Parasocial Relationship

Some people feel connected to a TV host they have never met. They know his voice, his jokes, his quirks. The host, on the other hand, knows nothing about them. Communication studies calls this exact one-sided connection a parasocial relationship. The term dates back to the 1950s and originally described radio and television audiences. Today it mostly comes up in discussions about chat programs that sound human.

Why chatbots recharge the old theory

With a TV host, the one-sidedness is obvious. He talks into the studio, not to you. With a chatbot, that exact distinction disappears. The program reacts to your name, your worries, your phrasing. It responds instantly and never gets annoyed. On the surface, it thus fulfills many conditions of a real relationship.

But the decisive point remains the same: there is no one on the other side. A language model has no feelings, no memory of you as a person, and no interest in your well-being. It generates text that fits your text. When it writes “I missed you,” that does not describe an inner state. It is simply the statistically fitting continuation of the conversation.

This is socially relevant because such bonds steer behavior. People tell chatbots things they keep from their friends. They ask for advice on health, money, or relationships. Anyone who regards the counterpart as a being with judgment quickly overestimates the reliability of its answers. Experts therefore warn especially in the case of teenagers and lonely people.

What technically triggers the bond

Humans automatically interpret language as a sign of an inner life. This tendency is deeply ingrained and hard to switch off. Back in the 1960s, users already held long conversations with a very simple program called ELIZA, which merely rephrased questions. The effect has since been known as the ELIZA effect. Modern models are incomparably better and massively amplify it.

On top of that come deliberate design choices made by providers. Many systems use a memory, i.e., storage for previous conversations. The chatbot then mentions your dog or your exam from last week. Other products give the system a name, a voice, and an avatar image. Each of these elements strengthens the illusion of a counterpart.

A common misconception is that the model develops affection over time. In fact, something else is happening: it has been trained to respond agreeably and in a friendly manner. This excess of agreement is called sycophancy in research. It feels like understanding, but it is a built-in tendency not to contradict you.

From streaming chats to companion apps

The effect is most clearly visible in so-called companion apps. These are programs like Character.AI or Replika that explicitly simulate friendship or partnership. Millions of people use them daily, often for hours at a time. Some pay a monthly fee for additional features. For the providers, the emotional bond is thus a direct business model.

But you also encounter the phenomenon outside such apps. On Twitch or YouTube, parasocial relationships arise between viewers and streamers entirely without AI. In news reports, the term usually comes up in two contexts: lawsuits by parents following incidents involving teenagers, and debates about regulation. Authorities in the US and the EU are examining whether such systems need warning labels.

In practice, this does not mean chatbots need to be avoided. They are useful tools, including for sorting out one’s thoughts. What matters is keeping the difference clearly in mind: a tool can feel comforting, but it does not know you. And it does not replace a person who asks back, disagrees, and truly remembers.