a simple assumption: patients wait until something feels wrong, they find their way to a doctor, and they follow the plan they’re given. That assumption no longer holds, according to the ZS Impact Institute’s 2026 Future of Health Report, a survey of nearly 9,000 healthcare consumers and more than 1,400 licensed physicians across the U.S., Germany and China, supplemented by qualitative interviews with 19 primary care physicians and 12 patients living with chronic conditions.
The report’s topline finding won’t surprise most physicians, even if the scale might: patients now trust generative AI nearly as much as they trust their own doctors, and they’re acting on that trust well before they ever book an appointment.
“Patients are changing faster than the system designed to serve them,” said Jon Roffman, ZS principal and lead contributor to the report. “AI has put medical knowledge directly in patients’ hands, but the healthcare system still assumes patients will come to it first. That model has changed. Patients are more informed and empowered. To keep them engaged, the system must reduce friction with its own set of AI and technology tools to deliver more connected, supportive care.”
A first opinion that no longer starts in the exam room
More than 40 million people use ChatGPT alone every day to search for health information, not counting other large language models or AI-powered search, according to the report. Consumers rate the helpfulness of AI-generated health information nearly on par with that of their physicians, the survey found: 89% of U.S. respondents who use AI for health information said they found it helpful, compared with 96% who said the same of their primary care physician — a narrower gap than most physicians would probably guess. In China, AI-generated information actually edged out search engines, with 94% of users rating it helpful, per the report.
The downstream effect shows up in the exam room itself. Sixty-eight percent of physicians surveyed said they’ve seen an increase in patients requesting a specific therapy by name — a sign, the report’s authors wrote, that patients are arriving with a working diagnosis and a preferred treatment already in mind, not just questions. “Patients are coming in with a diagnosis already,” one U.S. physician, identified in the report as Dr. Terry V., said.
The tension physicians are navigating is more complicated than simple AI adoption, though. A separate Annenberg Public Policy Center survey found that primary care physicians remain the most trusted source of health information overall, with 90% of respondents expressing confidence in their doctors — well ahead of government health agencies like the CDC or FDA, whose confidence levels sat closer to two-thirds. Yet that same survey found patients are far less comfortable with physicians relying on AI themselves: nearly half said they weren’t comfortable with a doctor using AI instead of independent clinical judgment, and only about a third said they were comfortable with AI playing a role in clinical decisions at all. Researchers there cautioned that AI tools aren’t necessarily current, and that “skepticism is warranted.” Few patients bring what they’ve found online back to the visit either — only about one in eight said they discuss it with their physician regularly, most often citing a lack of appointment time, according to that survey. Physicians retain the deeper trust, in other words, but patients want that trust demonstrated through judgment, not through visible AI use — and they’re increasingly out of the loop on the research shaping what they want from a visit in the first place.
Where the journey actually breaks
Patients are falling out of the system at every stage, and the pattern holds with remarkable consistency across three very different health systems, according to the report — a consistency the authors say points to something structural rather than a quirk of any one country’s insurance or delivery model.
At the prevention stage, 41% of U.S. patients, 48% of German patients and 38% of Chinese patients haven’t had a routine health check in three or more years, the survey found. At engagement, 45% to 68% of patients across the three countries said they delay seeing a doctor until symptoms become disruptive enough to force the issue. At diagnosis, more than a third of U.S. patients report waiting a year or longer for an answer, and 42% wait three or more months just to see a specialist. At treatment initiation, roughly a quarter to a third of patients never fill or start what’s been prescribed, according to the report. And even among those who do start, more than half of U.S. patients — 58% — say they stop treatment prematurely.
The disconnect between physicians and patients is starkest around diagnosis coordination, the data show. Fifty-eight percent of U.S. patients said their primary care physician helped coordinate their care across specialists; 83% of physicians said they did the same. That’s not necessarily anyone being wrong, the report notes — it may simply mean the coordination is happening in ways patients don’t recognize as coordination. But the practical result is the same: patients who don’t feel guided end up coordinating their own care, often badly. “I had to play coach and quarterback of my own care,” said one patient profiled in the report, who was eventually diagnosed with lupus nephritis after more than two years of fragmented visits.
Physicians and patients disagree on why treatment stalls
The most actionable finding for practicing physicians, according to the report, involves treatment initiation — the point where common assumptions about patient behavior turn out to be measurably wrong. Across all three countries, physicians and patients rank the barriers to starting therapy differently, sometimes dramatically so, the survey found.
In the U.S., physicians overwhelmingly cite cost as the top reason patients don’t start a prescribed treatment, according to the report. Patients rank cost third, behind side effects and treatment burden. In Germany, physicians and patients agree on the top two barriers — side effects and burden — but doctors still underestimate how much cost weighs on patients, the data show. In China, physicians underestimate efficacy concerns while overestimating burden and cost, according to the survey.
That mismatch doesn’t mean cost is a minor factor — separate data suggests the opposite trend line. A national survey commissioned by Imagine360 found that 38% of insured adults skipped or postponed necessary care or medication in the past year because of cost, up sharply from 27% in a comparable 2023 study — and of those who delayed care, 42% said their condition got worse as a result. High premiums and out-of-pocket maximums topped the list of reasons patients gave for care feeling unaffordable, even with employer-sponsored coverage. “Rising costs lead to Americans not accessing the care they need,” said Jeff Bak, CEO of Imagine360, in comments accompanying the survey. For practices, that trend shows up as missed appointments and worsening chronic conditions, along with a growing volume of billing disputes that fall to already-stretched front-office staff. Taken together with the ZS findings, the picture for physicians is less “cost doesn’t matter” and more “cost matters differently than assumed” — patients may not name it as their top barrier to starting a specific therapy, but it’s clearly reshaping how and whether they seek care in the first place. The practical takeaway is the same either way: Rather than defaulting to a co-pay card or assuming affordability is settled once insurance is in place, it may be worth asking patients directly what’s stopping them from starting or continuing treatment.
Closing the distance
Patients don’t experience individual touch points with the health system, the report argues — they experience the gaps between them, and right now no single stakeholder owns those gaps. Closing them doesn’t require a system overhaul so much as a shift in posture, the authors write: from assuming patients will show up and follow the path laid out for them, to actively drawing them in, carrying them forward and designing care that earns their continued participation.
For physicians and health systems, that means deploying AI not just for efficiency at the point of care but to extend triage and continuity beyond it — routing patients to the right level of care sooner, coordinating hand-offs so referrals don’t quietly stall, and giving patients trusted guidance before they turn to an unregulated chatbot instead. The report cites Baylor Scott & White Health as an example: Its first-contact navigation platform, which triages symptoms and routes patients algorithmically before they ever reach the emergency department, cut typical time-to-resolution from eight to 10 weeks down to about one week, while redirecting 80% of would-be ED visits to lower-acuity settings, according to the report.
The stakes extend well beyond any single practice’s patient satisfaction scores. ZS’s analysis estimates that shifting just 10% to 15% of patients from later-stage to earlier-stage diagnosis across four high-burden disease categories — metabolic disease, cardiovascular disease, cancer and neurological disorders — could unlock close to $500 billion in annual direct medical savings in the U.S. alone.
The bottom line for physicians
Patients aren’t rejecting medicine for AI, the report concludes. They’re responding rationally to a system that makes disengagement easier than participation — and every additional point of friction, whether it’s a hard-to-get appointment, an unexplained lab result or an unaffordable premium, pushes more of them toward workarounds that offer speed but sacrifice clinical oversight. Empowerment without support isn’t progress, the authors write. It’s an invitation to leave.
For physicians already stretched thin, the highest-leverage use of AI in a practice may not be diagnostic support during the visit, but closing the gaps before and after it, so patients aren’t left to coordinate their own care in the dark.
