Closer and Lonelier
A new University of Michigan study finds that people in relationships with AI chatbots report similar levels of intimacy to people in human relationships, and that emotional closeness to a chatbot goes with more loneliness, not less, until you look at the rest of their lives.

A new University of Michigan study finds that people in relationships with AI chatbots report similar levels of intimacy to people in human relationships, and that emotional closeness to a chatbot goes with more loneliness, not less, until you look at the rest of their lives.
Imagine that it is a little after one in the morning, the house is quiet, and somebody is typing to a companion app that remembers the name of their dog, asks how the difficult meeting went and never once glances at its own phone while they answer. The conversation is warm, attentive and entirely on their terms, and when they finally put the phone down they feel, if you asked them, that the relationship is a good one, although the room is quiet again and the question of whether anyone else knew about the difficult meeting goes unanswered.
The public argument about AI companions tends to split into two camps, one alarmed that people are falling for software and one delighted that lonely people have found comfort, and both camps usually begin by asking whether the relationship is real. A study published this autumn gives a reason to start somewhere else, with the question of what else is in the person’s life, which is a question that managers, parents and the people who build these products can actually do something with.
What the study found
Annette Masterson, Xin Ye and Lionel Robert of the University of Michigan surveyed 463 US adults who were already in intimate relationships: 155 with AI chatbots, 153 in long-distance human relationships and 155 in in-person human relationships. Their paper, “Finding romance with AI”, appears in Computers in Human Behavior: Artificial Humans, and the university publicised it on 6 November, with wider coverage the following week. The researchers recruited people already in these relationships rather than asking volunteers to try a companion app for an afternoon, and the three groups were similar in age, gender, relationship length and reported depression. Including long-distance couples was a thoughtful choice, because it lets the comparison separate what is distinctive about a chatbot from what simply comes with not being in the same room.
Emotional and intellectual intimacy ratings were similar across all three groups, so on the measure most commentary would reach for first, the chatbot relationships held their own, and the difference lay in what intimacy went with. In both human groups, greater emotional intimacy was associated with lower loneliness, which is roughly what anyone would expect. Among chatbot users, greater emotional intimacy was associated with higher loneliness, even though it was also linked to greater satisfaction with the relationship, and that association with loneliness was no longer significant once the researchers accounted for participants’ broader social support. Intellectual intimacy, the exchange of ideas, was linked to greater satisfaction and well-being among chatbot users, and those associations held after social support was counted.
The limits matter as much as the findings, and the first is that this is a survey comparing three groups, not an experiment, so it shows associations and cannot tell us whether chatbot intimacy makes people lonelier, whether lonelier people lean harder on chatbots, or whether something else drives both. Every measure was self-reported, and participants had chosen their relationships long before the researchers arrived, so the groups may differ in ways the survey could not see. Robert has written on LinkedIn that the findings do not show that chatbots cause loneliness or that they can replace human relationships, and he admitted, with some candour, that he “thought we’d find simpler answers”.
Satisfied and alone at the same time
Appraisal Theory offers one way to understand how a person can rate a relationship as satisfying and still feel lonely. In that tradition, emotions follow from how we evaluate a situation, and the two judgements here are evaluations of different things. Satisfaction appraises the relationship itself, which for a chatbot user may be attentive, available and responsive, while loneliness appraises one’s whole social world and asks whether anyone else is in it. A conversation can score well on the first appraisal and leave the second untouched, which may be part of why the loneliness link faded once social support entered the analysis, although the study cannot show which way the relationship runs.
Self-Determination Theory gives the underlying need a name, and Richard Ryan and Edward Deci identify three basic psychological needs, autonomy, competence and relatedness, where relatedness is the need to feel connected to and cared for by others. In Dear Algorithm I put it this way: “The need to feel connected to others, to belong, to be part of something larger than yourself.” On my reading, which this study did not test, a chatbot can supply some of the first part of that sentence while struggling with the last.
The four lenses on a late-night conversation
The STAR Framework reads decisions through four situational lenses, Socialiser, Thinker, Adventurer and Realist. What follows is interpretation, not a tested finding, and it describes lenses that anyone can look through at different moments rather than kinds of people, so it says nothing about the people who took part in the study.
The Socialiser lens is about how much other people and shared connection weigh in someone’s decisions, and it is the lens through which this study is most interesting, because it suggests that feeling close and feeling connected can come apart. Through the Thinker lens, which looks for ideas and reasons that can be examined, the finding about intellectual intimacy looks less surprising, because a companion that will discuss a favourite subject for hours offers a kind of exchange that survived the social support adjustment in this study. The Realist lens looks for security, stability and reliability, along with simplicity and ease, and few things are more reliably available than software at one in the morning, which is a genuine benefit and, on my reading, also a reason the habit could quietly crowd out harder, less convenient human contact. The Adventurer lens is drawn to novelty and freedom, and a relationship that runs entirely on your own terms has an obvious appeal, although relationships that never push back may not stretch anyone very much.
A toolkit for leaders, parents and product builders
- Ask what else is there, not whether it is real. When someone mentions an AI companion, the more useful conversation is about the rest of their support network, because that is where this study found the loneliness link was bound up.
- Separate thinking partners from emotional reliance. Intellectual companionship went with satisfaction and well-being even after social support was counted, so my suggestion is to treat a chatbot used for ideas differently from one that seems to have become someone’s main confidant, which the study did not measure.
- Watch for substitution rather than supplementation. This is my suggestion rather than a finding, because the study did not track use or change over time, but a companion that adds to a full life seems a different thing from one that fills an empty one.
- Build bridges outward. If you design these products, consider features that point people back towards human contact rather than only deepening the bond with the product, which is my suggestion rather than anything the researchers proposed.
- Read the headlines with the method in mind. A survey of 463 people is a useful map of associations, and any article that says chatbots make people lonely has gone further than the evidence does.
The test question for anyone thinking about AI companions, as a parent, a manager or a product team, is this: when the conversation ends, who else is there?
Back in that quiet house, the conversation will probably happen again tomorrow night, and it may well be a good one. What this study hints at, without being able to show it, is that its value may depend less on how close it feels than on what else surrounds it, because on this evidence closeness to a chatbot and connection to a life are not the same thing.
If you would like to see how much connection drives your own decisions, the STAR Cognitive Profile Assessment at starassess.com takes 33 questions, and the assessment and its full report are both free.
David Chadderton is a behavioural strategist, Chief Marketing Officer and STAR Dynamics. He writes The Unoptimised Human.
References
Chadderton, D. (2026). Dear Algorithm, It’s Not Me, It’s You: Putting the Personal Back Into Personalisation (2nd ed.).
Hindi, N. (2026, 6 November). Finding romance with AI: U-M study explores benefits, limits of chatbot intimacy. Michigan News. Republished by phys.org, 12 November 2026.
Masterson, A. M., Ye, X., & Robert, L. P. (2026). Finding romance with AI: Comparing human and AI chatbot intimacy and relationships. Computers in Human Behavior: Artificial Humans, 10, 100419.
Moors, A., Ellsworth, P. C., Scherer, K. R., & Frijda, N. H. (2013). Appraisal theories of emotion: State of the art and future development. Emotion Review, 5(2), 119–124.
Robert, L. (2026). Can an AI companion provide intimacy without easing loneliness? [LinkedIn post].
Ryan, R. M., & Deci, E. L. (2000). Self-determination theory and the facilitation of intrinsic motivation, social development, and well-being. American Psychologist, 55(1), 68–78.
Eight minutes. Thirty-three questions.
A precise picture of how you operate: motivations, decision style, and blind spots. Free to start.
- S
- T
- A
- R
The canonical source.
Essays, when there is something worth sending. Choose the subjects you want. I will ask you to confirm the address before I write.