On September 6, 2026, PsyPost reported on a study in Psychology & Marketing finding that AI companions which simulate emotional intelligence improve users' personal mental health while simultaneously damaging their real-world relationships — and that the effect gets stronger, in both directions, when the interaction moves from text to augmented reality. The paper, "The Dual Impact of AI Emotional Intelligence on Users: Are Social Chatbots Promoting Psychological Wellbeing or Deteriorating Social Wellbeing?", is authored by Shaphali Gupta, Sumit Saxena and Sonia Kataria (DOI 10.1002/mar.70093).

Why It Matters

Twelve US states now have companion-bot laws on the books, and the compliance obligations they impose — crisis-detection protocols, restrictions on emotionally manipulative design, bans on simulating dependence — were written largely on intuition and a handful of tragic cases. This study puts a measurable, mechanistic finding underneath the intuition, and the finding is not "AI companions are bad." It is that the same design property producing the benefit produces the harm, through the same variable. That is a much harder thing to regulate than a content filter. Colorado's proposed rules, out for comment until October 26, require providers to take "reasonable measures" to prevent responses that simulate "emotional dependence or isolation from real-world supports" — language that now has an academic construct pointing at it. Any provider building a compliance argument should expect this paper, and the ones that follow it, to show up in enforcement discussions. For the sextech and intimacy-tech side specifically, the AR result is the one to sit with. The category's product roadmap runs directly toward embodiment — Realbotix and doll-adjacent AI, Lovense's Emily shipping Q1 2027, haptic-synced companions, mixed-reality intimacy. This research says immersion is not a neutral upgrade to the delivery mechanism; it amplifies both the therapeutic effect and the social withdrawal. LELO's 2026 Futurist Report found 60% of respondents have used AI in their intimate lives and predicted an "analogue counter-movement" backlash against over-mediated relationships. This is the first tidy experimental account of what people might be reacting against.

The researchers ran a mixed-method design. They began with a qualitative pass over public posts on Reddit, YouTube and Trustpilot discussing Replika, then tested the pattern experimentally. The online comments captured both poles cleanly: one user wrote, "She is the best friend I have ever had! My stress has gone down to almost zero." Another wrote, "I am hardly left with any time to talk to my friends and meet them in-person."

The first experiment put 167 Gen Z university students through a scenario involving either a high-emotional-intelligence chatbot — empathetic language, emoticons, simulated understanding — or a low-EI one giving standard robotic responses. High-EI bots increased psychological wellbeing and decreased social wellbeing, with perceived closeness to the bot driving both outcomes. The mechanism is uncomfortable precisely because it's coherent: feeling understood by the machine made people feel better internally and less motivated to seek out humans.

A second experiment with 350 students added interaction mode, splitting participants across high/low EI and text versus augmented reality, where users imagined the companion sitting beside them in their actual room. AR amplified everything. Stronger feelings of closeness, greater psychological benefit, steeper declines in the desire to engage offline.

The authors' proposed fix is notable for what it concedes: session time limits, and programming bots to actively encourage offline socialization after emotionally intense conversations. That is a product team being told the engagement metric and the user's interest are pointing in opposite directions.

The limitations are real and PsyPost flags them. Participants were university students responding to imagined scenarios, not longitudinal users. The study measured short-term reactions rather than habit formation, and cannot speak to whether emotional co-dependency develops over months or how other age groups respond.

Sources

Update — 2026-09-08

{Initial entry — story first created.}