Agentic Patient 9: She built an AI companion for breast cancer patients - and won't upload her records to ChatGPT
Faces of Digital HealthShow Notes
"I am actually quite wildly uncomfortable with patients using LLMs." She built an AI companion for breast cancer patients — and she means it.
Ellyn Winters-Robinson was diagnosed with breast cancer in March 2022, months before ChatGPT launched. She wrote a book about it on her iPhone during chemotherapy. That book became AskEllyn, an AI companion used across a hundred countries. In this episode of The Agentic Patient — a Faces of Digital Health series on how patients actually use AI, which prompts, which guardrails — she talks to Tjasa Zajc about what an AI companion can hold that a clinician cannot, and why she still worries about where patient data goes.
Guest: Ellyn Winters-Robinson, CEO of The Lyndall Project and AskEllyn, author of "Flat Please Hold the Shame"
What the conversation covers:
- Building an AI companion from a book written on an iPhone during chemotherapy
- Why she keeps AskEllyn strictly non-medical, and how that guardrail held up under health-insurer review
- Whether one woman's lived experience can support patients with different cancers, cultures and languages
- Why traditional cancer support groups can become "places of collective trauma"
- Scanxiety, and what happened when she used her own chatbot during a CT scare
- Why she is uncomfortable with patients uploading medical records to ChatGPT or Claude
- Patient data rights, desperation, and the risk of being "victimized again" by AI tools
- The Canadian Cancer Society-funded study now testing whether AI companions actually help
- "The patient is the workflow" — lived experience as an untapped resource in health system design
- How clinicians can coach patients to use AI safely instead of pretending they aren't
Chapters:
00:00 Intro: why The Agentic Patient series exists
04:00 Meeting Ellyn Winters-Robinson
05:26 Diagnosed in 2022, before ChatGPT existed
07:36 From a book written on an iPhone to an AI companion
08:53 Why nurses and social workers started recommending it
09:54 Can one woman's story support every patient?
12:21 Shame, language, and cultures where breast cancer isn't discussed
13:25 Scanxiety — and taking a pep talk from your own chatbot
15:57 How AskEllyn is built on top of the LLMs
18:19 The non-medical guardrail, and how it held up under insurer review
21:51 Why patient AI use is outpacing the system
29:25 "The patient is the workflow": lived experience as untapped data
34:05 Inside the Canadian Cancer Society study
41:03 Why she's uncomfortable with patients uploading records to LLMs
45:54 The trauma healthcare never sees
6 tips on using AI as a patient: https://youtu.be/DGGVXxB4ygI?si=7m7HqCLKow51KSlQ
Faces of Digital Health:
Website: https://www.facesofdigitalhealth.com
LinkedIn: https://www.linkedin.com/company/faces-of-digital-health
Spotify: https://open.spotify.com/show/4cElKJHrauyP6QJQaCkvdY
Apple Podcasts: https://podcasts.apple.com/gb/podcast/faces-of-digital-health/id1194284040
Newsletter: https://fodh.substack.com
The Agentic Patient series: https://www.facesofdigitalhealth.com/agentic-patient
AskEllyn: https://askellyn.ai
#DigitalHealth #AIinHealthcare #BreastCancer #PatientAdvocacy #CancerSurvivorship #HealthTech #TheAgenticPatient
More from Faces of Digital Health
Three in ten AI pilots at Sheba Medical Center never scale. Eyal Zimlichman on why, and what Europe's regulation holds back.
When prevention works, nothing happens. That's why it gets cut. Joanna Lane on financing prevention as long-term investment.
Nearly €200m, twenty years of EU health data projects. Henrique Martins on what they built and what EHDS needs next.
One ultrasound tech, four scanning rooms. NVIDIA's David Niewolny on what "augmented autonomy" asks of clinicians.
Most interoperability debates focus on exchange formats. In this discussion with Dr Nils Hellrung, medical informatician and CEO Strategy & Operations at vitagroup, Nils focuses on the need for shaping hospital IT with governance and financing.