Names, faces, first-person language and expressive timing make digital systems easier to read as social actors. That can make an interaction feel warmer and more present. It can also lead people to expect more intelligence, empathy or reliability than the system can deliver.
Across chatbot and automation studies, one pattern is especially useful for character design: anthropomorphism works best when the behaviour carries meaningful information. A friendly surface without a capable, honest experience underneath is a promise the product may not keep.
- 5
- studies behind the customer-anger findings
- 461,689
- real chatbot sessions examined in the first study
- 98
- participants in the trust-calibration experiment
The invitation
Human-like cues can create social presence.
In an experiment using working chatbots, Theo Araujo found that human-like cues such as a name and conversational language increased perceived anthropomorphism. Under the right framing, those cues also increased social presence; that sense of presence helped explain a stronger emotional connection with the company.
This is the opening a character can create. It signals that the interaction has a voice and a point of view, giving people a more approachable place to begin than a blank utility interface.
Anthropomorphism earns its place when it makes capability and uncertainty easier to understand.
The expectation cost
The more human the promise, the sharper the disappointment.
Crolic and colleagues combined a large real-world chatbot dataset with four experiments. For customers who were angry, anthropomorphic treatment of a chatbot reduced satisfaction, company evaluations and purchase intentions when the service did not resolve the problem. The proposed mechanism was expectation violation: human-like cues had inflated expectations of efficacy.
The negative effect did not appear in the same way for customers who were not angry, and one experiment found it disappeared when the chatbot resolved the issue. Context and performance—not human-likeness by itself—determined the result.
Useful expression
Communicate certainty, limits and the next step.
A 2024 human-factors experiment separated cosmetic anthropomorphism from meaningful communication. An avatar and expressive voice made an automated assistant feel more human, but appearance alone did not improve trust or trust calibration. Meaningful vocal cues that communicated certainty and uncertainty did.
For an AI character, expression should therefore do real interface work. A pause can show processing. A change in posture can acknowledge uncertainty. A clear transition can signal that a person is taking over. These behaviours make the system more predictable instead of merely more charming.
- Introduce the character’s role and limits in plain language
- Match warmth to the emotional state and stakes of the journey
- Use expression to reveal system state, confidence and progress
- Make escalation visible before frustration becomes a dead end
A calibrated relationship
Design for appropriate trust, not maximum trust.
The goal is not to make everyone trust an automated character as much as possible. It is to help people understand when the system is useful, when its answer needs checking and when another route is better.
That changes the success criterion. The strongest character is not always the most human-looking one. It is the one whose appearance, language and behaviour set an expectation the product can consistently meet.
References
