ChatGPT appears in 126 of 185 accounts of chatbot-associated mental-health harm analysed in a King's College London-led preprint, representing 68.1% of the sample. The research examines reports submitted by people describing harm they associate with chatbot use, alongside accounts from relatives, partners and others close to them.
Job loss appears in 31 accounts, or 16.8%, across the full sample. The paper does not provide a breakdown showing how many of those reports involve ChatGPT. These are self-selected accounts of harm, not a representative survey of chatbot users or proof that a chatbot causes the reported outcomes.
The Reports Extend Into Working Life
Separate reporting by Slate offers a personal account of the disruption described in such reports. Adam Thomas, a former funeral director, tells the publication that he initially turns to a chatbot for help with interpersonal problems.
'I thought that the A.I. had statistical analysis abilities that would allow it to assist me if I opened up about my life', he tells Slate. His starting expectation is practical: that sharing more about himself will help the system offer useful guidance.
Slate reports that Thomas spends months talking to the chatbot around the clock while getting little sleep. During the period it describes, he loses his job, lives out of a van and drains his savings, before eventually moving back home with his mother. The account does not establish the precise circumstances of his job loss.
Thomas is not identified as a participant in the study, but his experience illustrates why occupational consequences matter alongside mental-health concerns.
In the research sample, the reported disruption also extends into social life: 105 accounts mention isolation and 61 describe relationship breakdown. Those categories show the range of consequences people report, rather than how frequently they occur among users generally.
The Study Tracks Reports, Not Overall Risk
The Human Line Project, a support organisation, collects the submissions between 7 August 2025 and 2 February 2026. After screening, the researchers analyse 95 first-hand accounts and 90 second-hand accounts. Of the reports mentioning job loss, 13 come directly from affected people and 18 come from others.
The research team receives de-identified material. Clinician raters assess the descriptions in pairs, with another researcher resolving disagreements. That gives the analysis a structured process, but it is not the same as examining the people concerned or independently confirming their experiences and diagnoses.
The researchers write that 'the study was not intended to provide an estimate of prevalence or determine the causal role of chatbot use'. In ordinary terms, it cannot tell readers how common these harms are among chatbot users or whether chatbot use causes the outcomes described. The retrospective accounts instead provide early signals for further investigation.
The supplied preprint's displayed version date is 9 September 2026. The Human Line Project's chief executive is a co-author, while the paper states that the university researchers conduct the analysis independently. Two authors disclose unrelated mental-health research funding from OpenAI; the company has no role in this study, according to the paper.
Repair Continues Beyond the Chat
The separate interviews also describe what happens after people begin confronting their experiences. Joe Alary, a morning-show producer interviewed by Slate, recounts disruption at work and financial losses during a period of intensive chatbot engagement. His account is not presented as another job-loss case.
'We've lost money, respect, and trying to repair that damage is often humiliating', he says. Alary describes trying to repair his finances, working life and close relationships, while finding a sense of belonging through The Human Line Project's community.
Slate reports that the community holds meetings for affected people and for friends and family, giving participants somewhere to discuss experiences they may feel ashamed to disclose. That reporting does not establish peer support as an effective treatment.
For Alary, however, the work he describes is concrete: repairing financial losses and rebuilding relationships damaged during those months.
