How we use AI

What AI does here

AI helps us draft plain-language explanations from authoritative sources, organize information around the moment a reader is in, suggest related pages, and run automated checks for problems like unsupported claims or missing context. It helps us cover more ground, more clearly, than a small team could alone.

Where AI is not trusted to decide

AI does not get the final say on anything that affects your safety. We aim for medical claims to trace to a named public source, such as the National Cancer Institute, and we check pages against their cited sources as part of an ongoing process that has not yet reached every page. Pages can still contain errors, and they are education, not medical advice. We do not use AI to diagnose, to recommend treatment for an individual, or to interpret your personal results.

To be plain about what that does and does not mean: most pages on higher-risk topics are published without waiting for a clinician, and every one of them carries a visible notice telling you what review it has and has not had. A smaller number we judge unsafe to publish that way are held back from search entirely until they have had an editorial check against their cited sources. That check is not clinician review. We would rather tell you the proportion than let “held for review” imply a safety net that covers more than it does.

Our source requirement

If a fact isn’t supported by a cited source, it doesn’t belong on the page. You can see the sources behind each page, and we encourage you to read the originals. See our sources, and editorial process for the full picture.

We also publish the measurements rather than only the promise: how checkable our sourcing actually is counts the pages that cite a specific document you can open and compare, and the pages that name only the publisher. Right now the second group is larger. That is our problem to fix, not yours.

Why most pages are not physician-reviewed

We label pages honestly. Most pages are written from authoritative sources and checked — but they have not been individually reviewed by a physician, and we do not claim they have. When a page carries a “medically reviewed” label, it means a named, credentialed reviewer checked it on the date shown. No page carries that label today. We will never invent a reviewer or a credential. Read more about our labels on how we review content.

How Ask Cancer Explained stays safe

Our “Ask Cancer Explained” tool does not use AI to write answers: it searches our own pages and shows the matching passages as written. It follows six safety rules:

• Results only from our pages: Passages only from published Cancer Explained pages and the NCI sources they cite — never from pages we are holding back for review, and never an invented resource or citation.
• Declines to diagnose: Will not diagnose you or choose a treatment for you.
• Shows its sources: Links each passage to its Cancer Explained page and to the original source.
• Flags emergencies: Watches for signs of an emergency (for example, a fever during chemotherapy, called neutropenic fever) and points you right away to 911 or your care team’s urgent line (and to 988 for thoughts of self-harm).
• Protects privacy: By default, never stores your personal health story or data that identifies you, and never sends your question to an AI company.
• Lets you report problems: If a result looks wrong or unsafe, tell us on the corrections page.

Other places AI appears

Audio narration of articles is an AI voice reading the published text, and some pages fall back to your device’s own text-to-speech voice. Translations into other languages are machine translations with automated checks and no documented human review. Many of the photo-style images, and some short video clips, are made with AI. They show general scenes, are not photographs of a specific real patient, and where an image is a Reprint Memories example it is labeled as AI-generated. We never use AI on the photos families send to Reprint Memories. Some scanned pages of donated works (such as a family’s hand-drawn book) were cleaned with AI image tools, and the work’s page says so; its narration voice and music are also AI-made.

How we handle what you type

Our interactive tools are built to minimize data. Please do not enter your name, medical record numbers, date of birth, address, or other identifying medical information. The symptom guide, screening checklist, and appointment and question builders run entirely in your browser and do not send what you enter to a server. The “Ask” tool sends your typed question to our own server to look up matching passages, never to an AI company, and is designed not to store the question text — it keeps only anonymous, non-identifying usage information. We do not claim HIPAA compliance, and these tools are not for transmitting a full medical record. See our privacy policy.

Limitations

AI can make mistakes, and no automated system catches everything. Our content is general education, not medical advice, and it is not a substitute for your care team, who know your situation. Pages here are not signed off by a clinician before they publish. That is not an oversight we are quietly working around: we restate published guidance and cite it, so the authority belongs to the source rather than to us, and every page names where its claims come from — you can verify us instead of trusting us. Where a volunteer clinician has reviewed a page, their name and credentials appear on it; where no name appears, no clinician has checked it. We are glad to have reviewers and are recruiting them, and we do not hold pages back waiting for one. Use this site to understand your situation and to ask better questions of the people treating you. If something on a page seems wrong, unclear, or unsafe, please tell us — see corrections and error reports or email corrections@cancerexplained.org.

Why transparency matters

AI lets a small nonprofit turn dense public health information into clear, free plain-language education. Being open about that — and about our limits — is how we try to earn trust rather than assert it.