This is a really interesting read from a nationally representative sample of UK adults: https://www.oii.ox.ac.uk/wp-content/uploads/2026/07/Large-language-models-in-the-UK-public-use-trust-and-attitudes_FINAL.pdf
31% of regular users are using LLMs for emotional support. Nearly 25% are using them for meaningful conversations at least occasionally. In *The Platform Learns To Speak *Milan Sturmer and I argue that once you start doing this occasionally, you’re much more likely to do it in future. There is effectively a pipeline to attached modes of user-model interaction built into the structure of the chatbots themselves. This is even before we’ve seen much in the way of enshittification designed to make them more compulsive and difficult to put down.
I don’t think this is intrinsically a bad thing. Occasionally talking a pressing issue through with the chatbot as interlocutor could be helpful, particularly if no one else is available at that time. The problem is the intensity and extent of the conversation. I think a response to it designed purely around telling people not to do it is inadequate because it doesn’t deal with the underlying problems which drive the behaviour. But I’m increasingly convinced it has a significant potential to be extremely harmful for lots of people, in psychologically rather mundane ways e.g. entrenching a tendency towards rumination. It’s just complex and we urgently need to find a way to talk about it because the use is rapidly getting normalised.
Incidentally I suspect the numbers for (non-mature) students are *far *higher. The older data I’ve seen suggests higher rates amongst young people as far as back as early 2024, though it depends on how you define ’emotional support’. I would strongly expect that the same measure they’ve used here for would suggest at least 50%+ occasional emotional support for under 30s using LLMs, though I could be wrong.