Patients seeking information about the use of AI in cancer care encounter few resources, many of which are low quality, difficult to understand, and written above recommended reading levels, according to findings from a cross-sectional analysis presented at the 2026 ASCO Annual Meeting (Abstract 9000).  

Additionally, fewer than 20% of patient-facing webpages addressed potential risks associated with AI, such as hallucinations and misinformation. 

“What we found about the misinformation risk was probably the biggest story,” said presenting author Pearl Subramanian, MD, an internal medicine resident at the Hospital of the University of Pennsylvania, in an interview with ASCO AI in Oncology during the meeting. She noted her surprise at the amount of publicly available information that failed to adequately communicate the risks associated with AI.  

The findings from this cross-sectional analysis will also be published in a forthcoming report in JCO Clinical Cancer Informatics.  

Background and Study Methods 

Polls have indicated that about one in three adults turn to AI tools, particularly chatbots, for health information.  

Given the growing use of AI tools for health-care information, researchers sought to evaluate the quality and readability of publicly available resources on AI in cancer care, as well as how effectively those resources explained AI safety concepts.  

The research team conducted a cross-sectional analysis of patient-facing content identified through Google and YouTube searches in August 2025, using top search terms from Google Trends related to AI and cancer. Two independent reviewers screened the first 170 webpages and 150 videos identified through the searches. Discrepancies between the two reviewers were resolved by a third reviewer.  

The researchers used the DISCERN instrument, a validated tool for evaluating consumer health information, to assess the quality of the webpages and videos. Readability was evaluated using the Flesch-Kincaid, Gunning Fog, and SMOG indices. To assess AI safety concepts, the reviewers examined whether the resources addressed hallucination and misinformation risks, clinician oversight, bias and equity, and transparency.  

Key Findings 

After the initial screening, 239 records were excluded because they were not truly patient-facing or were unrelated to AI and cancer. A total of 81 resources (25%) were included in the analysis, comprising 52 webpages and 29 videos.  

Interobserver agreement for screening decisions was substantial for webpages (κ = 0.65) and high for videos (κ = 0.88).  

According to all three readability indices, patient-facing webpages were difficult to read. Although the recommended reading level for patient education materials is eighth grade, all resources exceeded this threshold, with a median reading grade level of 12.8.

According to the DISCERN instrument, most patient-facing webpages and YouTube videos showed average quality scores of 2.5 and 2.25, respectively, while a score of 4 or higher was considered indicative of high quality.  

Among patient-facing webpages, more than 50% addressed bias and equity concerns, and approximately 80% discussed clinician oversight and transparency. However, fewer than 20% addressed  the risks of hallucinations and misinformation in AI and cancer care.  

“Patients are actively using AI; however, this is without adequate guidance,” Dr. Subramanian said during her presentation. “A lot of our resources currently focus on how clinicians should use AI, such as when using OpenEvidence, or how we use screening algorithms, but not on the risk that patients face when using AI themselves.”  

“This is particularly notable when it comes to the misinformation risk, which is a critical gap, especially during this stressful time,” she continued. “There’s a lot of content out there, but not much guidance on how our patients and their families should be using this information.” 

Limitations 

Dr. Subramanian noted that the analysis captured only a single point in time, while the use of AI in cancer care continues to evolve rapidly. She also acknowledged that the DISCERN tool was developed and validated long before the emergence of AI in cancer care, thus it may not fully capture AI-specific aspects of content quality.  

Additionally, the findings may have limited generalizability because the analysis was restricted to resources identified through Google and YouTube searches in the United States. Dr. Subramanian noted that patients may also seek information about AI and cancer through social media platforms such as TikTok, Instagram, and Reddit. 

Call to Action 

Dr. Subramanian said organizations such as ASCO, the National Cancer Institute, and the National Comprehensive Cancer Network are well positioned to partner with patient advocacy groups to develop higher-quality, plain-language, patient-facing resources.  

“We’re looking forward to how we can prioritize patient resources that are accessible, that discuss what safety risks are out there, and how we can improve oncology care for patients with cancer,” she said.  

Expert Insights 

ASCO Discussant Eleonora Teplinsky, MD, FASCO, the Head of Breast and Gynecologic Medical Oncology at Valley Health System, further emphasized the need for patient education on the use of AI in cancer care.  

She noted that only 33% of analyzed webpages and fewer than 40% of YouTube videos were classified as high quality. In addition, only 15% of resources addressed the risks of AI hallucinations and misinformation.  

“We know that AI can provide misinformation and hallucinations, but if you think about how few resources are available, this will further the misinformation epidemic that we live in currently,” Dr. Teplinsky said.  

She noted that the YouTube videos included in the analysis had an average view count of about 127, suggesting that information about AI and cancer may have limited reach.  

Dr. Teplinsky recommended developing AI-specific quality standards, expanding AI health literacy education, providing guidance on the safe use of AI for both clinicians and patients, and establishing frameworks for addressing conflicts between AI-generated recommendations and clinical judgment.  

She also highlighted several unresolved questions surrounding the use of AI in cancer care, including who should be held accountable when AI-generated misinformation causes harm, how AI literacy can be improved without exacerbating existing inequities, where patients and clinicians should turn for trustworthy AI education, and whether simply creating more online resources is the best solution.  

She further questioned who bears responsibility for AI education in cancer care, suggesting that potential stakeholders include professional societies, national organizations, clinicians, health systems, AI companies, regulators, and patient advocacy groups. The answer remains unclear and is likely to be the subject of ongoing debate, she said. 

DISCLOSURES: For full disclosures of the study authors, visit asco.org. Dr. Teplinsky reported a consulting or advisory role with AbbVie, Daiichi Sankyo/AstraZeneca, Immunogen, Novartis, and Pfizer. 

ASCO AI in Oncology is published by Conexiant under a license arrangement with the American Society of Clinical Oncology, Inc. (ASCO^®). The ideas and opinions expressed in ASCO AI in Oncology do not necessarily reflect those of Conexiant or ASCO. For more information, see Policies.