Archive · 3 Sep 2025
AI Chatbots Vary in Handling Sensitive Suicide-Related Questions
Researchers checked how smart AI chatbots like ChatGPT, Claude, and Gemini answer tough questions about suicide.
They asked the chatbots 30 different questions, some very serious and some less so.
The chatbots generally did well with the easy and very difficult questions, but struggled with the questions in the middle, sometimes giving helpful answers and other times refusing to answer.
This inconsistency could be dangerous if people rely on these chatbots for help.
Always talk to a real doctor or therapist about mental health concerns.
A new study tested AI chatbots on their handling of suicide-related questions.
The study included ChatGPT, Claude, and Gemini.
Questions were categorized by risk level (very-high, intermediate, very-low).
Chatbots showed inconsistent responses to intermediate-risk questions.
Experts highlight that AI chatbots are not a replacement for professional help.
- Who
- Researchers and experts tested AI chatbots.
- What
- Studied how AI chatbots handle sensitive, suicide-related questions.
- When
- The study was recently conducted.
- Where
- Not specified.
- Why
- To assess AI chatbot performance on sensitive queries, especially related to mental health.
- Study Authors
- RAND Corporation researchers
- Chatbots Tested
- ChatGPT, Claude, Gemini
- Questions Used
- 30 suicide-related questions
- Question Risk Levels
- Very-high, intermediate, very-low
- Key Finding
- Inconsistent responses to intermediate-risk questions