Prompting My Way Through Cancer

Illustration of a man caught in a vaguely intestine-like maze with healthcare documents sprinkled about.

Let's get the most important parts of this story out of the way:

  1. I've recovered and I'm back to my normal life now, so no need to worry if this is the first you're hearing about my diagnosis.
  2. I'm definitely not a doctor, so please don’t take this as advice if Grok tells you to try horse dewormer or whatever.

From Training Plan to Treatment Plan

One random afternoon in late March, I went to the doctor about a stomach pain that was getting in the way of my routine. I had big plans for this year culminating in a 1500km London-Edinburgh-London cycling event in August, and I didn’t have time for my training schedule to get derailed.

Within hours, the doctor’s visit had escalated into emergency surgery. Initially diagnosed with appendicitis, surgeons found themselves unexpectedly removing part of my colon along with a mystery mass, which was identified as colon cancer about a week later.

Shortly after waking up following the operation, my surgeon debriefed me, casually mentioning the unplanned colon removal before swiftly exiting the room to continue his rounds. Stunned and full of unanswered questions, I started dumping everything into my notes app for the next time the surgeon checked in on me. As my list of questions grew, I realized I had enough content to fill a multi-hour podcast, and some of my questions were likely beyond the scope of his domain knowledge.

And so, like any nerd would do in 2025, I turned to ChatGPT.

Before you get your hackles up about how LLMs are hallucination machines and remind me that Google’s AI recommended everyone should eat one rock per day for health: Yes, I get it. I’ve been working with LLMs long enough to have a good understanding of where they succeed and fail. I wasn’t going to fire my doctors and blindly follow whatever ChatGPT told me, but I hoped that I’d get some value out of the exercise by using ChatGPT to prepare, process, and refine my questions while I was consulting with human experts.

What I wasn’t prepared for was just how effective ChatGPT would be at helping me navigate everything that followed. AI didn’t replace my care team, but it made me a much better patient.

Here’s how ChatGPT transformed this entire bizarre experience:

🔍 AI as First-Line Diagnostician

About a year before my surgery, I’d noticed my cycling performance unexplainably tanking. ChatGPT initially flagged iron-deficient anemia, which blood tests confirmed. This boosted my confidence in AI’s diagnostic prowess, although I glossed right over the part of ChatGPT’s anemia hypothesis that mentioned cancer as a common root cause of anemia for men with my background (oops).

As I continued regular blood testing, ChatGPT’s instant analyses prepared me to ask smarter, more relevant questions at appointments. With its 2025 memory update, ChatGPT started remembering all of my past tests and conversations, making interactions feel like chatting with an entire expert medical team at once. Questions about nutrition took blood tests into account, questions about cycling training had context from my surgical records.

Performance rating: Pretty damn good

There were a few times early in the process where it hallucinated some past test results, but once the memory feature was rolled out, that went away completely. Using ChatGPT to get an early read of my tests let me walk into appointments with a higher baseline, letting doctors spend their time helping me making decisions, not bringing me up to speed.

📖 AI as Medical Translator

A huge portion of my medical LLM usage has just been looking up what different medical terms mean. Going into this adventure, I was blissfully naive about almost everything related to cancer, including the condensed, technical shorthand you’ll find in lab reports.

While I wasn’t quite “George Costanza getting a biopsy result” level of naive, I had no idea what I was looking at when I opened my surgical pathology report. “pT3 N0 M0, AJCC Stage IIA” was gibberish to me, but ChatGPT helpfully told me that was the code for stage II cancer (I think I was expecting a giant “YOU’VE GOT CANCER” notification). Using LLMs to understand my pathology report beforehand gave me space to emotionally process the results so I could show up to my appointments in the right headspace to make decisions.

Performance rating: Lifesaver

Although this is a task I could have done with good ol’ fashioned Google, using ChatGPT made it frictionless to keep asking clarifying questions about a topic until I understood it. I can’t imagine how much time and patience it would have required for me to get to the same level of knowledge doing this with a human expert.

Pro tip: A great way to go deep quickly on a topic is to use Elicit to find relevant academic papers and feeding them into Google’s NotebookLM to ask questions. I first learned that trick here.

🛋️ AI as Therapist

Amusingly, I started using ChatGPT to process some of the mental challenges that came with my diagnosis at the exact time OpenAI accidentally made ChatGPT too sycophantic. Initially comforting, this exaggerated praise quickly became suspicious. Thankfully it was pretty easy to fine tune the model’s personality in the ChatGPT settings, and it went back to being supportive but not saccharine.

Journaling prompts generated by ChatGPT encouraged me to dive deeper emotionally, creating meaningful daily conversations. This culminated in creating a “North Star Manifesto,” something I’d typically roll my eyes at, but ChatGPT managed to help craft an authentic and genuinely helpful guide during tough moments.

Performance rating: Surprisingly therapeutic

I’ve done similar journaling work with a therapist before, and ChatGPT managed these conversations just as well as my human therapist did. And like every other role here, ChatGPT had a clear advantage in being always available, no appointment needed.

🚴 AI as Rehab Strategist

One of the first questions I asked ChatGPT after I was admitted to the hospital before my surgery (but before I knew it was cancer), was how this was going to affect my preparation for London-Edinburgh-London. After pasting in my current TrainerRoad training plan and event calendar, ChatGPT generated a return to sport plan that looked reasonable, and reassured me that I could still achieve my goals despite this setback.

That initial plan was quickly obsolete as my situation evolved, but ChatGPT was always able to generate new plans and strategies that met me where I was at that moment. Over the last few months, I’ve generated at least a dozen iterations of my training and rehab plan as I’ve made progress or experienced setbacks.

ChatGPT’s rehab judgement validated itself when I ignored its counsel against doing an hour of easy riding on the trainer only one week after surgery. After I recovered from that fiasco a day or two later, we developed a structured walking program that helped me rebuild my endurance and let me feel like I was making progress. It even suggested that I post the walks to Strava to give me the same sensation of doing “real” workouts. This felt a little silly but it worked in keeping me off the bike until I was ready.

Screenshot Sample chat about my training, click for full size

Performance rating: Remarkably personalized

It felt like having a sports physiologist, coach, and physical therapist on speed-dial. This weekend I did a 100 mile bike ride seven weeks after surgery, which I’m confident I could not have completed if I had just relied on the generic post-surgery rehab handout given to me as I was discharged from the hospital. My doctors gave me the clearance, but ChatGPT helped me turn that into a personalized, data-informed plan to follow.

⚠️ Reality Check: Limits and Lessons Learned

So, has AI been perfect in managing my cancer journey? Am I ready to toss aside those pesky humans and hand the reigns over to our silicon overlords? Obviously not quite, but it’s been a more effective enhancement to my treatment than I would have predicted a few months ago.

Here are a few basic tips I’d recommend for anyone using LLMs to augment their healthcare:

  • Provide as much context as possible. Every time a new report or test result appeared in my patient portal, it went straight into my ChatGPT “Cancer” project folder. I’d also add regular updates about my progress, as if I was checking in with my doctor. More data equals fewer hallucinations.
  • Verify all conclusions and suggestions, especially if they could have negative consequences. I had very few instances where I felt like ChatGPT was leading me astray, but I made sure to wait until I got the final word from my doctors before making any major decisions.
  • Approach the process with curiosity and openness. Have a question about something that you think is silly? ChatGPT doesn’t care. Have a passing thought that isn’t really urgent? Doesn’t matter, go ahead and ask. Quite a few of the trivial things I asked about ended up leading to more significant insights.
  • Keep additional notes externally. ChatGPT’s interface is great for an open-ended conversation, but it’s pretty bad for rapid information retrieval. If I had an important thing I wanted to remember and access easily, I’d put it into my notes app with a link to the chat. I really hope OpenAI adds more functionality to organize ChatGPT content in the near future.

🤖 Conclusion: A Patient Experience Designed by the Future

All things considered, I got really lucky with my cancer. I was lucky that we found it relatively early. Lucky that it was treatable with surgery, and there weren’t any serious complications. Lucky that this happened while I was living in Canada and not the United States (as an immigrant to Canada from the US, I still can’t wrap my head around the fact that I’ve paid $0 for my treatment). I’m especially lucky that I didn’t have to follow the surgery up with chemotherapy, and experience everything that goes along with that. And yes, I feel lucky that I started experimenting with ChatGPT as a clinical companion to help me navigate all of this.

It feels like a cliche to say that having ChatGPT during this process was like having a whole team of experts on hand 24 hours a day, but honestly, that’s exactly what it felt like. That’s not to diminish the amazing work my team of actual human professionals contributed to my outcome – I’m eternally grateful for them. But augmenting their skills and availability with an omnipresent and omniscient surprisingly capable AI made me a much more informed and much less terrified patient. I have no doubt that my recovery has been sped up by regularly consulting with my AI tools, and I’m better prepared to do what I can to remain disease free in the future.

👷 P.S. – A Note for the Builders

One last impression from this experience that I want to emphasize:

However hyped AI is, I think it’s still being underestimated.

I’ve been playing and working with LLMs since ChatGPT-3.5 launched in late 2022. But the last three months have been a step change in how much I rely on it for things I never would have considered before. And it works.

The models have gotten scary good, yet the products built around them are still treating LLMs like smarter search bars or productivity toys. If you’re on a product team right now, whatever you scoped six months ago probably isn’t ambitious enough for what’s possible today.

As William Gibson said, “the future is already here – it’s just not evenly distributed.” Nowhere is that truer than with AI. And I say this as someone who’s been living on the edge of that distribution curve. Start experimenting with things that feel slightly impossible. You’ll be surprised how often they’re not.

Mark Allen

Written by Mark Allen, a product manager and designer currently based in Toronto. Say hello on Bluesky or send me a message.

Receive new posts by email