People Are Using ChatGPT as a Therapist — Until an Update Breaks It

A team of researchers spent eight months — April to December 2025 — studying how 18 ordinary Americans use everyday chatbots like ChatGPT, Claude and Gemini for emotional support, not just for homework or emails. They ran interviews, a four-week diary study where people logged their chats, group discussions, and final interviews. The headline finding: this kind of comfort-seeking rarely starts on purpose. People asked about a recipe or a work problem, felt understood, and slowly started sharing more personal things — especially when friends, family or a therapist weren't available.

Over time, users built routines. Some chatted every night before bed. Others set their own ground rules: no talking about their ex, no asking for medical advice, keep it to 20 minutes. These habits gave them real comfort and a sense of being heard. But the study found those routines were fragile. When the company quietly changed the AI's personality or responses with a software update, people felt like a friend had suddenly gone cold. Reading news stories about AI harms made them second-guess themselves. Big life changes — a breakup, a move, a new job — also shifted what they needed.

The important insight is that these relationships don't exist in a vacuum. They sit inside a person's whole support network, history and daily life. The researchers argue that tech companies should think about this before pushing out updates, because a small tweak can accidentally remove something someone was leaning on. For now, the paper doesn't say AI therapy is good or bad. It says the experience is real, it's growing, and it deserves more careful attention than a simple on-off switch.