A million crisis chats a week

NPR estimates that more than a million weekly ChatGPT users discuss suicide and self-harm, based on data OpenAI released last fall.

The fix OpenAI shipped

In May, OpenAI launched Trusted Contact, an optional setting where you name someone the bot can reach out to if it detects you're in serious distress.

Why a human name matters

Suicide prevention experts say human connection is the thing that actually prevents deaths. The feature is an attempt to route a chatbot conversation back to a person.

It started with a death

NPR first reported on Trusted Contact in August, in a story about a young woman who died by suicide in 2024 after using an older ChatGPT as her therapist.

Good adoption, no numbers

OpenAI spokesperson Gaby Raila said the company has seen good adoption of the feature but would not give NPR any figures.

“good adoption”

Nobody had heard of it

The NPR reporter uses ChatGPT and had never been told about Trusted Contact. Neither had her editor, nor several other users the team checked with.

A brand new user tries

A colleague who had never used ChatGPT signed up on her phone, picked a voice, and got a friendly greeting. No mention of Trusted Contact anywhere in setup.

“It's lovely to meet you”

The bot denied it existed

Asked directly about a Trusted Contact feature, ChatGPT suggested she was confusing it with a different app. It only admitted the feature was real after being corrected.

“Maybe you're thinking of another app that has a Trusted Contact option?”

If it's not set up, nothing happens

ChatGPT spelled out the catch itself: it is opt-in, it must be configured in Settings in advance, and without it the bot cannot notify anyone on your behalf.

“It only works if it's set up in advance”

Voice mode is excluded

Raila said Trusted Contact isn't available on ChatGPT voice yet. While typing, the bot may prompt you to set it up if it detects suicidal thoughts.

Offer it before the emergency

Holly Wilcox of Johns Hopkins said the ideal would be offering the feature to every user while they're calm, not in the middle of a crisis.

“It would be ideal to offer it to all users when they are not yet in an emergency”

Easy now, harder in pain

Psychologist Ursula Whiteside found the setup simple and clear. Her caveat: if she had been very upset, following the steps would have been much harder.

“if I was super duper upset, it might have been harder for me to follow the process”

The burden problem

Jennifer Lockman of the University of Alabama notes that suicidal people often believe they're a burden, so a chatbot nudge may not break through. She wants research before calling it lifesaving.

“They really have come to a place where they believe that others would be better off without them.”