I can confirm this, caught my father using ChatGPT as a therapist a few months ago.

The chats were heartbreaking, from the logs you could really tell he was fully anthropomorphizing it and was visibly upset when I asked him about it.

They don't even have to be "cooked", people generally are pretty similar which is why common scams works so well at a large scale.

All AI has to be is mildly but not overly sycophantic and as a supporter/cheerleader to someone, or who affirms your beliefs. Most people like that quality in a partner or friend. I actually want to recognize OAI courage in deprecating 4 because of it sycophancy. Generally I don't think getting people addicted to flattery or model personalities is good

Several times I've had people speak about interpersonal arguments and them having felt vindication when chatgpt takes their side, I cringe but it's not my place to tell them chatgpt is meant to be mostly agreeable.