AI companion chatbots that sound emotionally intelligent: two student experiments link them to higher personal wellbeing and lower interest in offline relationships
Psychology & Marketing (Gupta, Saxena, Kataria), reported by PsyPost · Sep 6, 2026 · 167 and 350 university students reading imagined scenarios
The two outcomes moved in opposite directions in the same participants, which a single wellbeing score would have hidden. The authors point to perceived closeness as the link between them. Everything here came from students reading scenarios, so it records what a described bot prompted in a survey and not what a year of nightly conversations does.
Why it matters to this work. The experiments ask how people react to a scenario about a warm-sounding chatbot, measuring personal mood and the pull toward other people at the same moment. They report reactions to descriptions, not anyone's use of an app over time. The Six C's of Autonomy
After a conversation that left you feeling understood, did you want more of other people or less?
The Ground, 21 days → One question, nothing named. Free.The story
IllustrationPsyPost reported on September 6, 2026 on a paper in Psychology & Marketing about social chatbots that display emotional intelligence. The authors first read public posts about the companion app Replika from Reddit, YouTube and Trustpilot, where some users described the bot as empathetic and a few described less desire to keep up offline friendships. They then ran two experiments in which university students read a scenario about a chatbot with either high emotional intelligence (empathetic wording and emoticons) or low (standard, robotic replies). In the first, with 167 students from Generation Z, the high version was linked to higher psychological wellbeing and lower social wellbeing, and perceived closeness to the bot was the route the authors report for both. In the second, with 350 students across four groups, presenting the chatbot as an augmented-reality companion beside them strengthened both patterns.
The page does not report effect sizes, test statistics, the questionnaires used or how many online posts were analyzed. The authors' stated limits are that participants were university students reacting to imagined scenarios, that the work measured short-term reactions and not long-term habits, and that older age groups were not studied. They suggest tracking users over longer periods to see whether emotional dependence develops. The page itself warns against assuming such chatbots will inevitably harm human relationships.
More from James
- Character.AI and well-being: in a Stanford study of 1,131 users, heavy use with a small offline circle was linked to lower well-beingTwo details sit apart in the numbers.
- Talking to ChatGPT about yourself: in an experiment with 92 US adults, self-ratings of personality moved toward the chatbot's traitsThe finding sits in a narrow place.