Using AI as a Second Opinion: a Patient’s Playbook

Conno Christou is one of the most health-tracked founders you’ll meet. He wears a Whoop, cross-checks it against an Oura ring, and runs nearly 100 biomarkers every year. Then, as TechCrunch AI reports, the 35-year-old was diagnosed with an aggressive non-Hodgkin’s lymphoma that no wearable predicted, a rare cancer hitting roughly one in 420,000 people. What stands out is what he did next: he ran his treatment like a data problem and used Claude to ask sharper questions than the system handed him.

This is significant because a third of American adults now use chatbots for health information, according to a March poll cited by TechCrunch AI. Here’s how Christou actually did it, step by step, plus the cautions experts attach to it.

Quick Start: You’ll learn how one patient used AI as a research partner during a cancer diagnosis. What you need: your own medical records (bloodwork, scans, journals), access to a capable AI model, and the discipline to verify everything with real doctors. AI here is a question-sharpener, not a doctor.

📋 The Playbook

  1. Track your baseline data. Christou had four straight years of annual bloodwork and wearable history. Why it matters: when something goes wrong, you already have a personal baseline to compare against, not a blank chart.
  2. Don’t accept the first recommendation. His first oncologist suggested the lighter chemo regimen. The night before his first infusion, he sought a second opinion. That doctor recommended the harder regimen, pushing his odds from roughly 60% to around 85%. “As founders, we hold the wheel,” Christou says. “You don’t have to follow the first advice.”
  3. Gather many opinions, not two. Over two days he collected 12 opinions total, calling hematologists and oncologists in the US and abroad. The vote came back 11 to 1 for the harder path. He took it. Why it matters: one expert can be wrong; a weighted consensus is harder to argue with.
  4. Log everything as data. He kept a symptom journal using voice transcription, recording every side effect, medication, and counter-medication. His Whoop flagged the days his immune system would bottom out, sometimes before symptoms showed. Why it matters: structured records turn a chaotic experience into something an AI can actually analyze.
  5. Narrow your focus. He tracked three variables: sleep, nutrition, and, above all, psychology. “It moves the needle more than anything,” he said. “I never asked ‘why me,’ not once. That question has no useful answer.”
  6. Feed it all into the model to ask better questions. He fed blood results, scan data, wearable output, and journal entries into Claude. His framing is the key lesson: “It didn’t replace the doctors,” he says, but it “helped me ask the right questions.” For a cancer an oncologist might see once a year, a model trained on the full medical literature beat a Google search.
  7. Use AI to pressure-test ambiguous results. His final PET scan came back unclear, and his oncologist raised radiotherapy near his heart and lungs. Christou learned the false-positive rate on these scans runs around 60%. He fed three PET scans and an MRI into Claude, which flagged thymus rebound, a known phenomenon in patients under 40 where the gland reactivates after chemo and mimics active disease. The model put it at roughly 90%.
  8. Verify with humans before acting. He sought three more opinions. The fourth doctor confirmed thymus rebound. No active disease. No radiotherapy needed.

⚠️ The Caution: Experts are not sold on chatbots for diagnosis. Danielle Bitterman, clinical lead for data science and AI at Mass General Brigham, told the New York Times that general-purpose chatbots are frequently wrong and “have not been thoroughly evaluated” for personalized diagnoses. Christou agrees. The model informed his questions; doctors made the calls.

Next steps: Start a structured health journal now, before you need one. Keep copies of your own scans and labs. And if you ever face a major diagnosis, treat AI as a second researcher who helps you interrogate the experts, never as the final word. More details are in the original TechCrunch AI report.

Scroll to Top