Can AI help me understand my cancer diagnosis? What it can and can't do

Short answer: Yes — for translation. AI is genuinely good at turning pathology-report jargon into plain English, drafting the questions you should ask your oncologist, and helping you write updates to family. It is not reliable for diagnosis, staging, prognosis, or treatment decisions, and it should never be the last word before an appointment. Use it to walk in prepared, not to decide anything. The safe and unsafe uses are listed below.

Patients are already doing this. Oncologists now routinely hear "I asked an AI and it said…" in the exam room. But a May 2026 Penn Medicine study presented at ASCO found that the online information available to patients about using AI for cancer is scarce, mostly low quality, written well above a typical reading level, and — most importantly — usually silent on the risks. So people are using these tools with almost no guidance on where they break.

This guide is that guidance: what AI does well with a diagnosis, where it fails, and how to use it without getting hurt.

The one-line rule

AI is a translator, not a doctor.

If a task is "help me understand what this says" or "help me ask a better question," AI is useful. If the task is "tell me what I have," "tell me how long I have," or "tell me what treatment to pick," it is out of bounds — not because the answer will always be wrong, but because you have no way to tell when it is.

What AI can genuinely do for you

1. Translate your pathology report into plain language. Words like invasive ductal carcinoma, grade 2, ER+/PR+, HER2-negative, margins negative, 1/3 lymph nodes positive are a foreign language on the worst day of your life. AI is very good at rendering each term into a sentence you can actually read. See how to read a pathology report you don't understand for the manual version.

2. Turn your report into questions for your oncologist. This is the highest-value use by far. Paste the report and ask: "What should I ask my oncologist about this?" You get a list you can bring to a 15-minute appointment. Cross-check it against questions to ask after any new diagnosis.

3. Explain a proposed treatment before you consent to it. What a port is. What "adjuvant" means. What neoadjuvant chemo is for. What a sentinel node biopsy involves. This is textbook explanation, and AI handles it well.

4. Prepare you for the logistics. Insurance-appeal letters, FMLA paperwork, what to ask an employer, how to organize records, what to pack for infusion day.

5. Help you tell people. Drafting the text to your sister, the email to your manager, the words for your kids. See what to say to someone with cancer and how to talk to kids about a parent's cancer.

6. Decode the appointment you just left. You retained maybe 20% of what the oncologist said. Typing up your notes and asking for them in plain language is a legitimate and safe use.

What AI cannot do — and where it fails quietly

1. It cannot diagnose you. It has never seen your slides, your scans, or your body. A description of symptoms is not a workup. Anything it says about "what this probably is" is pattern-matching on text.

2. It cannot stage your cancer. Staging requires imaging, pathology, and often surgery, interpreted together. An AI guessing at a stage from a partial report can be off by enough to change everything.

3. It cannot give you a prognosis. Survival statistics describe large groups of people from years ago; they do not describe you, your biology, or the treatments available now. This is the single most damaging thing to ask an AI, because the number will lodge in your head and it is not about you.

4. It cannot choose your treatment. Treatment choice depends on biomarkers, comorbidities, kidney and heart function, prior treatments, trial eligibility, and what you personally want out of the next year. Your oncologist has that context. A chat window does not.

5. It gets things confidently wrong. AI can invent a drug interaction, misstate a dose, mangle a statistic, or cite a study that doesn't exist — in the same calm, fluent tone it uses when correct. There is no visible difference between the two. This is the central risk, and it is the one the Penn Medicine researchers found is most often missing from the content patients read.

6. Its knowledge has a cutoff, and oncology moves fast. Approvals, biomarker-driven therapies, and standards of care change within months. A confident answer can simply be out of date.

7. It is not private by default. A general consumer chatbot is not a medical record. Whatever you paste may be retained or used to train models depending on the product and settings. Strip names, dates of birth, and MRNs before pasting anything, and read the tool's privacy terms.

Safe uses vs. unsafe uses

| Use AI for this | Don't use AI for this | | --- | --- | | Defining terms in your report | Deciding whether you have cancer | | Writing questions for the doctor | Deciding your stage | | Explaining a treatment your team proposed | Choosing between treatments | | Summarizing what happened at an appointment | Estimating your survival odds | | Drafting insurance and work letters | Adjusting or stopping a medication | | Finding what a clinical trial term means | Deciding whether to enroll in one | | Practicing how to say it out loud | Replacing a second opinion |

How to use AI safely: a five-step method

1. Strip identifiers first. Remove your name, DOB, MRN, and your doctors' names before pasting anything.
2. Ask for translation, not verdicts. "Explain what this means in plain language" — not "what do I have" or "how bad is this."
3. Ask it to flag uncertainty. Add: "Tell me which parts of this you're unsure about and what a doctor would need to answer instead." Good answers will name their own limits.
4. Convert everything into questions. End with: "Turn this into questions I should ask my oncologist." That's the output that changes your care.
5. Verify with your care team before acting. Every single time. AI output is a draft agenda for a conversation, never a substitute for one.

Red flags: stop and call your care team

• The AI suggests changing, delaying, or stopping a treatment or medication.
• It gives you a survival number or timeline.
• It contradicts something your oncologist told you.
• It recommends a supplement, alternative protocol, or "natural" therapy.
• It tells you a symptom is nothing — new symptoms during treatment are a phone call, not a prompt.

What to say to your oncologist about it

Don't hide it. Most oncologists would much rather see your list than have you act on something silently. A version that works:

"I used an AI tool to help me understand my report and write down questions. I know it isn't a doctor. Can we go through the three that matter most?"

That framing does two things: it signals you know the limits, and it gets your questions answered instead of dismissed.

The honest summary

The reason patients turn to AI isn't that they distrust their doctors. It's that the appointment was 15 minutes, the report was written for another clinician, and the results landed in the portal at 9pm on a Friday. AI fills the gap between getting information and understanding it.

That's a real, useful job. Keep it in that job. Translation, preparation, and questions — yes. Diagnosis, staging, prognosis, and treatment decisions — no, and no exceptions.

If you want to do exactly that with your own documents, ask Cura — you can paste a report or upload a photo of it and get it back in plain language, along with the questions worth bringing to your next appointment. Cura reads what you send to answer you; it doesn't save your documents.