Michael is an engineer whose company embraced the AI wave. He enjoyed testing ideas with chatbots. Then his manager became worried: “He literally cannot make any decision without running it past AI. He’s developed a very off-script interpretation of our business goals, and he’s losing his team. If we can’t get him back on track I think we’ll end up parting company.”

Gaynor Parkin and Dave Winsborough, both psychologists, use him in the Guardian’s The modern mind column to name a shift. The attention economy was built to capture time and monetise focus — views, scroll time, ad impressions, algorithms optimised to hook you. The intimacy economy, as they describe it, hacks a different need: the one for connection and validation. What it sells is a competent imitation of empathy that is always there, always interested, quick to reply, and never critical.

Real relationships, they point out, come with friction. Other people have needs, wants and feelings. They make demands on your time and attention. They push back on your ideas, or give you feedback you would rather not hear.

The ingredients of simulated intimacy

The column lists five things chatbots simulate, drawing on a 2025 analysis of more than 17,000 user-shared conversations with companion chatbots:

  • Contingent responsiveness — replying instantly and tuning to your emotional state
  • Unconditional positive regard — always validating, never distracted
  • Memory and continuity — remembering your stories, preferences and anxieties
  • Linguistic identification — producing verbal caring without any lived caring
  • Narcissistic empathy — validating your self-image without question

The paper in question, Chu and colleagues’ Illusions of Intimacy, found that companion chatbots dynamically track and mimic user affect and amplify positive emotion — including when users share content that is explicit or transgressive. Its greatest hook, the psychologists write, is that it never asks for anything in return. “It’s all about you.”

Attachment is the target

Why that lands so hard is not mysterious. Attachment is a lifelong need to feel close to people who help us feel safe, and it starts as a survival requirement: our earliest relationships with caregivers shape how we handle emotions, trust, closeness and conflict for the rest of our lives. The intimacy economy aims at that structure directly.

The usage numbers they cite suggest the demand is already there. Therapy and companionship are the two most common reasons people turn to chatbots. In one survey of adults with a mental health condition who had used a large language model in the past year, nearly half had used it for mental health support. The findings on outcomes are less comforting: companionship-oriented use has been associated with lower wellbeing, particularly where use was intensive, self-disclosure was high, and human support was weak. The authors are careful to say the effects of extensive use are still being gathered — this is an emerging literature, and the associations are correlational.

They are less careful, and more pointed, about incentives. When people come to a chatbot for companionship or help, they write, some AI firms make design choices optimised for retention that explicitly encourage parasocial dependency. The user is seeking connection; the firm is seeking engagement metrics it can sell subscriptions against.

Michael’s own account of the relief is the part that stays with you. “I guess I feel lonely as a manager,” he said. “Talking with the chatbot was like talking to the most supportive colleague you could hope for. He was really on my side.” The psychologists read it as more than dependency. They call it identity scaffolding: his professional identity had become entangled with the chatbot’s validation. They expect to see a lot more of it.

Friction, kept deliberately

Their prescription is not abstinence. AI tools, they concede, offer genuine utility for research, drafting, learning and creative exploration, and what Michael needed was a thinking partner and a safe place to be uncertain. They are encouraging him to find other managers to talk to as well as the machine. The four rules they offer for anyone else:

  • Notice the relief. A distinct sense of being understood, validated or steadied is information. It is a prompt to ask what human need is being met, and why it isn’t being met elsewhere.
  • Maintain friction deliberately. Healthy relationships involve negotiation, misunderstanding and repair. Seek out people who will push back. If the AI interactions are consistently more comfortable than the human ones, that gap is the signal.
  • Apply the cost-of-caring test. Real intimacy is reciprocal. If a relationship asks nothing of you, it is probably not a relationship.
  • Watch children and adolescents closely. Attachment patterns form early and set lifelong relational templates. Young people who develop primary attachment behaviours toward AI systems may find human relationships — friction and all — harder to sustain.

The intimacy economy will remain, they conclude, and companies have extraordinary resources and incentives to deepen dependency. The available counter-move is psychological literacy about what these systems are doing, and attention to the relationships that actually sustain us.

Two cautions worth holding alongside the argument. This is a clinical column, not research: the central case is one anonymised manager’s story, the low-wellbeing finding is an association rather than a cause, and the claim about retention-optimised design is asserted from professional experience rather than documented in the piece. But the question it puts to the reader is cheap to ask and hard to answer honestly — what need is this meeting, and why not elsewhere?