Consent: Patients want plain-language written materials they can review on their own time, paired with in-person conversations where they can ask questions — and they preferred a comprehensive consent form over a brief summary. Meaningful consent requires that patients can genuinely decline AI-assisted care; when that’s not possible, consent feels like a formality.
Accountability: Patients need to know that a human provider remains reachable and responsible for medical decisions, and worry that over-reliance on AI could lead to cognitive deskilling and diffused responsibility. When accountability is clear — including plain-language documentation of how AI is used, who oversees it, and how decisions can be contested — patients feel safer trying new approaches to care.
Trust: Patients trust their providers’ judgment more than they trust “AI,” and any introduction of AI should strengthen the patient-provider relationship rather than replace it. For lower-stakes decisions patients are comfortable with AI autonomy, but for higher-stakes ones they expect their provider to have the final say.
Autonomy: Patients report feeling powerless in healthcare systems and want more autonomy, particularly in decisions with direct personal impact like treatment choices. Tools that increase transparency — like patient portals — have made patients feel more empowered, and when autonomy is absent, patients turn to outside resources like consumer AI chatbots instead.
Fairness: Patients from historically marginalized communities expect AI to follow familiar patterns of exclusion, encoding existing biases and disregarding their perspectives — and they worry about AI displacing jobs. At the same time, many of these same patients are already turning to AI chatbots as a non-judgmental alternative, hopeful that AI might reduce the impact of provider bias and improve access to quality care.
Accuracy: Patients are aware of AI’s potential for errors and hallucinations, but are equally frustrated by existing inaccuracies in their care, such as misdiagnosis and insufficient examination. They’re hopeful about AI’s analytical potential, but only with robust human oversight and transparency about the limitations of any tools being used.
Privacy: Many patients feel general data privacy has already been significantly compromised and are willing to trade some privacy for better care — provided data use is transparent, de-identified, and serves a clear benefit to them or others. However, some patients remain wary, and those who feel in control of their data are more open to new tools and more likely to share sensitive information with their providers.
Transparency: Patients feel their current care already lacks adequate transparency, and want disclosure about AI involvement — its purpose, data use, error rates, known biases, and accountability structures — in detailed written form they can review at their own pace. They also want the conversational equivalent: providers who can answer questions and be a trusted source of information about how AI is being used in their care.
Beneficence: Patients are skeptical that AI in primary care is primarily for their benefit, and worry it will be deployed to increase efficiency and cut costs rather than improve care quality. They want honest, specific communication about how AI tools will help them — not vague claims about innovation — and expect that any efficiency gains translate into more provider time, greater access, and better outcomes.
Connection: Even patients who have had negative experiences in their medical care — feeling dismissed, judged, or misdiagnosed — don’t want their providers replaced by AI. Patients told us that good care is ultimately defined by a consistent, empathic relationship with a provider, and that well-designed AI could strengthen that relationship by freeing up more time for the interactions that make patients feel heard and cared for.
