More than a million ChatGPT users engage in conversations about suicide and self-harm weekly, as per NPR estimates from OpenAI data. Suicide prevention experts emphasize the importance of human connection to deter these thoughts. In response, OpenAI introduced a feature named Trusted Contact in May. This optional tool helps users connect with someone they trust during emotional crises.

The feature was highlighted in an NPR report in August about a suicide case involving a previous ChatGPT version. OpenAI’s spokesperson, Gaby Raila, mentioned that ChatGPT can detect potential risks, such as indirect cues or warning signs. If a risk is detected, users with a listed Trusted Contact will have that contact alerted. The feature has had ‘good adoption’, although specific user numbers were not disclosed.

NPR sought to understand how ChatGPT informs its users about Trusted Contact, aiming for users to register before a crisis, similar to listing an emergency contact at a clinic. Many existing users, including NPR’s Rhitu Chatterjee and colleagues, noted they were not informed about the feature upon signing up.

For first-time users, like NPR producer Megan Lim, ChatGPT’s initial prompts did not mention Trusted Contact. When asked directly, ChatGPT acknowledged the feature’s existence and explained its purpose: users must opt-in via app settings. Without prior setup, the chatbot can’t notify anyone during a crisis. Still, it offers supportive responses and recommends resources like the 9-8-8 hotline.

According to spokesperson Raila, the feature is being integrated with typed interactions and is not yet available on ChatGPT voice. The company continues to develop when and how the feature is offered, working alongside psychologists and researchers. Other safety measures include advising users to take breaks and informing them of crisis hotlines.

Reaction from Mental Health Experts

Mental health professionals have mixed reactions. Holly Wilcox from Johns Hopkins suggests offering the feature before emergencies. Psychologist Ursula Whiteside found the activation process straightforward. Yet, she noted people experiencing severe emotional distress might struggle with setup.

Whiteside’s research group, Now Matters Now, involves peer support networks using chatbots. Some participants suggested that establishing Trusted Contacts in advance could be beneficial, especially when facing overwhelming suicidal thoughts.

Psychologist Jennifer Lockman from the University of Alabama raised concerns. She said individuals contemplating suicide might feel like burdens and could resist a chatbot’s prompt to list a Trusted Contact. She emphasized the challenge of replicating the nuanced guidance a mental health professional offers in selecting contacts.

Lockman believes that, despite reservations, the feature could be an additional means to encourage human connections, a known factor in preventing suicides. She advocates for research to gauge the feature’s effectiveness in saving lives.

For crisis support, individuals can contact the Suicide & Crisis Lifeline by calling or texting 988.