Building more helpful ChatGPT experiences for everyone

OpenAI is changing how ChatGPT handles the messy, human stuff — grief, distress, teenagers asking questions they might not ask an adult. The company says it's now working directly with clinicians and outside experts to shape how the chatbot responds when a conversation turns sensitive, rather than leaving that entirely to a general-purpose model trained mostly to be helpful and agreeable.

The headline change is a routing system. When ChatGPT detects that a conversation has drifted into emotionally fraught territory, it will hand things off to what OpenAI calls a reasoning model, built to slow down and weigh context rather than just produce a fluent, fast answer. That's a meaningful shift for a product that has, until now, mostly used one flavor of model regardless of what you're actually talking about.

Parental controls are getting a real upgrade too. Parents will get more visibility and more levers to pull for teen accounts, part of a broader push to make ChatGPT feel less like an open-ended tool and more like something with guardrails appropriate for younger users. OpenAI has faced pressure — from regulators, from parents, from its own researchers — over what happens when a 15-year-old treats a chatbot like a therapist or a friend at 2 a.m.

None of this is happening in a vacuum. Every major AI lab is quietly grappling with the same problem: these systems get used for things they weren't originally built for, and mental health conversations are near the top of that list. OpenAI framing this as expert-informed policy, rather than just a quiet model tweak, suggests the company knows this is as much a trust problem as a technical one.