Young Cancer Survivors Say AI Chatbots Could Fill a Gap in Care, If They Can Be Trusted

For young adults who have finished cancer treatment, the hardest part often begins after the last infusion or surgery. Symptoms such as fatigue, pain, brain fog, and anxiety can persist for months or years, yet clinic visits are short and survivorship support is thin. A new qualitative study published in Supportive Care in Cancer suggests that this generation of patients, raised on smartphones and already experimenting with tools like ChatGPT, sees generative artificial intelligence chatbots as a plausible bridge across that gap. But the eighteen young adult survivors interviewed by researchers at the University of Michigan Rogel Cancer Center drew a sharp line: AI may guide, reassure, and inform, but it must never replace their doctors or make decisions on their behalf.

The research team, led by nurse scientist Yupawadee Kantabanlang with colleagues including Robert Knoerl, recruited eighteen post-treatment survivors aged 18 to 39 who had completed cancer treatment between one month and two years before enrollment. The median participant was 29 years old, and most identified as female and White. Breast cancer was the most common diagnosis, followed by lymphoma, sarcoma or bone cancer, and melanoma. Before sitting down for interviews, participants completed validated questionnaires including the PROMIS-29 profile, which measures physical function, anxiety, depression, fatigue, sleep, pain interference, and social participation, along with a self-efficacy scale for managing symptoms and a three-item Digital Health Literacy Scale.

The questionnaire results painted a picture of a digitally fluent group carrying a moderate symptom load. Anxiety scores fell in the mild to moderate range, with a median T-score of 59.55, while other symptom scores hovered near the general population mean of 50. Digital health literacy was uniformly high, with median scores of 4.0 out of a possible 4 on every item, indicating strong confidence in using applications, setting up video calls, and troubleshooting basic technical problems. That fluency matters, because the study’s central question was not whether young survivors could use AI, but whether they would trust it with something as intimate as their post-cancer symptoms.

Interviews, conducted by secure video over 15 to 30 minutes and analyzed with inductive content analysis in NVivo software, revealed a consistent backdrop of unmet need. Participants described a substantial mismatch between what they needed during and after treatment and what routine clinical care could provide. Time-limited visits left little room to discuss the breadth and persistence of symptoms, and information about long-term and late effects was often too sparse to prepare them for what came next. Fatigue and cognitive difficulties, in particular, were described as invisible or minimized, leaving some survivors feeling that their concerns were never fully acknowledged. Symptoms were unpredictable, persistent, and disruptive to work, social roles, and mental health, extending well beyond the end of active treatment.

Unsurprisingly, many survivors had already gone looking for answers elsewhere. Online peer support groups, social media, and plain internet searches helped them normalize their experiences, compare symptoms, and find self-management strategies, but participants also recognized the uneven quality of advice from informal channels. Most were already familiar with AI tools such as ChatGPT, Microsoft Copilot, Google Pixel AI, and OpenEvidence for everyday tasks like drafting text, organizing schedules, summarizing information, and planning travel. That prior exposure appeared to prime them for openness to AI-based symptom support, provided the technology was designed specifically for healthcare rather than repurposed from a general-purpose assistant.

The analysis surfaced three overarching themes. The first was a demand for personalized, emotionally supportive, and credible guidance. Survivors were frustrated by generic advice that ignored their individual treatment histories, symptom trajectories, and psychosocial circumstances. They envisioned chatbots that could integrate personal health information, including cancer type, treatments received, time since treatment, and current emotional state, to deliver tailored, evidence-based recommendations. Beyond information, they wanted empathy: validating responses in clear, nontechnical language, regular check-ins, and affirming messages that could reduce isolation and self-doubt when symptoms were ambiguous or fluctuating. Credibility hinged on transparency, with participants trusting chatbots described as evidence-based, curated by healthcare professionals, and willing to cite reputable references and explain the reasoning behind their recommendations.

The second theme centered on accessible, user-centered design. Although healthcare providers remained the primary and most trusted source of care, participants valued the idea of a supplemental resource available around the clock, especially for symptoms that struck unpredictably or during spikes of anxiety. They wanted adaptable communication styles, favoring professional, factual language for interpreting clinical information and a warmer, empathetic tone for emotional support. Preferred features included mobile apps, web-based interfaces, patient portal integration, and multimodal input such as voice or image uploads. Symptom tracking, visualization of trends over time, and easy retrieval of symptom histories were seen as particularly useful for self-management and for preparing for clinic visits. Practical considerations also surfaced: free or insurance-covered tools were generally viewed as more feasible for sustained use than paid subscriptions.

The third theme was the most cautionary. Despite their openness, a substantial proportion of participants worried about reliability, privacy, and data governance. They questioned whether a chatbot could accurately interpret complex, subjective, or overlapping symptoms, and they supported escalation to healthcare teams only under strict manual control. Automated transmission of their data without explicit consent was seen as undermining autonomy and potentially triggering unwanted clinical interventions. Most were comfortable sharing general symptom and treatment information but hesitated over personally identifying data. The boundary participants drew was consistent: chatbots should validate concerns, reinforce symptom monitoring, and suggest when contacting a provider might be warranted, but the final authority over whether and when to seek care had to remain with the patient.

The authors argue that these findings position generative AI as an adjunct to survivorship care rather than a substitute for it, supplementing limited clinic time with real-time guidance and emotional reassurance between visits. The emphasis on personalization aligns with a broader digital health literature showing that tailored interventions drive greater engagement among people with cancer and chronic conditions, and the dual preference for factual accuracy and empathy echoes emerging evidence that advanced conversational AI can deliver both. The privacy concerns, meanwhile, mirror wider ethical debates about AI governance, accountability, and misinformation in healthcare. The study does have limits the authors acknowledge: the sample was small and relatively homogeneous in sex, race, ethnicity, and digital literacy, and perceptions of a technology evolving as fast as generative AI may date quickly.

Looking ahead, the researchers call for prospective, longitudinal trials evaluating chatbot interventions in young adult survivors, measuring feasibility, acceptability, symptom knowledge, self-management, and patient-reported outcomes. A key design question remains whether evidence-based symptom support should live inside widely used general-purpose AI platforms, which offer scale but risk generating inaccurate information, or in standalone, clinician-curated oncology tools, which offer oversight but demand investment in training, implementation, and privacy safeguards. Initiatives such as the Multinational Association of Supportive Care in Cancer’s free Ask, Understand chatbot, which draws on clinical practice guidelines and does not retain user responses, illustrate one path forward. If future trials confirm value, implementation research will be needed to weave these tools into clinical workflows, protecting both patient engagement and the autonomy that survivors in this study insisted upon.

Subject of Research: Young adult cancer survivors' perspectives on generative AI chatbots for post-treatment symptom management

Article Title: Exploring young adult cancer survivors’ perspectives on generative AI chatbots for symptom support

Article References: Kantabanlang, Y., Kanzawa-Lee, G., Pozzar, R. A., & Knoerl, R. (2026). Exploring young adult cancer survivors’ perspectives on generative AI chatbots for symptom support. Supportive Care in Cancer, 34(10), Article 1004. https://doi.org/10.1007/s00520-026-11254-0

Image Credits: AI Generated

DOI: 10.1007/s00520-026-11254-0

Keywords: generative AI, chatbots, cancer survivors, adolescent and young adult oncology, symptom management, survivorship care, qualitative research, digital health, patient privacy, personalization, PROMIS, University of Michigan