The report about your own body was never written for you to read. A major new review shows AI can fix that — if it's done responsibly.
For decades, a radiology report has been a conversation between two clinicians. A radiologist reads the scan and writes for the referring physician. The language is precise, dense, and built for a medical audience — "no acute intracranial abnormality," "3.2 cm spiculated mass," "nonspecific T2 hyperintensities."
Then the 21st Century Cures Act changed who reads it first. Results now land in the patient portal the moment they're signed — often before a clinician has had the chance to call. So the person whose body the report describes opens it alone, on a phone, on a Friday night, and tries to decode a language that was never meant for them.
A new systematic review in The Lancet Digital Health puts hard numbers behind what patients have experienced for years — and points to a credible way forward for patient-friendly radiology reports.
What the Lancet study found about AI-simplified radiology reports
The review, led by Dr. Samer Alabed at the University of Sheffield and Sheffield Teaching Hospitals NHS Foundation Trust, pooled 38 studies published between 2022 and 2025. Together they generated 12,922 radiology reports simplified by large language models, evaluated by 508 assessors — 387 members of the public and 121 medical professionals. Most of the simplifications (92% of studies) used OpenAI's GPT models, with GPT-4 the most common.
The headline result: across patients, the public, and clinicians, LLM-simplified reports were consistently rated more understandable than the originals — while clinicians still judged them largely accurate and complete. In practical terms, readability dropped from roughly university level to the comprehension of an 11–13 year old, making the reports almost twice as easy to understand. Patients also read them faster and with lower cognitive workload.
For anyone who has watched a patient spend a weekend Googling a phrase from their own scan, this is a meaningful validation of a simple idea: comprehension is a solvable problem, not an inevitable one.
The finding that matters most is the caveat
The same review is careful about the risk. A small but real proportion of AI-simplified radiology reports contained clinically significant errors. And most of the underlying studies were small, single-center, and skewed toward younger, more educated, English-speaking participants — which limits how far the results generalize.
This is not a footnote. It's the whole design brief.
An error rate near zero is the difference between reassurance and harm. A raw chatbot, handed an unedited report and pointed at a frightened patient, is not a patient-communication strategy — it's a liability. The lesson from the evidence isn't "let AI rewrite reports." It's "simplify reports in a way that keeps clinical accuracy and human oversight at the center."
That distinction — between simplification as a novelty and simplification done responsibly — is where the real work is.
Why unreadable reports are a health-system problem, not just a patient one
The comprehension gap doesn't stay with the patient. It flows straight back into operations.
When patients don't understand a result, they call. Front-desk staff and nurses field clarification questions that could have been answered by the report itself. Anxiety drives unnecessary follow-up appointments. Non-adherence rises when instructions aren't understood. And in an era where health systems are, for the first time, ranking patient experience as their top strategic priority, the moment a patient opens an unintelligible result is one of the most overlooked breakpoints in the entire journey.
Access to information has been solved. Understanding of that information has not. Faster access to a report a patient can't interpret can actually make the experience worse.
What patient-friendly radiology reporting should look like
The Lancet review points to the ingredients of a system worth trusting — and they're the principles we design around at FlexReport:
Accuracy first, always. Simplification can never introduce or drop clinical meaning. Clinical oversight has to be part of the workflow, not an afterthought. The technology translates the report; it does not replace the clinician who interprets it or the physician who owns the diagnosis.
Plain language, not dumbed-down language. The goal is an explanation a patient can follow without losing what the report is actually saying — the balance the study measured as "more understandable while remaining accurate and complete."
Meeting patients where they are. The review flagged that its evidence skewed toward educated, English-speaking participants. Real patient populations don't. That's exactly why multilingual support, visual aids, and audio matter — comprehension can't be a privilege reserved for the most health-literate patients.
Reducing anxiety and load — for patients and staff. Lower reading time and cognitive workload for patients; fewer clarification calls and follow-ups for providers. Both were signals in the data.
The report isn't the finish line
The most useful reframe from this research is quiet but important: delivering a result is not the same as delivering understanding. For years we've treated the signed report as the end of the diagnostic process. For the patient, it's the beginning of a question — what does this mean for me?
The evidence now suggests we can answer that question at scale, in plain language, without sacrificing accuracy — if we build it carefully. That's the opportunity in front of every hospital, diagnostic center, and radiology network right now.
At FlexReport, that's the entire premise: turning complex clinical reports into patient-friendly explanations, so patients understand their health information with confidence — and clinicians keep their time and their authority.
Frequently asked questions
Can AI make my radiology report easier to understand? Yes. The 2026 Lancet Digital Health review found that AI-simplified radiology reports were rated nearly twice as easy to understand as the originals, lowering the reading level from university-level to roughly that of an 11–13 year old — while clinicians still rated them largely accurate and complete.
Are AI-simplified radiology reports accurate? Mostly, but not perfectly. The review found that a small proportion of AI-simplified reports contained clinically significant errors. That's why responsible tools keep clinician oversight in the loop rather than sending raw AI output directly to patients, and why AI should translate — not interpret or diagnose.
Does an AI-simplified report replace my doctor? No. A plain-language explanation helps you understand what a report says, but your physician remains responsible for interpreting the findings, answering questions, and deciding on any next steps. Always discuss your results with your care team.
Why are radiology reports so hard to read in the first place? Radiology reports are written for referring physicians, not patients. They use precise clinical terminology and typically read at a university level — far above the average patient's health literacy — which is the gap patient-friendly reporting is designed to close.
Reference: Alabed S, et al. "Large language models for simplifying radiology reports: a systematic review and meta-analysis of patient, public, and clinician evaluations." The Lancet Digital Health, 2026. https://www.thelancet.com/journals/landig/article/PIIS2589-7500(25)00142-6/fulltext
This article is for general information and does not constitute medical advice.
