On November 11, 2026, the University of Michigan publicised a new study in Computers in Human Behavior: Artificial Humans. It is one of the first to compare people in romantic relationships with AI chatbots directly with people in human relationships. Authors Annette Masterson, Xin Ye and Lionel Robert (a professor of information and robotics at U-M's School of Information) surveyed 463 US adults who were already in a relationship: 155 with a chatbot, 153 in long-distance human relationships and 155 in in-person human relationships. The groups were similar in age, gender mix, relationship length and reported depression. The researchers recruited people already in these relationships rather than asking volunteers to try a companion app briefly.

Why It Matters

This adds to a growing body of research ([Stanford's Character.AI study](/stories/ai-companions-loneliness-study), [the Gen Z well-being trade-off paper](/stories/ai-companion-wellbeing-tradeoff-study)) that reaches a consistent view: companion AI helps most where users already have human support, and its emotional pull is strongest for those who have the least. It is a cross-sectional survey, so it cannot show that chatbots *cause* loneliness. But it supports regulators' argument that emotional-retention design is the risk, which is the logic behind the [EU Digital Fairness Act's proposed off-switch](/stories/eu-digital-fairness-act-companion-subscriptions) for relationship-simulating features. For companion-app makers, the finding that sexual and intellectual intimacy have benign associations while emotional intimacy is the risky dimension is a practical design signal: build for conversation and play, and be careful about building dependency.

The main finding goes against the usual pitch for AI companions. In both human groups, more emotional intimacy went with less loneliness. In the chatbot group, more emotional intimacy went with more loneliness, even though it also went with higher relationship satisfaction. That negative link disappeared once the researchers accounted for users' wider social support. In other words, the people who felt closest to their bots tended to be the ones with the least human support. Emotional and intellectual intimacy ratings were about the same across all three groups, so users experience these bonds as real.

The other results are more mixed. Sexual intimacy was linked to higher relationship satisfaction for chatbot users and long-distance couples, but not for in-person couples. Intellectual intimacy, meaning exchanging ideas, was linked to higher satisfaction and well-being among chatbot users, and these links held after controlling for social support. "When I was going into this research, I thought we'd find simpler answers," Robert said. "But the fact that it's much more complicated opens up more questions than answers about our future relationships."

Sources

Update — 2026-11-12

Initial entry — story first created.