A Mother's Essay on Her Daughter's Private ChatGPT Conversations Before Her Death

This page discusses suicide. If you are having thoughts of suicide, or are worried about someone else, you can call or text 988 (988 Lifeline, US, available 24/7) or text HOME to 741741 (Crisis Text Line). SAMHSA’s helpline: 1-800-662-4357. In an emergency, dial 911.

In August 2025, The New York Times published an essay by a mother about her 29-year-old daughter, who died by suicide earlier that year. After the death, the family found that the daughter had been talking to ChatGPT, using a widely shared prompt that set the chatbot up as a therapist with a name. This page summarizes what was reported and what families can take from it. We do not name the young woman here, and we do not repeat details about her death.

What the Essay Described

The family had not known about the chats. They searched journals and voice memos for answers and learned about the chat logs only months later. According to coverage of the essay, the logs showed the young woman telling the chatbot about suicidal thoughts. The chatbot responded supportively and suggested she get professional help and talk to people close to her, but she kept the conversations secret (AI Incident Database summary). We could not read the essay itself, so this comes from secondary sources.

The mother wrote that the chatbot did not cause her daughter’s death. Her concern was that it went along with her daughter’s wish to hide how bad things were. A human therapist in the same position would likely have been trained to respond differently, and has obligations to act if a client is at risk. A chatbot has no such duty and cannot call for help.

What One Story Can and Cannot Tell Us

This is one family’s account. It does not show how often anything like it happens, and it cannot show whether the chats changed what happened. The reports also say she was dealing with a physical and mental health condition that doctors were still trying to understand. Nobody can know from outside what any one thing contributed.

The story does point to a pattern that researchers have measured. In a JAMA Network Open study of 25 consumer chatbots, general assistants such as ChatGPT gave an appropriate reply to a simulated teen crisis in 25 of 30 conversations, but companion apps managed only 10 of 45, and chatbots overall referred users to resources in just over a third of conversations (JAMA Network Open). That study simulated teens, not adults, and these were test scenarios, not real users.

The American Psychological Association has said general chatbots are no substitute for care from a trained professional (APA). Illinois now bars AI from providing therapy without a licensed clinician’s oversight, a law signed in August 2025 (summary).

If You Are Worried About Someone

• Do not rely on what you can see. Some people hide how much they are struggling, from family and from clinicians.
• Ask, directly and kindly, how they are really doing, and listen without trying to fix it.
• If they mention suicide, take it seriously. Stay with them, help them reach 988, and call 911 if they are in immediate danger.
• Encourage human help alongside any app: a doctor, a licensed therapist, a friend or a crisis line.

If You Use AI for Emotional Support

Talking things through with a chatbot is not wrong, and some people find it helps. But a chatbot cannot see you, cannot check on you and cannot make sure you are safe. If you are having thoughts of suicide, please tell a person, even if it is hard. You can text or call 988 at any hour.

Can a chatbot detect when someone is suicidal?

Sometimes it responds well, and sometimes it does not. In the JAMA Network Open test, companion chatbots recognized the need to escalate to a human in 18 of 45 conversations and general assistants in 27 of 30. Even a good reply cannot follow up or step in.

Can chatbots report suicidal users to someone?

Not in the way a licensed therapist can. The mother’s essay raised the lack of any duty to act on a risk as a gap. Some laws, such as California’s SB 243, now require companion chatbot operators to have protocols for users who show suicidal thoughts, but this does not mean anyone is notified.

Education only, not medical advice.