Researchers say long-term effects of chatbot use on isolation are unknown

Three Cornell researchers, Qian Yang, Ned Cooper and Ibrahim Emara, argue in a STAT First Opinion piece published Oct. 26, 2026 that no one yet knows whether months of chatbot use gradually pull people away from friends and family. Their core point is that a chatbot "could be safe in every single conversation" and still, over time, displace the human relationships that keep people healthy. They say the data to answer the question sits largely with AI companies and is not yet available to independent researchers.

The authors note that current safeguards target acute harm. OpenAI recently said it is tracking signs of user distress across sessions, not only within them. Other measures across the industry include parental controls, usage limits, break reminders and crisis referral. According to the piece, a dozen states now require companion chatbot operators to disclose that their tools are not human and to set up crisis referral procedures. The authors argue that gradual withdrawal from human relationships may show few signs of distress, so these tools may not catch it, and that a user may find each session helpful.

The evidence is thin, and the authors say so. They cite early studies suggesting companionship and emotional support use is concentrated among a subset of heavy users, and that heavier use is associated with reduced socialization and increased loneliness. They describe this evidence as preliminary and debated. This is an argument for measurement, not a finding of harm.

For health organizations, the relevance is direct. Patient-facing chatbots for scheduling, triage, follow-up and support are becoming part of the patient experience, and the piece names primary care utilization and social prescription programs among the public costs of isolation. The authors also warn that per-chat disclosures such as "not talking to a human" could later be cited as a defense if accountability for cumulative harm arrives, which means a disclosure label alone is a weak posture for a care organization whose purpose is connection.

The move for the Chief AI Officer, patient experience and marketing leaders: review every patient-facing chatbot you run or plan to run. Build a clear, easy handoff to a human, and instrument repeat use so you can see which patients return often and whether they also reach your staff. That gives your teams an early signal, keeps the bot as a bridge to people rather than a substitute, and puts you ahead of the measurement question the researchers are raising.

You are a patient experience and AI operations advisor to the executive team of [your organization], a [type of health or wellness organization].

Context: In a STAT First Opinion piece published Oct. 26, 2026 (STAT Health Tech, "Opinion: Could long-term AI use lead to isolation? We need to find out"), Cornell researchers Qian Yang, Ned Cooper and Ibrahim Emara argue that no one yet knows whether months of chatbot use gradually pull people away from friends and family. They say a chatbot "could be safe in every single conversation" and still displace human relationships over time. Current safeguards target acute harm. According to the piece, a dozen states now require companion chatbot operators to disclose that their tools are not human and to set up crisis referral procedures. The authors warn that per-chat disclosures alone are a weak posture. The evidence is preliminary and debated. This is an argument for measurement, not a finding of harm.

Task: Review each patient-facing chatbot we run or plan to run. For each, design a clear, easy handoff to a human and a way to measure repeat use, so we can see which patients return often and whether they also reach our staff. Aim for something the team can finish this week.

Output: A draft workflow for the executive team, one section per chatbot, listing the handoff trigger and path, the repeat use measure, the owner for each step, and the checks that confirm it works. End with open questions.

My inputs: [list of patient-facing chatbots and what each does], [your current human handoff path for each], [the team or person who owns each bot], [the analytics or reporting tool you use], [the staff who would receive handoffs].

Guardrails: Use only the facts above and the inputs I fill in. Cite STAT Health Tech by name for any claim from the piece. Flag anything uncertain or that needs verifying, including state requirements that apply to us. Do not request or include patient data. Nothing here is medical or legal advice. Label this a draft for the team to check.