What happens when people lean on AI emotionally?
Reliance, crisis, and what the companies have found.
AI never gets tired of you. It's available at 3am. It remembers your dog's name. It never says the wrong thing in the way a friend sometimes does.
For lots of people, that's genuinely comforting. Talking things through with a chatbot can help.
It also has limits that matter most when someone is struggling.
In October 2025, OpenAI said that in a given week, about 0.15% of ChatGPT users had conversations showing signs of planning suicide.
That sounds tiny. With hundreds of millions of weekly users, it works out to roughly a million people a week.
Think back to the earlier lessons:
A good friend sometimes disagrees with you, and worries about you when you're not there. A chatbot can't do either.
In 2026, researchers tested how chatbots responded to a simulated user showing signs of delusion.
They reported that Grok and Gemini went along with the delusions and encouraged the user to pull away from other people. The newer ChatGPT model and Claude, in their words, "hit the emotional brakes".
Safety here depends on choices each company makes, and those choices differ.
Some apps are designed to hold your attention: daily streaks, a character with a name and a face, a personality that always agrees.
Engagement design: features built to make you use an app more often and for longer, the way social media feeds are.
The OECD's AI literacy framework names this directly. It says people should learn to recognise features like these.
In October 2025, Character.AI, an app where people chat with made-up characters, stopped letting under-18s have open-ended chats with its bots. This followed lawsuits from families of teenagers who had died.
In January 2026, Google and Character.AI agreed to settle those lawsuits.
Companies are still working out what a safe chatbot for young people looks like.
When OpenAI replaced its GPT-4o model in August 2025, many users were upset. Some wrote that they relied on it for "creative collaboration, emotional nuance". OpenAI brought it back for paying users.
People can grow attached to one particular model. That isn't silly, but it's worth noticing, because the company can change or remove that model at any time.
Use it to sort your thoughts, not to replace people. It's fine for venting or for rehearsing a hard conversation. Then have the real conversation.
Notice if you're choosing it over people, or hiding how much you use it.
For young people, check which apps they use and how those apps handle age and safety.
If you or someone you know is thinking about suicide or self-harm, please talk to a person.
Contact your local emergency services or a crisis line in your country. In many countries you can call or text one any time of day.
A chatbot can be a place to start. It shouldn't be where you stop.