The Kaiser Family Foundation published results from a nationally representative poll of 1,343 U.S. adults in April 2026, conducted in late February and early March. One in three American adults, 32 percent, say they used AI chatbots for health information in the past year, a share now equal to those who use social media for health advice, according to KFF. About 29 percent sought guidance on physical health. About 16 percent sought mental health guidance. Among those who turned to AI, 19 percent did so because they could not afford to see a healthcare professional, and 18 percent because they had no regular physician or could not get an appointment. AI chatbots are not supplementing healthcare for these patients. They are replacing access that healthcare failed to provide.

The Trust Paradox Inside the Data

The surface reading of the trust numbers is alarming. Among Americans who have not used AI for health information, roughly 80 percent say they trust it not at all or not very much, according to the KFF poll. That figure has been cited widely as evidence that patient distrust is a barrier to healthcare AI adoption.

The more important finding is what happens after first use. Among Americans who have actually used AI chatbots for health information, 69 percent trust them a great deal or a fair amount for physical health guidance, and 62 percent trust them for mental health information. The distrust is concentrated in non-users. Use changes the outcome substantially.

This pattern is familiar to anyone who tracked telemedicine adoption. Before broad adoption, patient skepticism about video appointments was high. After use, satisfaction rates climbed consistently. The barrier was not the quality of the technology. It was the first experience, and who controlled that experience.

The Access Gap Underneath the Trust Gap

The 19 percent who turned to AI because they could not afford care and the 18 percent who could not get an appointment represent a structural failure that long predates large language models.

According to additional health survey data, an estimated 14 million adults who used AI for health information subsequently skipped a doctor's visit. For some, the AI guidance was sufficient. For others, it may have delayed care they needed. Either way, the pattern reflects a patient population making rational decisions in a system where care access is rationed by cost and availability.

According to survey findings from Ohio State University's Wexner Medical Center, only 42 percent of Americans are open to AI being involved in their care as of January 2026, down from 52 percent in 2024. Trust in AI is declining even as use is rising. The divergence makes sense when you look at who controls the AI patients encounter. Patients using general-purpose AI chatbots for health advice are making a cost and access decision, not an AI preference decision. Their distrust of health systems deploying AI is a separate issue rooted in institutional credibility.

According to a 2026 Edelman Trust Barometer special report on health, 66 percent of patients report low trust in their health care system to use AI responsibly, and 58 percent doubt their health system would ensure an AI tool would not harm them. These numbers do not describe distrust in AI technology. They describe distrust in the institutions that would deploy it.

What Health IT Actually Needs to Build

The data points toward a specific and actionable platform gap. Patients who use AI for health information become more trusting of it. Patients distrust health systems deploying AI. The path forward is health-system-controlled patient-facing AI that is transparent about its nature, validated against clinical protocols, and connected to real care access.

Some health systems have moved in this direction. UnitedHealthcare's Avery, launched in March 2026, is available to 6.5 million commercial members for appointment booking, benefit navigation, and care coordination, with expansion to 20.5 million members planned by end of 2026. Tools that reduce appointment friction directly address the access problem driving patients toward unvalidated chatbots in the first place.

The governance requirements for patient-facing AI are specific. Patients need to know they are interacting with AI. They need a clear path to a human clinician within a small number of steps. The AI needs validated clinical guardrails, not just general language model reasoning. And the system needs to escalate proactively when a patient reports concerning symptoms rather than waiting for the patient to know what to ask.

Health systems that meet these requirements will build trust over time because patient experience with their AI will be positive and safe. Systems that deploy opaque AI in patient channels without disclosure will accelerate the institutional trust erosion the Ohio State data already shows.

What To Do About It

1. Prioritize patient-facing AI that reduces access friction. Scheduling automation, symptom triage, benefit navigation, and care coordination tools address the root cause driving patients to unvalidated chatbots. These are higher-priority deployments than additional back-office AI tools that deliver no patient-visible benefit. Close the access gap and you reduce the pressure that sends patients to uncontrolled AI sources.

2. Require transparent AI disclosure in every patient-facing deployment. Any system routing patient inquiries through AI should identify itself as AI, explain its limitations clearly, and provide a path to a human clinician within two steps. The Edelman data identifies transparency as the primary driver of healthcare AI trust. Absence of disclosure is not a minor gap. It is the fastest way to confirm patients' worst assumptions about how health systems use their data.

3. Measure patient-reported trust as a deployment metric. Add validated trust instruments to your AI deployment measurement framework. The Ohio State Wexner number, 42 percent of patients open to AI in care and declining, is your industry benchmark. Your health system should be tracking its own version of that figure annually and treating it with the same seriousness as patient satisfaction scores.

4. Build clinical escalation protocols before going live. Patient-facing AI that cannot escalate to a nurse or care navigator when a patient reports a concerning symptom is a patient safety liability. Define the clinical rules for when the AI transfers to a human. Test those protocols before go-live. Audit them quarterly. The protocol is not a feature. It is the governance control that separates responsible deployment from reckless deployment.

HRIM's Take

The KFF data does not describe a patient population that rejects AI. It describes a patient population that adopted AI because their health system left them without a better option. The trust problem is real, but its root cause is an access problem that healthcare created. Patient-facing AI built to clinical and transparency standards can simultaneously close the access gap and rebuild the institutional trust that has eroded significantly over the past five years. Health systems that treat this as a platform strategy will convert a public health challenge into a competitive advantage. Those that continue deploying AI only for physician documentation and revenue cycle optimization will watch their patients seek health guidance from sources outside their clinical network, with clinical outcomes they cannot observe or influence.