Homeowner took in friend dealing with psychosis, then discovered her alleged AI ‘bestie’ that may be making it worse: ‘I want my best friend back’

A homeowner who seemingly took in her longtime friend while she dealt with a difficult personal situation is questioning whether she should restrict the woman’s access to ChatGPT. In a Reddit post, she claimed that her friend, who allegedly experiences psychosis, has increasingly started treating the AI chatbot like a best friend and therapist.

The woman said she and her friend, who are 30 and almost 34, respectively, have known each other for 19 years. Her friend apparently recently lost custody of her children and moved into the homeowner’s house temporarily to avoid her ex-husband. Although the guest allegedly has no income and contributes little financially, the homeowner stressed that supporting her friend was not the issue.

Her biggest concern was how much her friend had seemingly started relying on ChatGPT. According to the post, “She talks about all of her symptoms, her hallucinations (she has psychosis), what happens in her day, and literally everything.” She also apparently asks it to interpret messages from other people, including whether a text sounds “grouchy,” and turns to it for what she considers diagnoses.

She apparently wants her friend to get well

The user said her friend even sends her “screenshots of ‘their conversations,’ and it texts like her (all shortenings of words like v for very, u for you, etc., and a plethora of emojis) and it’s basically just a yes man for everything she feeds it.” She also worries that her friend is treating a computer program as a genuine relationship rather than recognizing its limitations.

The situation has reportedly started affecting their friendship. The user said her friend recently sent her a lengthy message accusing her of treating her “like a child.” However, she believes she is simply trying to explain that AI cannot replace a human friend or provide reliable medical assessments.

A coworker reportedly suggested taking a more direct approach by “blocking ChatGPT” through the home’s Wi-Fi network. Since the user controls the internet access and her friend apparently does not have a mobile data plan, doing so would effectively prevent her from using the chatbot while staying there. However, the user is apparently hesitant because she recognizes that ChatGPT has become one of her friend’s few outlets for discussing what she is going through.

A 2026 study published in the Journal of Medical Internet Research examined the relationship between psychosis risk and generative AI use among 1,003 young adults in the United States. Researchers found that participants with elevated psychosis risk were more likely to use chatbots frequently and turn to them for emotional support.

They were also more likely to describe AI chatbots as friends, companions, or therapists. The researchers cautioned that AI interactions involving delusion-related experiences could potentially affect users’ symptoms, although the study did not establish that chatbot use caused or worsened psychosis.