Servant, exchange partner, or counterpart: What relationships chatbots offer

The tool-or-friend question falls short

Conversational systems are often marketed as assistants. They perform tasks for their owners and are supposed to respond to instructions. At the same time, people tend to attribute social properties to computers and to develop bonds beyond the mere owner-tool relationship.

Tschopp and colleagues used the Relational Models Theory to describe this diversity more precisely. Instead of only measuring trust or a single closeness metric, they examined various patterns according to which people order their relationship to the system.

The results from a total of 729 frequent users show no uniform role. Conversational AI can simultaneously be experienced as a subordinate assistant, as a partner in an exchange, and, to a lesser extent, as a peer-like counterpart.

The master-assistant model is built in

In the authority ranking model, a hierarchical order is central. The human gives instructions, the system obeys. Terms such as assistant, copilot, or voice command already make this structure visible linguistically.

The study found that the perception of a master–assistant relationship was hardly related to system assessment or user characteristics. It thus appears to be less individual and more deeply anchored as a fundamental pattern of the technology.

For a conversational offering, this hierarchy has an advantage: the human retains control. At the same time, it can lead to unrealistic expectations. A system that is supposed to fulfill every wish is easily confronted with tasks for which it has no competence or responsibility.

The exchange relationship often remains invisible

Market pricing describes a relationship as an exchange. Users receive answers, convenience, or entertainment and give money, data, attention, or product loyalty in return. Even a free chat is therefore not free of consideration.

This level is particularly sensitive in mental health conversations. Personal content has a different character than ordinary click data. If it is used for improvement, analysis, or personalization, the exchange must be understandable and voluntary.

The study found stronger correlations between the exchange relationship and human-like system perception than for the servant role. The more social the chat seems, the easier it is to overlook that an economic product relationship still exists.

Partnership perception is smaller but significant

The peer bonding model describes a relationship on equal footing. In the results, it was less pronounced than hierarchy and exchange. Nevertheless, it had greater predictive power for human-like perception than the pure servant role.

A mental companion often aims precisely at this experience: not a lecturing expert, not a subservient tool, but a counterpart that thinks along and can also contradict. Linguistically, such a position can be created.

Technical reciprocity, however, remains limited. The chatbot has no interests of its own and bears no consequences. Equal footing can therefore be a form of conversation, not the complete description of a relationship.

A product should not blur this distinction through artificial vulnerability. If the bot claims to be disappointed, lonely, or dependent on the person's return, it turns the conversational form into emotional pressure. Contradiction and character are possible without feigning a need for attachment. This boundary particularly protects people who already experience the chat as an important counterpart.

A system can switch between roles

In the same conversation, different models can appear. Those who ask for a summary treat the chat as an assistant. Those who pay for a subscription are simultaneously in an exchange relationship. Those who talk about loneliness may experience the system as a counterpart.

These switches are not contradictory. They explain why users may give a chatbot a rough instruction in one moment and thank it for a personal response in the next. The social meaning arises from the current function.

Product design should enable these switches without manipulating them. A factual task does not need artificial intimacy. Conversely, a personal conversation should not come across with subservient platitudes, as if the chat were merely fulfilling a command.

Trust depends on the relational context

The review by Bach and colleagues of 23 empirical studies found multiple definitions of trust in AI. Socioethical conditions, technical and design features, and user characteristics all had an influence.

The relational models help explain this diversity. In a master–assistant relationship, trust may mean reliable execution. In an exchange, it is about fairness and data protection. In a partnership-like perception, continuity, respect, and appropriate responses matter.

A single trust score cannot distinguish among these requirements. Evaluation should therefore ask which relationship the product offers in a given situation and which expectation was met or violated in that context.

Attachment adds a temporal dimension

Yang and Oshio developed a scale for experiences in human–AI relationships along the dimensions of attachment anxiety and attachment avoidance. Attachment anxiety manifests as a need for emotional reassurance and concern about insufficient responses; avoidance manifests as discomfort with closeness.

These dimensions complement the relational models. A person may experience the chat as peer-like while still preferring distance. Another person may react emotionally strongly to unreliable responses in an exchange relationship.

This makes clear that product roles are not determined solely by names. They evolve over the course of the conversation and encounter different needs. Settings for tone, memory, and closeness are therefore essential relational functions.

The desired role must be designed honestly

A conversational chat may choose a clear role. It can appear as a tool, a structuring assistant, or a mental companion. It becomes problematic when the visible role does not match the actual relationship—for instance, when a seemingly independent friend primarily generates data and attachment for a business model.

For a companion on equal footing, honesty means: a natural tone without claimed human feelings, disagreement without authoritarian posturing, and personalization without artificial neediness. The user remains the owner of their data and can limit closeness.

Servant, exchange partner, and counterpart are not fixed categories. They are models for expectations. Good design makes it recognizable which one currently applies and ensures that a pleasant social surface does not conceal the economic and technical underpinning.