A large study published in JAMA Network Open, surveying more than 6,700 radiologists in China, found that using AI diagnostic tools did not reduce burnout among radiologists — and in some cases was actually associated with a higher risk of burnout, particularly among radiologists carrying heavy caseloads and those who reported low acceptance of AI tools in the first place. The finding complicates a widely repeated industry narrative that AI assistance would ease radiologists’ workload by automating repetitive parts of image review.
The study’s authors found the association between frequent AI use and burnout was most pronounced specifically among radiologists with the heaviest workloads, suggesting AI tools may be adding friction rather than relief for exactly the clinicians who need help most. A leading explanation researchers point to is that AI-assisted workflows often increase postprocessing and interpretation time rather than reducing it — radiologists still need to review, verify, and often correct AI-generated findings, adding a new task layer on top of, rather than instead of, their existing workload.
What Burnout Actually Costs a Radiology Department
Burnout among radiologists is not simply a workplace wellness concern, it is linked in prior research to higher rates of diagnostic errors, slower turnaround times, and elevated staff turnover, each of which carries direct financial and patient-safety costs for the hospitals and imaging centers employing burned-out radiologists. A department that adopts AI tools expecting efficiency gains, only to see burnout rise among its highest-volume readers, risks incurring exactly the kind of downstream costs, mistakes, attrition, recruitment expense, that the AI investment was originally meant to help the department avoid.
Why This Contradicts the Standard AI Pitch
The dominant industry pitch for diagnostic AI has always centered on efficiency: automate the tedious parts of image review so radiologists can focus on complex cases and spend less time per study. The JAMA findings, along with a companion literature review published in European Radiology characterizing AI’s effect on radiologist burnout as “AI’s true black box,” suggest the evidence for that efficiency claim is far more mixed in practice than in marketing materials — with the review concluding that available data does not clearly support the claim that AI improves the core drivers of radiologist burnout.
A Trust and Verification Problem, Not Just a Workflow One
Part of what may be driving the counterintuitive burnout finding is that radiologists with low AI acceptance still have to work with AI-generated outputs in mixed clinical environments, creating friction between the tool and the clinician using it — verifying an AI finding you don’t fully trust arguably takes more cognitive effort than simply making the initial read yourself. That dynamic points to an underappreciated adoption challenge: AI tools succeed or fail not just on model accuracy, but on whether the radiologists using them actually trust the outputs enough to integrate them efficiently into their existing diagnostic process.
The Radiologist Shortage Backdrop
This research lands against the backdrop of a persistent, global radiologist shortage that has been one of the primary arguments for accelerating AI adoption in imaging in the first place — the logic being that AI-assisted reading could help a shrinking radiologist workforce keep pace with rising imaging volumes. If AI assistance is instead contributing to burnout among the radiologists already working the heaviest caseloads, it risks accelerating attrition from the field rather than easing the underlying staffing shortage it was meant to help solve.
What AI Vendors and Health Systems Are Saying
Companies building diagnostic AI tools, including firms like Aidoc that are working with a dozen health systems to co-design clinical AI workflows and gather real-time feedback, argue that burnout outcomes depend heavily on implementation quality — how well an AI tool is integrated into existing picture archiving systems, how much extra clicking or verification it demands, and whether radiologists were meaningfully involved in choosing and configuring the tools they’re asked to use. Poorly integrated AI, in this view, isn’t evidence AI itself increases burnout, but evidence that deployment matters as much as the underlying algorithm.
What Needs to Happen Next
Radiology leaders reviewing this research are calling for more granular studies that separate the effects of specific AI tool designs, workflow integration approaches, and institutional support structures, rather than treating “AI use” as a single undifferentiated variable. Until that more detailed picture emerges, health systems rolling out diagnostic AI tools in 2026 are being advised to pair any new AI deployment with explicit attention to radiologist workload and trust, rather than assuming efficiency gains will materialize automatically.
