Two new Pew Research Center studies released August 25, 2026 offer the clearest public-opinion picture yet of how Americans feel about artificial intelligence in their medical care — and the findings land in a California workers’ compensation system where AI-assisted medical review is already a live regulatory question. Both studies draw on the same survey of 3,488 U.S. adults conducted June 22–28, 2026, through Pew’s American Trends Panel.

The first report, Americans want transparency when AI is used in their healthcare, found that 72% of U.S. adults consider it extremely or very important that a doctor or other healthcare provider tell them when AI is being used in their care, with majorities saying so across every age, gender, education, and racial group surveyed.

Crucially for anyone administering medical review, the demand for disclosure tracks how directly a task affects clinical decisions. Roughly eight in ten say they should be told when AI is used to analyze medical scans (81%), make a diagnosis (81%), or explain lab results (80%). Support drops but stays substantial for background tasks: 72% for AI note-taking during an appointment, 64% for prescription refill ordering, and 56% for appointment scheduling — the only task where a meaningful share (33%) said disclosure isn’t necessary.

Americans also report feeling out of the loop. A 53% majority say they have little or no say over whether a provider uses AI in their care, and respondents were three times as likely to want more input as to say they’re comfortable with the input they have (63% vs. 21%). Nearly half (46%) simply don’t know whether AI has been used in their own care at all; only 16% say it has. Even among those who know AI was used, just 22% say they understand well how it was used, while 32% say they don’t understand it well.

The companion report, From Diagnoses to Treatments, Why Americans Use AI Chatbots for Health, shows a parallel trend from the patient side: 34% of U.S. adults now use AI chatbots for at least one health-related purpose. A quarter (25%) use them to figure out what’s causing symptoms, 28% for speed, 22% to learn more about a doctor’s diagnosis, 22% for treatment information, 20% to understand lab results, and 15% to decide whether to see a doctor at all. Another 18% turn to chatbots for health topics they’re uncomfortable discussing with a person, and 22% cite low or no cost as a motivation.

Users find the results useful: 47% say chatbot health information is extremely or very helpful and another 48% say somewhat helpful, with only 5% finding it unhelpful. But comfort with data-sharing is split roughly evenly — 29% are extremely or very comfortable sharing personal health information with a chatbot, 26% are not, and 42% land in the middle.

Does this map onto UR, IMR, and QME/AME? The short answer is that it maps onto the utilization review and independent medical review side quite directly, and onto the QME/AME side more as an emerging question than a settled one.

California has already legislated on the UR question. Senate Bill 1120, the Physicians Make Decisions Act, took effect January 1, 2025 and directly regulates AI in utilization review and utilization management for health care service plans and disability insurers. It does not ban AI, but it requires that a determination of medical necessity be made only by a licensed physician or licensed health care professional competent to evaluate the specific clinical issues, after reviewing the requesting provider’s recommendation and the patient’s individual clinical circumstances; that AI not deny, delay, or modify services based in whole or in part on medical necessity; that AI analysis not rest solely on group datasets; and that plans disclose how AI tools are used in their utilization review policies. In other words, California law already treats AI in UR as a decision-support tool rather than a decision-maker — which is essentially the arrangement Pew’s respondents say they want, provided they’re told about it.

A companion statute, Assembly Bill 3030, also effective January 1, 2025, addresses the transparency question head-on for clinical communications: health facilities, clinics, physician’s offices, and group practices using generative AI to produce written or verbal patient communications about clinical information must include a prominent disclaimer that the communication was AI-generated, plus clear instructions for reaching a human provider. Critically, that requirement drops away entirely if a licensed human provider reads and reviews the communication before it goes out — a design choice that mirrors the intuition in Pew’s data, where disclosure demand rises with the degree to which AI is actually shaping the medical output rather than assisting a human who remains accountable.

Where the fit is looser is the workers’ compensation-specific machinery. SB 1120 amends the Knox-Keene Act and the Insurance Code, which govern group health plans and disability insurers — not Labor Code section 4610, which governs workers’ compensation utilization review, nor the IMR process under section 4610.5 that resolves UR disputes through Maximus’s independent reviewers. AB 3030 similarly reaches health facilities and physician practices, not the medical-legal evaluation process. Whether and how those statutes’ principles extend to comp UR, IMR, or QME and AME reporting is not settled by their text, and RAND’s recent DIR-funded review of California’s UR system did not address AI-assisted review at all — a gap worth noting given that the same report recommended DIR build out standardized UR data infrastructure it currently lacks.

For QME and AME practice, the chatbot findings raise a distinct issue evaluators are likely to encounter with increasing frequency: injured workers arriving at evaluations having already used AI to interpret their own symptoms, imaging, or lab results, and sometimes to form expectations about diagnosis, causation, or work restrictions before the evaluation begins. With a third of adults using chatbots for health and use running highest among younger workers — the demographic filing the largest share of comp claims — that dynamic is likely to shape the evaluation encounter itself, independent of whether the evaluator uses any AI tool.

And on the evaluator side, the question of whether a QME or AME may use AI to assist in drafting a medical-legal report, and whether that use must be disclosed, remains substantially unaddressed by California’s existing AI health statutes or by DWC regulation. Given that a medical-legal report is a formal evidentiary document rather than a patient communication, and that the Board has already sanctioned an attorney in a workers’ compensation matter this year for filing AI-fabricated case citations without verification, the disclosure and verification norms now settling into legal practice may well arrive in the medical-legal context before regulators formally address them.