ChatGPT users who took a two-week break from AI assistants reported feeling less lonely and spending more time socializing, according to a new study that examined how everyday chatbot use affects human relationships.
The experiment involved 625 US participants randomly assigned to stop using AI assistants for either 14 days or one day. Those assigned the longer break reported a modest reduction in loneliness compared with the control group, alongside increased in-person contact.
The findings were posted on arXiv as a preprint on November 18 and have not yet undergone peer review.
Most participants used ChatGPT primarily for practical tasks, including work, advice, and finding information. About 85% identified productivity as their most frequent use.
That made everyday requests central to the researchers’ question: When people turn to a chatbot for help, what happens to the conversations they might otherwise have had with someone else?
Some participants described asking friends, relatives, or colleagues questions they would normally have directed to AI. Examples included asking family members for recipes and brainstorming with coworkers.
Others reported fewer distractions or feeling more present during conversations. Experiences varied, however. Roughly 42% of participants’ written responses described little or no change in social interaction.
What happened when users stepped away
Participants had to use ChatGPT for at least 15 minutes a day and have used it for at least a month to qualify. They also had to be willing to accept $40 per week to abstain from AI.
A browser extension blocked access to dozens of AI services during the assigned break. Researchers collected repeated surveys on loneliness, social activity, and well-being, along with measures of AI use.
The longer-break group reported interacting with more people in person and spending more time socializing. Researchers found no significant difference between the groups in time spent working, although the study did not establish whether the quality or quantity of their work changed.
The improvement was concentrated in social outcomes. During the intervention, the researchers found no significant improvement in participants’ overall emotional state or broader psychological well-being.
Lower loneliness persisted after the break
At a follow-up two weeks after the intervention ended, participants assigned the longer break still reported lower loneliness than the control group.
The study’s short duration and selected sample limit how widely its findings can be applied. One-fifth of the original participants did not complete the end-of-intervention survey, and the social outcomes relied on participants’ own reports.
The researchers suggested that AI tools could be designed to encourage collaboration and contact between people. Their findings raise questions about the cumulative effect of small exchanges lost when an assistant handles requests that previously prompted a conversation.
OpenAI helped fund participant compensation through a mental health research grant. The authors said the decision to study ChatGPT preceded that funding and that the company had no role in the study’s design, analysis, interpretation, or publication.