An avatar alone does not create social presence

Design and conversational style were varied separately

The study used a between-participants design. The presence of an avatar and a social or task-oriented interaction style were varied independently of each other.

This makes it easier to see which component influences perception. In real products, figure, language, and functions often appear together, making their individual effects hard to separate.

The data came from 193 millennials in Italy. The chatbot examined was for retail, not for psychological support or medical advice.

Transferring this to other contexts is therefore an assessment, not direct evidence of efficacy. Nevertheless, the experiment provides a useful design question: Do people respond to the image or to the way the system interacts with them?

The social style increased presence

A socially oriented interaction style increased perceived social presence. People experienced the counterpart as more socially present when their communication was designed accordingly.

In the model tested, social presence influenced perceived enjoyment, trust, and ultimately the attitude toward the chatbot. Conversational behavior can thus shape the overall impression of the product.

A social style consists of more than just friendly greetings. It is reflected in whether responses build on previous statements, are clearly formulated, and pick up on the rhythm of the counterpart.

For supportive chats, an additional layer comes into play: the system must offer closeness without feigning human reciprocity or a therapeutic relationship.

Avatar presence was not significant

The study found no significant effect of the mere presence of an avatar on social presence. A figure alone did not automatically make the chat more social.

This does not mean that visual design is unimportant. An avatar can create recognition, orientation, and atmosphere. Its contribution may lie more in brand identity or emotional access than in measured social presence.

Problems arise when the figure promises a more human impression that the conversation does not deliver. A warmly designed companion that rigidly repeats itself or forgets corrections intensifies the gap between expectation and behavior.

Design resources should therefore not be invested solely in animation and expression. At least equally important are trajectory logic, response selection, loading behavior, and a consistent reaction to user boundaries.

Humanness and interactivity influence trust

Ding and Najaf surveyed 343 customers from southern China in a time-delayed online procedure about their use of e-commerce chatbots. Interactivity and perceived humanness had a significant effect on trust.

Trust, in turn, mediated the relationship between these characteristics and the intention to use chatbots. Perceived enjoyment also played a role as a moderating factor.

The results complement the avatar experiment: humanness is not created by a face alone. It also emerges from responsiveness, dialogue flow, and the experience that the system actually engages with user input.

In a healthcare or mental health context, however, more trust is not automatically better. It must match actual reliability so that social design does not create overtrust.

Trust influences intention and actual use

Choudhury and Shamszare surveyed 607 adults in the United States who used ChatGPT 3.5 at least monthly. Common purposes were information seeking, entertainment, and problem solving; 44 individuals mentioned health-related questions.

Trust had a significant direct effect on intention to use and actual use. Part of the effect on use was mediated by intention.

The statistical model explained considerably more variance in intention to use than in actual use. Thus, between a positive attitude and sustained behavior lie additional factors such as usefulness, habit, access, and concrete experiences.

The authors also warn against blind trust, especially because ChatGPT was not originally developed for health applications. Trust must be accompanied and calibrated.

A companion may be visibly designed

For young adults, a distinct character can make an abstract AI service more accessible. Colors, light, and movement can create a calm space without the character having to appear childish.

However, the avatar should not obscure the central information or feign an artificial emotional response. A slight blink or a calm movement can signal presence; dramatic facial expressions quickly claim more understanding than technically exists.

On small screens, the conversational content must take priority. The companion can become more compact while input, history, and AI labeling remain legible and operable.

Accessibility also requires reduced movement, sufficient contrast, and a function that is fully understandable without visual cues.

Social presence needs clear boundaries

A social style can foster enjoyment and trust. With a conversational partner used over a longer period, the question also arises of what expectation of a relationship this creates.

The system should not suggest that it is exclusively available, that it has human needs, or that it misses the person outside of technical use. Such statements reinforce attachment in a way that obscures its role.

Clear labeling as AI is compatible with warmth. It should be visible at first contact and recognizable within the conversation space, without burdening every response with an intrusive standard notice.

Credibility emerges when role, design, and behavior align: a recognizably technical companion that responds respectfully and does not claim human reciprocity.

Even a freely chosen name for the companion does not change this role. It can make the space feel more personal, but it should not be staged as evidence of an independent personality or consciousness.

Behavior is the true character

The study with Millennials shows a clear difference: social interaction style increased social presence, while mere avatar presence did not. A face is therefore not a shortcut to relationship.

Subsequent experiments link interactivity and perceived humanness with trust and usage. This trust is valuable, but it must remain tied to verifiable quality.

For development, this means not defining the companion only visually. Its character shows in whether it listens, carries forward user corrections, does not automatically give advice, and actually makes the chosen tone tangible.

An avatar can open the door. Whether someone stays in the conversation is decided primarily by what happens behind it. Social presence is earned over the course of the conversation, not glued to the surface with a figure.