Online information about artificial intelligence (AI) and its cancer research and treatments for both patients and general audiences is limited, and available web pages and videos are mostly of low quality, difficult to read, and often omit the risks of using AI, according to a new study presented today at the 2026 American Society of Clinical Oncology (ASCO) Annual Meeting and led by researchers at Abramson Cancer Center (ACC). Penn State and Penn’s Perelman School of Medicine.
This study highlights the opportunity to create more accessible, high-quality resources as AI is increasingly incorporated into cancer research and patient care and used by patients to understand diagnosis, treatment options, and prognosis.
In our clinics, we know our patients are using this new technology because we are constantly hearing questions from our patients about what our AI tools are telling us. Clinicians are used to educating patients about the risks of treatments, but they are unaware of the risks of misinformation that can come with using AI tools in the context of cancer treatment. Our goal was to gain a basic understanding of what patients would find online if they sought information about using AI to learn about cancer treatment. ”
Henry Litt, MD, Hematology-Oncology Fellow and senior author of this study
Most online information about AI and cancer treatment is of low quality and difficult to understand
The research team, led by internal medicine resident Dr. Pearl Subramanian, screened the first 320 web pages and videos identified through Google and YouTube searches using common keywords related to cancer and AI. Web content included everything from news articles to articles published on hospital and healthcare organization websites, educational resources published by government agencies and medical societies, and videos from influencers.
After removing content not related to AI or cancer treatment or not intended for a general audience (such as academic papers), 52 web pages (31 percent of Google search results) and 29 videos (19 percent of YouTube search results) were included in the final analysis. The team used validated tools to evaluate the readability of web pages and the overall quality of web pages and videos in terms of consumer health information best practices. We also tracked whether the content covered key AI safety concepts, such as the need for clinical supervision and the potential for inaccuracy.
Only 17 web pages (33 percent) and seven videos (23 percent) in the sample were considered high quality. Although the American Medical Association (AMA) and the National Institutes of Health (NIH) recommend a 6th to 8th grade reading level for consumer health information, the median readability of web pages was at college level. Only 15% of webpages mentioned the risk of AI hallucinations.
Many of the resources that discussed AI safety practices did so in terms of how AI is being integrated into clinical care—how doctors and hospitals are using AI tools, rather than focusing on the risks faced by patients as direct users of the technology.
For example, if a patient asks a chatbot whether the side effects of a treatment they are experiencing are normal, they are likely to receive a well-researched and accurate answer. Or they may provide hallucinatory (fabricated) information that patients may ignore, rather than address potentially serious side effects with their treatment team. Alternatively, without a patient’s medical history, the chatbot may miss important context that could change its response.
“While we always encourage patients to consult their care team as their primary source of information about their care, we recognize that patients always refer to other resources as well,” Subramanian said. “We know that patients will use AI to ask questions about their cancer treatment, and they need access to resources to help them learn how to safely interact with these tools.”
Opportunities to improve patient resources on AI and cancer care
The authors say the findings provide a clear call to action for health systems, cancer centers, and oncology organizations to develop high-quality, public-facing resources for patients and to set standards for the development of these resources, including appropriate reading levels and best practices for including safety information.
“Given that only one in four of our search terms was considered relevant to patients, and only one in three of these was of high quality, patients may have a hard time finding useful, plain information,” said co-author Ronak Mamtani, MD. He holds the David J. Vaughan, MD Professorship in Oncology at GU and served as the research supervisor. “As AI becomes more integrated into oncology, patient education should be prioritized as a key part of the AI implementation strategy.”
sauce:
University of Pennsylvania School of Medicine
