91% of Patients Require Consent Before AI Touches Their Diagnosis — Healthcare Systems Aren't Built for That

By Hector Herrera · November 4, 2026

Nine out of ten patients want explicit permission before AI is used in their diagnosis. The healthcare systems deploying AI, meanwhile, are mostly focused on somewhere else entirely: processing claims. A new survey has exposed a fundamental mismatch between what patients fear about AI in healthcare and where providers are actually spending their AI budgets — and it has significant implications for how trust in medical AI gets built or broken.

The findings come from a Healthcare IT Today survey published November 1, which queried both patients and healthcare providers about AI adoption, consent, and trust. The gap the survey captures is not incidental — it reflects a structural disconnect in how the industry is sequencing its AI investments.

What the Survey Found

The headline numbers:

• 91% of patients say they require explicit permission before AI is used in their diagnosis or clinical care
• 64% of healthcare providers say they are actively deploying AI for claims and denials management — administrative and billing workflows
• A much smaller share of providers report deploying AI directly in diagnostic or clinical decision-support roles, where patient anxiety is highest

The pattern reveals two parallel realities. Patients are most worried about AI in the examination room — in the moment a doctor reaches a conclusion about their health. Providers are most actively deploying AI in the billing office, where AI reviews claims for coding errors, flags likely denials, and automates appeals. Neither group is wrong about what AI can do in their domain. But the investment and the anxiety are aimed at different targets.

Why the Gap Matters

The consent expectation that 91% of patients hold is not currently built into most healthcare AI workflows. When a radiologist uses an AI-assisted imaging platform to detect anomalies, there is typically no patient-facing disclosure that AI contributed to that read. When a clinical decision support system flags drug interactions during prescribing, the patient at the bedside generally does not know. Most healthcare AI operates in the background, embedded in systems that clinicians use — and there is no standardized mechanism for surfacing that use to patients or obtaining their agreement.

That gap is about to become a regulatory problem. Several states — including California's clinical AI audit bills signed in October — are moving toward mandatory disclosure requirements for AI in clinical settings. The federal patient AI act currently moving through Senate committees would require consent for AI in diagnostic workflows at institutions receiving Medicare and Medicaid funding. The survey's 91% figure gives that legislation a data anchor: patients are not ambivalent, they are nearly unanimous.

Where Providers Are Actually Deploying AI

The 64% of providers deploying AI for claims and denials management reflects a rational business calculation. Claims processing is a high-volume, rule-bound workflow where AI can demonstrate clear ROI quickly: faster claim submissions, fewer coding errors, more successful first-pass approvals. The administrative burden of denials management — which consumes significant staff time at most health systems — is a natural target for automation.

This is not AI touching a patient. It is AI handling a form. But the revenue logic of healthcare AI means the fastest deployments are administrative, while the clinical deployments that patients care most about — diagnostics, treatment recommendations, risk scoring — move more slowly because the regulatory path is longer and the liability exposure is higher.

The result is a trust debt accumulating in advance of the deployments that will matter most. If patients first encounter healthcare AI through news stories about diagnostic errors or through opaque disclosures when they finally do receive AI-assisted care, the absence of a consent culture will be visible and damaging.

What Providers Can Do Now

The consent infrastructure needed to honor the 91%'s expectations does not yet exist at scale, but several health systems have begun building it:

• Pre-visit disclosure in scheduling confirmations that lists which AI tools may be used in the patient's care
• EHR-integrated consent workflows that capture patient preference for AI participation in clinical decisions at registration
• Clinician-facing AI transparency — requiring that AI-generated recommendations be labeled as such when presented to care teams, so the loop closes if patients ask

None of these are technically complex. They are workflow and policy choices. The survey data makes clear that the demand for them is essentially universal among patients. The question is whether health systems build consent infrastructure voluntarily, or wait until legislation requires it.

What to Watch

The Senate's patient AI act is in committee markup; a floor vote is expected in Q1 2027. If it passes with the current consent-for-diagnostics language, every major health system will need to build the disclosure workflows they currently lack. Watch also for the first significant malpractice case where AI's role in a diagnosis was undisclosed — that case, when it arrives, will accelerate institutional adoption of consent practices faster than any survey.

Source: Healthcare IT Today — Bonus Features November 1, 2026