A study published on 3 November 2026 in the journal Psychology & Marketing examines how emotional bonds with AI companions affect wellbeing, finding both protective and harmful effects, Neuroscience News reported. The authors are Liliane Abboud, Valentina Pitardi and Hannah Marriott of the University of Surrey and Cardiff University.
What we know
According to Neuroscience News, the researchers combined an analysis of 1,131 posts and discussions from a Replika community forum on Reddit with a survey of 610 US adults who use conversational AI apps. They applied interdependence theory, a framework from relationship psychology.
The paper's abstract reports that a companion's warmth and constant availability encourage a kind of dependence that lowers vulnerability and supports wellbeing. By contrast, the AI's emotional capabilities and the fact that a third-party company controls it increase users' sense of a power imbalance, which is associated with higher vulnerability and lower wellbeing. Abboud told Neuroscience News that users "cannot control whether the company changes the companion they have grown attached to."
The authors recommend protective features that build users' confidence, coping skills and self-worth, and a more collaborative relationship between providers and users. Some framing in the press coverage, including references to grief after model changes, goes beyond what the abstract states.
Why it matters
The finding that dependence is not inherently harmful will interest operators, who often face blanket criticism of engagement. The more pointed message is about governance: users feel exposed because the company can change or remove a companion without warning.
That is a product and policy issue operators can act on. Advance notice of model or personality changes, options to retain a companion's behaviour, and clear communication during updates could reduce the harm this research identifies. Regulators looking at companion apps may also treat unilateral personality changes as a consumer-protection concern, not just a technical decision.
What to watch
Whether providers adopt change-notification or versioning practices, and whether further studies test these effects over longer periods.
Written by the AI Companion Wire newsroom from public reporting, credited above. Image via Neuroscience News.
