AI Cancer Information Gaps Put Patients in the Hot Seat

AI cancer information gaps are now a patient safety problem, not just a web quality problem. A new Penn Medicine analysis found that patient-facing pages and videos about artificial intelligence in cancer care are scarce, hard to read and weak on safety, leaving people who ask chatbots about symptoms, side effects or treatment options with too few warnings before they act.

The study, presented on May 31, 2026 at the American Society of Clinical Oncology (ASCO, the major cancer research meeting) Annual Meeting, catches oncology at a tense moment. Artificial intelligence (AI, software that can generate text, predictions or recommendations from data) is moving into cancer clinics, while patients are already using consumer tools at home.

The Search Result Became the Safety Problem

Penn Medicine researchers did something simple and revealing. They looked at what a patient or family member might find after typing common cancer and AI search terms into Google and YouTube. The team screened the first 320 webpages and videos, then removed material that was irrelevant, too technical or not meant for lay readers.

That last number matters because cancer questions are rarely casual. A patient may ask whether a fever during chemotherapy is normal, whether a new symptom sounds urgent or whether a treatment option fits their diagnosis. A polished answer can feel reassuring even when it lacks the medical history, lab values and treatment plan needed to be safe.

What Penn’s Screen Captured

The Penn study fits a wider pattern in cancer communication research. Different teams have tested different doors into the same problem: search results, chatbots, reading level and actionability. The details vary, but the patient burden keeps showing up in the same place.

The table shows why the Penn result lands harder than a standard bad-webpage finding. The problem is not only that information is thin. It is that the tool a patient may turn to next can sound more complete than the pages warning them how to use it.

The Missing Risk Label

A useful patient page about AI in oncology has to do more than celebrate efficiency or explain how hospitals use algorithms. It must speak to the person alone at night with a phone, a diagnosis and a question they are afraid to ask out loud.

Large language models (LLMs, AI systems trained to generate human-like text from patterns in data) can summarize general cancer concepts. They can also give confident answers that omit context or invent details. The World Health Organization (WHO, the United Nations health agency) has urged caution when LLMs are used for health information, noting risks tied to bias, inaccurate answers, privacy and plausible but wrong outputs in its WHO guidance on safe AI for health.

That is the risk label patients need in plain words. A chatbot can help prepare a list of questions for an oncologist. It should not decide whether a side effect is safe to ignore, whether treatment should stop or whether a prognosis applies to a specific person. Confidence is not clinical oversight, and fluent wording can hide missing facts.

Readability Is a Clinical Bottleneck

Penn’s team found that the median webpage in its sample was written at a college reading level, even though consumer health information is commonly urged toward much lower grade levels. That is not a style complaint. In cancer care, the difference between plain and complex language can change whether a patient knows when to call the clinic.

The Agency for Healthcare Research and Quality (AHRQ, a U.S. health quality agency) says almost a quarter of U.S. adults have low literacy, a third have low numeracy and the average adult reading level is around eighth or ninth grade. Its AHRQ health literacy materials tool also warns that health education materials are often written well above that level.

Good web medicine is not just shorter sentences. The Centers for Disease Control and Prevention (CDC, the U.S. public health agency) offers a CDC Clear Communication Index with four introductory questions and 20 scored items to assess public messages. It asks whether the main message is obvious, whether the audience is defined and whether the action step is clear. For AI and cancer pages, reading level is a safety feature.

A Safe AI Page Needs More Than a Warning

Many hospitals already publish patient pages about chemotherapy, radiation, imaging and clinical trials. AI guidance should sit beside those resources, not in a technical corner of a website. The Food and Drug Administration (FDA, the U.S. medical products regulator) notes that AI and machine learning can help medical devices by drawing insights from health data, while also requiring careful management across a product life cycle in its FDA artificial intelligence software guidance.

Consumer chatbots are a different category from many FDA-reviewed medical tools, which makes patient education more important. A hospital page does not need to teach model architecture. It needs to teach judgment.

• Set the boundary – say that AI can help organize questions, translate jargon or summarize general topics, but cannot replace an oncology team.
• Name urgent scenarios – tell patients to call for severe pain, fever during treatment, breathing trouble, uncontrolled vomiting, bleeding or new neurologic symptoms.
• Explain hallucinations – use examples, such as invented drug names, outdated treatment claims or false reassurance about side effects.
• Protect privacy – warn patients not to paste medical record numbers, full names, addresses or sensitive documents into public tools unless they understand the privacy policy.
• Give a better prompt – suggest asking for a plain-language summary and a list of questions to bring to a clinician, not a treatment decision.

The strongest pages will also say what AI is already doing inside the health system. Patients can handle nuance. What they cannot use is vague comfort that says AI is promising without telling them where the edge is.

The Clinic Becomes the Backstop

Henry Litt, MD, a hematology-oncology fellow at Penn Medicine and senior author of the study, said patients are already bringing AI-generated information into clinic conversations. Pearl Subramanian, MD, an internal medicine resident at Penn Medicine and the study’s lead researcher, focused the analysis on material meant for lay audiences. Ronac Mamtani, MD, the David J. Vaughn MD Professor of Genitourinary Oncology and faculty advisor on the study, pointed to a need for plain-language resources as AI moves further into oncology.

The practical shift is small but important. Oncologists and nurses already ask patients about supplements, second opinions and internet searches. They may now need one more question: have you asked an AI tool about this? That question should not shame the patient. It should open the door to correcting bad advice before it shapes a decision.

Cancer centers also have a chance to make their own materials easier to find than weaker sources. A short, plain guide on safe chatbot use can live on diagnosis pages, patient portals, after-visit summaries and survivorship resources. The guide should be reviewed as models and cancer guidelines change.

If cancer centers make safe AI use part of patient education, chatbots can become a prompt for better conversations. If they leave the lesson to search rankings, the next answer a frightened patient sees may carry more authority than it deserves.

Frequently Asked Questions

What Did Penn Medicine Find About AI Cancer Information Online?

Penn Medicine found that patient-facing online information about AI and cancer care was limited, often low quality, hard to read and frequently missing safety risks such as hallucinations.

Should Patients Use AI Chatbots for Cancer Questions?

Patients should treat AI chatbots as a support tool for organizing questions or understanding general terms, not as a source for treatment decisions, urgent symptom advice or personal prognosis.

What Is an AI Hallucination in Cancer Care?

An AI hallucination is a false or made-up answer that sounds confident, such as an invented treatment claim, outdated guideline, incorrect side effect warning or reassurance that does not fit a patient’s case.

Why Does Reading Level Matter for Cancer Information?

Reading level matters because cancer decisions often involve risk, timing and numbers, and materials written at college level can leave many adults unsure about what action to take.

What Should Cancer Centers Add to AI Patient Guides?

Cancer centers should add plain-language boundaries, urgent symptom rules, privacy warnings, examples of hallucinations, safer prompts and a clear reminder to bring AI answers to the oncology team.

Disclaimer: This article is for informational purposes only and does not provide medical advice. AI tools can produce inaccurate or incomplete health information, especially in cancer care. Patients should speak with a licensed clinician or oncology care team before acting on symptoms, side effects, treatment choices or prognosis information. Figures are accurate as of publication.