Interactivity and Human-Likeness Build Trust—but Not Truth

The study used time-lagged surveys

343 customers from southern China participated through a survey company. The data collection was conducted online in Chinese and used a time-lag approach.

Time-lagged measurements can reduce some problems of simultaneous self-reports, but they remain observational. They show statistical relationships between perceptions, trust, and intention.

The context was online retail. There, users expect quick product information, orientation, and service. This differs from an open emotional conversation.

The psychological mechanisms of interactivity and human-likeness are nevertheless of interest for other chat offerings, as long as their transfer is treated as a hypothesis and not as direct evidence.

Interactivity affected trust

The more interactive the chatbot was perceived to be, the higher the trust. Interactivity means more than the existence of an input field. It encompasses the impression that the system responds promptly and appropriately to the user's input.

In a real conversation, this is demonstrated by referencing specific details, asking meaningful follow-up questions, and the ability to change behavior after a correction.

In contrast, superficial contingency quickly feels hollow: the chat repeats a word from the message but continues its standard template unchanged. Linguistic mirroring alone does not prove responsiveness.

Technical measurements should therefore not only consider response time and message count. They must check whether new information actually influences the subsequent course of the conversation.

This includes tests with contradictory information and later corrections. If a person clarifies that they sleep well but only briefly, the chat must not continue to assume poor sleep onset. Such small differences determine whether interactivity is actually experienced or merely simulated graphically.

Perceived humanness also had an effect

Perceived humanness also significantly influenced trust. People respond to social features such as natural language, politeness, rhythm, and possibly a personalized style.

Human-likeness is an attribution in this context. The system is perceived as more human without thereby becoming a human or possessing human experience.

In sensitive applications, this attribution can foster positive openness, but it can also lead to an overestimation of capabilities. A naturally sounding piece of advice appears more easily as individually tailored and professionally substantiated, even if it is generated from only a few pieces of information.

Clear labeling of AI should therefore not be viewed as a disruption. It keeps the role visible, while tone and design can remain respectful and accessible.

Trust mediated the intention to use

The study found a significant mediating effect of trust between interactivity or human-likeness and the intention to use the chatbot.

This means: these characteristics influenced adoption not only directly as pleasant product features, but also through the fact that people were more likely to trust the system.

For e-commerce, this can encourage the use of a service channel. In a conversation about psychological distress, there is additionally a responsibility for the quality of what is accepted on the basis of this trust.

A product should therefore not exclusively optimize conversion or dwell time. It must also capture whether users correctly understand the role and adequately scrutinize risky statements.

Joy changed the relationships

In the e-commerce study, perceived joy had a moderating influence. A pleasant user experience can thus change how people translate chatbot features into trust and adoption intention.

Joy is an obvious target metric in commerce. For a mental health chat, the appropriate experience does not always have to be positive or playful. A serious, calm space can be more helpful than gamification.

The transferable principle is not maximum entertainment, but emotional fit. Design should support the task without trivializing sensitive content.

A companion avatar can also create warmth. It should not overshadow the content of the conversation and should not impose a childish or overly human role.

A social style has a stronger effect than an avatar alone

De Cicco, Costa e Silva, and Alparone varied both avatar presence and social versus task-oriented interaction style among 193 Italian millennials.

The social style increased perceived social presence. No significant effect was found for the mere presence of the avatar.

Social presence was linked to enjoyment, trust, and a positive attitude in the model examined. This finding complements the e-commerce study: behavior and interaction quality shape the social impression more strongly than a figure alone.

Visual design remains important for recognition and atmosphere. However, its promise must be fulfilled through the dialogue.

More trust does not automatically lead to better usage

Choudhury and Shamszare found significant effects of trust on usage intention and actual usage among 607 regular ChatGPT users. The model explained intention much better than actual usage.

The authors warn against overtrust in health questions. ChatGPT was not originally developed as a health system, but is used for such questions by a portion of people.

A supportive chat must therefore recognize which concerns it can handle safely. This distinction must not appear only at the end as a general disclaimer, but must influence the specific form of the answer.

Trust is productive when it matches performance. If it is built up more through human-likeness and interactivity than through technical reliability, a dangerous gap emerges.

This gap should be examined directly in user tests. After a conversation, participants can assess which statements they understood as verified, personally tailored, or merely as a general perspective. Misunderstandings about the role are product defects in their own right.

Good design calibrates rather than persuades

The three studies show how interactivity, social style, perceived humanness, and enjoyment influence trust and adoption. They come from e-commerce, retail, and general ChatGPT use.

For sensitive conversational systems, these mechanisms should not simply be adopted as growth levers. The goal is not a maximally humanized chat that people follow as often as possible.

Good design enables natural communication, keeps the AI role visible, and respects corrections. It makes sources, uncertainty, and data paths understandable where they are relevant.

Interactivity and humanness build trust, but not truth. Only when social quality is combined with verifiable performance, clear boundaries, and genuine freedom of choice does this trust sustain a responsible service.