Patients say AI can help, but only if doctors keep a hand on the steering wheel, Salesforce research finds

Salesforce on Wednesday pointed to a growing gap between what patients want from healthcare AI and how it is often deployed. In new consumer research on connected health, the company found that patients are largely willing to use “agentic” AI, but they are not bargaining on removing humans from care delivery. Instead, they want clear paths to a real person, stronger privacy assurances, and enough transparency to understand how AI decisions are made.

The report is based on a survey of more than 3,200 health consumers across eight countries, conducted March 24 to April 10, 2026. Salesforce said that patients’ day-to-day experiences with the healthcare system are driving the issue, with widespread complaints about confusing digital processes, slow scheduling, and prescription delays.

According to the findings, 46% of patients delay care because the digital steps are too confusing, and 58% delay or skip necessary care because scheduling is difficult. Salesforce also reported that nearly half of patients, 49%, hang up after 10 minutes on hold with a doctor’s office to seek care elsewhere or avoid it. For ongoing care needs, the research found 66% of respondents have run out of medication while waiting for a prescription refill to be approved.

The survey data also highlights expectations around continuity of care between settings. Salesforce said 90% of patients want their primary doctor automatically notified after an emergency room visit, a level of coordination it says many systems do not reliably deliver. Against that backdrop, patients appear open to AI taking on parts of the workload, but only when the handoff to clinicians and staff is dependable.

When patients were asked about AI in healthcare, Salesforce said the top concerns centered on accuracy and privacy. Thirty-six percent cited accuracy of diagnosis or treatment as their primary worry, while 30% pointed to privacy and security of health data. Even so, patient comfort with AI is rising: Salesforce reported 61% of global patients say they are comfortable using agentic AI in healthcare contexts, and 64% would share their full medical history with AI for a faster diagnosis.

The company’s research also ties trust to where the AI lives and who stands behind it. Salesforce said patients are three times more likely to trust an AI agent integrated directly into their doctor’s secure portal than an AI option offered through a public chatbot or general health site. In other words, patients appear to view healthcare AI less as a standalone tool and more as an extension of provider accountability.

The human escalation requirement is central across multiple use cases. Salesforce reported that 89% of patients say a clear “escalate to human” option is essential for trusting AI-based administrative support, and 90% expect the same for AI medical support. Salesforce also said 91% want the right to opt out of AI-driven clinical recommendations entirely, indicating that convenience is not a substitute for control when clinical decisions are involved.

In the provider community, Salesforce cited UChicago Medicine Chief of AI-related communications, Tyler Bauer, who said trust and transparency are foundational to patient experience and that patients need to know information is protected with escalation to a live agent always available. Salesforce also quoted its own executives, including an SVP and GM for Agentforce Health, Amit Khanna, who described “transparent AI” as something built on a unified, governed data foundation so clinicians can see how decisions were made and patients can see sources and reasoning.

For healthcare organizations, Salesforce framed the takeaway as an implementation challenge: patients are ready for agentic AI, but only when it is deployed with governed data, auditability, and a reliable escalation path to clinicians. The company said the research also suggests “loyalty implications,” with 44% of patients saying a 24/7 agentic assistant would make them more likely to stay within a provider’s network for follow-up care, and 65% of patients with long-term conditions saying a 24/7 digital helper would make managing health significantly easier. Still, Salesforce emphasized that none of that convenience erases the expectation of human backup.

The main limitation is also straightforward: the findings are based on stated preferences and concerns, not on clinical outcomes or demonstrated accuracy. Salesforce does not break out how providers would quantify AI accuracy, how frequently human escalation occurs in real workflows, or how different regulatory or system environments would affect trust. Patients’ willingness to share data, for example, may depend on specific safeguards and consent language that are not detailed in the report summary.

Looking ahead, the healthcare AI market will likely be judged as much on operational design as on model capability. What to watch next is how providers define and measure “human backup,” what they disclose about data governance and explainability, and whether AI assistants are integrated into secure provider portals in ways patients clearly associate with accountability. If Salesforce’s survey announcement holds, trust will depend less on whether AI can respond, and more on whether it can be audited, bounded, and corrected in time.