The Artificial Personality Model: Theory, Implementation, Measurement and Experimental Validation in Conversational Agents

Alexander Dregger, Barbarella Petz, Erik Kuhn, Pia Volk, Andreas Oberweis, Stefan Hillmann, Jannik StrötgenarXiv 2026 · arXiv

TL;DR

This paper introduces the Artificial Personality Model (APM), a hybrid framework that combines human-centered traits with machine-specific characteristics to better define how we perceive conversational agents. Through three experiments and the development of a novel measurement tool (the APQ), the authors demonstrate that artificial personality significantly shapes user experience, though human personality models like the "Big Five" do not translate perfectly to the digital realm.

What Problem Does This Paper Address?

We are living in the era of the CASA paradigm (Computers are Social Actors). Whether it’s naming a robotic vacuum or arguing with a chatbot, humans instinctively anthropomorphize machines. For developers and researchers, this creates a challenge: if users will attribute a personality to your agent anyway, how do you design, implement, and measure it intentionally?

Existing approaches fall into two camps:

  1. Human-centered models: Using the "Big Five" (Extraversion, Agreeableness, etc.). These are often too abstract for system design and include irrelevant traits (like "Neuroticism").
  2. Machine-centered models: Lists of adjectives (e.g., "robotic," "serviceable"). these often lack theoretical depth and validated measurement tools.

The authors address this gap by proposing a hybrid model and a validated questionnaire to ensure that the personality a designer intends is actually what the user perceives.

Methods

The researchers refined the Artificial Personality Model (APM), which integrates dimensions from human psychology with machine-specific traits. To validate it, they developed the Artificial Personality Questionnaire (APQ), focusing on observable behaviors rather than abstract adjectives.

The original APM structure showing the blend of human and machine-specific dimensions.

They tested the model across three distinct experiments (Total N = 395):

  • Experiment 1 (Humour): A film recommendation chatbot using GPT-4o with zero-shot prompting to test different humor styles (affiliative, aggressive, self-defeating).
  • Experiment 2 (Service Orientation): A package return scenario using high-fidelity mock-up videos to vary the intensity of service-related cues.
  • Experiment 3 (Artificiality): An IT security chatbot using "Personality Infusion"—a structured prompting method—to vary how "machine-like" the agent appeared.

Datasets

The paper does not rely on a pre-existing machine learning dataset. Instead, it generates primary data from 395 human participants across three controlled experiments. The authors also utilized the BFI-2-S (a short version of the Big Five Inventory) as a benchmark to compare how well human-centric models perform against their new machine-centric APQ.

Key Insights and Ideas

  • The "Extraversion" Disconnect: While Extraversion is a cornerstone of human personality, it failed to emerge as a distinct dimension for machines. Users don't perceive "sociability" in a tool that doesn't initiate social interaction on its own.
  • The Rise of "Machiavellianism": A surprising finding was the emergence of a new dimension the authors call Machiavellianism—incorporating traits like manipulation, control, and intrusiveness. This suggests users are highly sensitive to whether an AI is trying to "push" them toward certain decisions.
  • Human Tests Fail for AI: The standard human personality test (BFI-2) showed poor reliability and an uninterpretable factor structure when applied to chatbots. This proves that simply "porting" human psychology to AI is a flawed strategy.
  • Humour is a Double-Edged Sword: Not all "funny" is good. While affiliative humor can boost social presence, aggressive humor (sarcasm/irony) significantly reduced user satisfaction and social presence.

Key Results

  • Measurement Validity: The newly developed APQ showed strong psychometric properties, with most dimensions achieving high reliability (Cronbach’s $\alpha > .80$).
  • UX Impact: Artificial personality had significant effects on user satisfaction, continuous use intention, and social presence. For instance, in Experiment 2, low service orientation drastically reduced the intention to use the chatbot again.
  • Social Cues: The paper provides a comprehensive taxonomy of how personality is expressed through verbal, visual, and even "invisible" cues like response time.
A taxonomy of social cues used to express personality in conversational agents.

Limitations and Future Work

  • Modality Constraints: The studies focused primarily on text-based chatbots. It remains to be seen if the APM holds up for social robots or voice-only assistants where auditory cues (pitch, tempo) dominate.
  • Cultural Bias: The participants were exclusively German-speaking. Personality perception is deeply cultural, and "serviceability" or "humour" might be interpreted differently in other regions.
  • Short-term Interaction: The experiments involved relatively brief encounters. Future research should look at long-term relationships—does a "Machiavellian" personality become more or less apparent after a month of daily use?
  • Prompting Sensitivity: The authors noted that LLMs tend to "over-express" prompted traits, making it difficult to achieve subtle, medium-level personalities. Refining "personality competence"—knowing when to be funny or serious—is a key next step.

Summary updated

Abstract

Artificial personality shapes user experience (UX) in human-machine interaction, yet coherent theoretical frameworks and validated measurement instruments remain limited. This paper refines and empirically evaluates the Artificial Personality Model (APM), a hybrid framework integrating human-centred and machine-specific personality dimensions. Across three experiments (N = 395), the model was operationalised and evaluated using the Artificial Personality Questionnaire (APQ). Factor analyses revealed that some human personality dimensions (e.g., agreeableness, conscientiousness) are partially transferable, whereas others (e.g., extraversion) did not emerge as distinct dimensions. Additional machine-specific dimensions were identified. The APQ provided encouraging initial psychometric evidence, while the human-centred BFI-2 showed poor reliability and an uninterpretable factor structure in Experiment 3. Experimental manipulations demonstrated significant effects of artificial personality on user satisfaction, continuous use intention, and social presence. The findings contribute an empirically refined framework and a novel instrument for measuring perceived artificial personality, offering practical guidance for designing conversational agents.
BibTeX
@misc{dregger2026artificial,
  author = {Alexander Dregger and Barbarella Petz and Erik Kuhn and Pia Volk and Andreas Oberweis and Stefan Hillmann and Jannik Strötgen},
  title = {{The Artificial Personality Model: Theory, Implementation, Measurement and Experimental Validation in Conversational Agents}},
  year = {2026},
  eprint = {2610.38608},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.38608},
}
Figure