Journal of Management Information Systems (2026) AI Processed Human Approved
Artificial Normality:How Conversational Agents’ Perceived Humanness Inhibits Error Attribution and Preserves Satisfaction
This study examines how designing conversational agents (CAs) to appear more humanlike influences user responses when errors occur during service interactions. Across two randomized online experiments, the authors evaluated how perceived humanness, situational normality, and error attribution impact user service satisfaction.
Problem
Conversational agents frequently make process errors that harm user satisfaction, as users tend to view algorithmic mistakes as unnatural and attribute fault to the system. Understanding the theoretical mechanisms that explain how humanlike design mitigates these negative reactions remains a critical gap in human-AI interaction research.
Outcome
- Greater perceived humanness in conversational agents preserves perceptions of situational normality when minor errors occur.
- Preserved situational normality reduces users' tendency to attribute error cause to the agent, thereby safeguarding overall service satisfaction.
- Errors made by humanlike conversational agents are processed similarly to errors made by human service agents, demonstrating that humanlike design can alleviate algorithm aversion.
- Preserved situational normality reduces users' tendency to attribute error cause to the agent, thereby safeguarding overall service satisfaction.
- Errors made by humanlike conversational agents are processed similarly to errors made by human service agents, demonstrating that humanlike design can alleviate algorithm aversion.
What it means for you
- CIO / IT Executive: On Monday morning, allocate budget to pilot the integration of more human-centric design principles (e.g., nuanced language, empathetic phrasing) into our core customer-facing conversational agent, focusing on areas where process errors are most common, to preemptively reduce user error attribution and maintain satisfaction.
- IT Manager: On Monday morning, task your development team to review our current conversational agent's dialogue logs for instances of process errors and identify the top 3 most frequent error types to begin redesigning their responses with more humanlike cues (e.g., slight pauses, conversational fillers) to normalize the situation for users.
- Business Strategist: On Monday morning, initiate a cross-functional workshop with marketing and product development to explore how to strategically position our conversational agents' perceived humanness in customer communications, emphasizing its role in creating a seamless experience even when minor technical glitches occur.
- Researcher: On Monday morning, begin designing a follow-up study that experimentally manipulates specific humanlike conversational features (e.g., varying levels of politeness, acknowledgment of difficulty) within a simulated error scenario to quantify their direct impact on error attribution and situational normality, building upon the findings of Artificial Normality.
- Policymaker: On Monday morning, convene a small working group to draft guidelines for the ethical development of conversational agents, recommending that AI design standards encourage the incorporation of humanlike qualities to enhance user experience and mitigate negative reactions to inevitable system errors, rather than solely focusing on pure efficiency.
Transcript
Host: Welcome to A.I.S. Insights — powered by Living Knowledge. I'm your host, Anna Ivy Summers. Today, we're diving into a fascinating new study titled "Artificial Normality: How Conversational Agents’ Perceived Humanness Inhibits Error Attribution and Preserves Satisfaction." Joined by our expert analyst, Alex Ian Sutherland, we're exploring how the design of AI chatbots affects customer satisfaction when things don't go strictly according to plan. Alex, welcome!
Expert: Thanks, Anna. It's great to be here.
Host: So, Alex, let's start with the big picture. We’ve all interacted with customer service chatbots that make mistakes—maybe it misunderstands a date, asks you to rephrase, or loses track of what you just said. What is the core problem this study addresses regarding those AI errors?
Expert: Well, Anna, those kinds of hiccups are known as "process errors." They don't necessarily stop you from finishing your task, but they make the experience frustrating. Historically, when a computer or an algorithm makes a mistake, humans view it as unnatural. We expect software to be flawless. When it errs, it breaks what researchers call "situational normality"—that background sense that everything is customary and working as expected.
Host: And when that sense of normality breaks, what happens in the user's mind?
Expert: It triggers a psychological process called "error attribution." Basically, the user stops running on autopilot and starts actively looking for something or someone to blame. Because humans naturally have a self-serving bias, we rarely blame ourselves. Instead, we point the finger at the chatbot, which drastically reduces our overall satisfaction with the service.
Host: That makes total sense. So how did the researchers behind this study approach solving that problem?
Expert: They conducted two randomized online experiments. In the first study, nearly three hundred participants interacted directly with an AI chatbot to complete a service task—booking an e-bike rental. The researchers varied two things: first, the chatbot's design, making it either very machinelike or humanlike using social cues like human names, conversational tone, and natural response delays. Second, they varied the magnitude of the error—from a minor misunderstanding to a larger disruption like forgetting previous answers.
Host: And what about the second experiment?
Expert: In Study 2, they wanted to see how a humanlike chatbot compared to an actual human customer service agent. They had over three hundred participants watch videos of identical chat interactions featuring either a human agent or an AI agent making the exact same errors.
Host: That sounds like a robust setup. What were the key findings from these experiments?
Expert: The results were really eye-opening. First, when a conversational agent is designed to appear humanlike, users perceive minor errors as far more "normal." It taps into the age-old mindset that "to err is human." This is what the authors call "artificial normality."
Host: So because it feels more human, the mistake doesn't disrupt the flow as much?
Expert: Exactly. And because situational normality is preserved, users don't feel the need to engage in deliberate error attribution. They don't actively sit there blaming the bot. As a direct result, their service satisfaction remains largely protected despite the error occurring.
Host: That's incredible. And how did the humanlike AI compare to the real human agent in Study 2?
Expert: Participants actually evaluated errors from a humanlike AI chatbot in almost the exact same way they evaluated errors from a human agent! There was no evidence of "algorithm aversion"—meaning people weren't harsher on the AI just because it was a machine, provided it presented itself with humanlike qualities.
Host: That brings us to why this matters so much for business leadership today. Alex, what are the key practical takeaways for leaders deploying conversational AI?
Expert: First, as companies shift toward Large Language Model or LLM-driven chatbots, errors like hallucinating facts or losing context during long conversations are inevitable. This study shows that humanlike design acts as a vital psychological buffer. Equipping your AI with appropriate social cues—such as warm language, an avatar, and natural pacing—can offset the negative impact of those minor process errors on customer satisfaction.
Host: But I imagine there's a fine line here. Companies shouldn't just use humanlike design to mask broken systems, right?
Expert: Spot on, Anna. The study emphasizes an ethical boundary. Humanlike design helps soften minor, routine process hiccups, but it won't save you from major outcome failures where the customer fails to achieve their goal. Leaders shouldn't use humanness as a shield to hide flawed technology. It must be paired with continuous efforts to improve the underlying system.
Host: That is such a vital distinction. To wrap up, designing AI to feel more human creates a sense of artificial normality that buffers against minor mistakes, keeping customer satisfaction intact while developers work on refining the tech. Alex, thank you so much for breaking down this study for us today.
Expert: My pleasure, Anna.
Host: And thank you to our listeners for tuning into A.I.S. Insights — powered by Living Knowledge. Join us next time as we continue exploring the intersection of business, technology, and strategy.
Expert: Thanks, Anna. It's great to be here.
Host: So, Alex, let's start with the big picture. We’ve all interacted with customer service chatbots that make mistakes—maybe it misunderstands a date, asks you to rephrase, or loses track of what you just said. What is the core problem this study addresses regarding those AI errors?
Expert: Well, Anna, those kinds of hiccups are known as "process errors." They don't necessarily stop you from finishing your task, but they make the experience frustrating. Historically, when a computer or an algorithm makes a mistake, humans view it as unnatural. We expect software to be flawless. When it errs, it breaks what researchers call "situational normality"—that background sense that everything is customary and working as expected.
Host: And when that sense of normality breaks, what happens in the user's mind?
Expert: It triggers a psychological process called "error attribution." Basically, the user stops running on autopilot and starts actively looking for something or someone to blame. Because humans naturally have a self-serving bias, we rarely blame ourselves. Instead, we point the finger at the chatbot, which drastically reduces our overall satisfaction with the service.
Host: That makes total sense. So how did the researchers behind this study approach solving that problem?
Expert: They conducted two randomized online experiments. In the first study, nearly three hundred participants interacted directly with an AI chatbot to complete a service task—booking an e-bike rental. The researchers varied two things: first, the chatbot's design, making it either very machinelike or humanlike using social cues like human names, conversational tone, and natural response delays. Second, they varied the magnitude of the error—from a minor misunderstanding to a larger disruption like forgetting previous answers.
Host: And what about the second experiment?
Expert: In Study 2, they wanted to see how a humanlike chatbot compared to an actual human customer service agent. They had over three hundred participants watch videos of identical chat interactions featuring either a human agent or an AI agent making the exact same errors.
Host: That sounds like a robust setup. What were the key findings from these experiments?
Expert: The results were really eye-opening. First, when a conversational agent is designed to appear humanlike, users perceive minor errors as far more "normal." It taps into the age-old mindset that "to err is human." This is what the authors call "artificial normality."
Host: So because it feels more human, the mistake doesn't disrupt the flow as much?
Expert: Exactly. And because situational normality is preserved, users don't feel the need to engage in deliberate error attribution. They don't actively sit there blaming the bot. As a direct result, their service satisfaction remains largely protected despite the error occurring.
Host: That's incredible. And how did the humanlike AI compare to the real human agent in Study 2?
Expert: Participants actually evaluated errors from a humanlike AI chatbot in almost the exact same way they evaluated errors from a human agent! There was no evidence of "algorithm aversion"—meaning people weren't harsher on the AI just because it was a machine, provided it presented itself with humanlike qualities.
Host: That brings us to why this matters so much for business leadership today. Alex, what are the key practical takeaways for leaders deploying conversational AI?
Expert: First, as companies shift toward Large Language Model or LLM-driven chatbots, errors like hallucinating facts or losing context during long conversations are inevitable. This study shows that humanlike design acts as a vital psychological buffer. Equipping your AI with appropriate social cues—such as warm language, an avatar, and natural pacing—can offset the negative impact of those minor process errors on customer satisfaction.
Host: But I imagine there's a fine line here. Companies shouldn't just use humanlike design to mask broken systems, right?
Expert: Spot on, Anna. The study emphasizes an ethical boundary. Humanlike design helps soften minor, routine process hiccups, but it won't save you from major outcome failures where the customer fails to achieve their goal. Leaders shouldn't use humanness as a shield to hide flawed technology. It must be paired with continuous efforts to improve the underlying system.
Host: That is such a vital distinction. To wrap up, designing AI to feel more human creates a sense of artificial normality that buffers against minor mistakes, keeping customer satisfaction intact while developers work on refining the tech. Alex, thank you so much for breaking down this study for us today.
Expert: My pleasure, Anna.
Host: And thank you to our listeners for tuning into A.I.S. Insights — powered by Living Knowledge. Join us next time as we continue exploring the intersection of business, technology, and strategy.