Source-linked AI summary
How should my chatbot interact? A survey on human-chatbot interaction design
Ana Paula Chaves, Marco Aurelio Gerosa
TL;DR
Chatbot research is fragmented across domains and often examines only limited social characteristics, despite users expecting socially capable interactions. This survey synthesizes 56 studies to develop a conceptual model, finding that 59% focused on task-oriented chatbots and identifying social-characteristic relationships and design strategies.
Problem
Evidence on chatbot social characteristics is fragmented across domains, with most studies examining only a single characteristic or small set despite users’ expectations for social capabilities.
Method
The authors surveyed literature on disembodied, text-based chatbots using a search strategy covering chatbot terminology while excluding embodied and speech-based agents.
Results
The survey included 56 studies across domains, with 59% focusing on task-oriented chatbots, and identified social-characteristic relationships, challenges, and design strategies.
Takeaways & Limitations
The conceptual model and 22 propositions can help designers select appropriate social characteristics and guide researchers’ further investigations.
Takeaways & Limitations
The survey excludes embodied and speech-based conversational agents, whose representations and vocal features may influence chatbot social characteristics and user experiences.
Abstract
from arXiv · showhide
Chatbots' growing popularity has brought new challenges to HCI, having changed the patterns of human interactions with computers. The increasing need to approximate conversational interaction styles raises expectations for chatbots to present social behaviors that are habitual in human-human communication. In this survey, we argue that chatbots should be enriched with social characteristics that cohere with users' expectations, ultimately avoiding frustration and dissatisfaction. We bring together the literature on disembodied, text-based chatbots to derive a conceptual model of social characteristics for chatbots. We analyzed 56 papers from various domains to understand how social characteristics can benefit human-chatbot interactions and identify the challenges and strategies to designing them. Additionally, we discussed how characteristics may influence one another. Our results provide relevant opportunities to both researchers and designers to advance human-chatbot interactions.
1. Introduction
Chatbots have transformed human-computer interaction but often fail to meet users’ expectations, motivating attention to social capabilities alongside functional performance. This survey brings together fragmented research on chatbot social characteristics to guide design choices and future investigation.
- Chatbot evolution and adoption: Chatbots now support general conversations and domain-specific tasks across messaging platforms, websites, and apps.Their widespread deployment reflects a new form of human-computer interaction.
- Motivation: Despite advances in functional performance and accuracy, chatbots still fail to meet users’ expectations.The literature therefore argues that chatbot interactional goals should also include social capabilities.
- Research gap: Research on chatbot social characteristics is dispersed across domains, while most studies examine only one or a small set of characteristics.This fragmentation makes it difficult to understand how social characteristics shape users’ perceptions and behavior toward chatbots.
- Survey contribution: The survey compiles research initiatives examining how chatbot social characteristics influence human-chatbot interaction.It addresses a gap left by extensive technical reviews that do not bring together these social characteristics.
- Implications: The results help designers decide whether characteristics are desirable for particular chatbots and select appropriate subsets, while identifying directions for further research.The survey also discusses how humanness and conversational context influence users’ perceptions.
2. Overview of the surveyed literature
The survey identified 56 studies on social characteristics in disembodied, text-based chatbots across multiple domains. Qualitative coding yielded 11 characteristics grouped into conversational intelligence, social intelligence, and personification.
- Literature search: The search used multiple chatbot-related synonyms because the literature lacks a coherent definition of chatbots.The review excluded studies focused on embodiment and speech input.
- Literature search: Google Scholar searches began with about one thousand papers and excluded studies that did not highlight social characteristics or duplicated other research.Social-characteristic terms were not included because studies seldom label their findings that way.
- Surveyed studies: 56 studies remained after filtering, with 25 papers from human-computer interaction and 8 each from learning and education and information and interactive systems.Other represented domains included virtual agents, artificial intelligence, and natural language processing.
- Surveyed studies: 35 out of 56 studies adopted real chatbots, including 18 analyzing third-party chatbot logs or user perceptions and 9 introducing self-developed architectures or dialogue management.The surveyed systems included Cleverbot, Talkbot, and Woebot among third-party chatbots.
- Conceptual model: Qualitative coding of chatbot behaviors and attributed characteristics produced 11 social characteristics grouped into conversational intelligence, social intelligence, and personification.The analysis focused on characteristics influencing how users perceive and behave toward chatbots.
3. Chatbots Social Characteristics
The section organizes chatbot social characteristics into three categories: conversational intelligence, social intelligence, and personification. It also groups these characteristics by the domains in which they were investigated.
- Conversational intelligence includes characteristics that help chatbots manage interactions.
- Social intelligence focuses on habitual social protocols in chatbot interactions.
- Personification concerns the chatbot’s perceived identity and personality representations.
- The characteristics are also grouped according to the domains in which they were investigated.
3.1. Conversational Intelligence
Conversational intelligence enables chatbots to participate actively while tracking the topic, evolving context, and dialogue flow beyond merely achieving a conversational goal. The surveyed literature identifies benefits, strategies, and challenges for proactivity and conscientiousness in sustaining useful, productive interactions.
- Proactivity: Proactivity adds useful information, inspires users, sustains conversation, and helps chatbots recover naturally from failures.Introducing new topics can prevent the chatbot from getting stuck when it cannot understand the user or find an answer.
- Proactivity: Proactivity improves task productivity by using follow-up questions to resolve and maintain context while reducing search time.In domain-specific interactions, it can also guide users, establish or monitor goals, and support engagement through prompts and reminders.
- Proactivity: Untimely, irrelevant, repetitive, or disruptive proactive messages can annoy users and compromise interaction success.Seven of 13 participants reported irritation, with frequent complaints that the chatbot directed them to specific places.
- Proactivity: Leveraging conversational context is the most frequent proactivity strategy, relating proactive messages to information provided during the conversation.This strategy is intended to increase the usefulness of proactive interventions.
- Conscientiousness: Conscientious chatbots provide meaningful answers, maintain topic continuity, and steer task-oriented conversations toward productive goals.Productivity was the main reason for chatbot use for 68% of participants, while complex tasks increase turns, mistakes, and correction effort.
3.2. Social Intelligence
Social intelligence enables chatbots to produce socially appropriate behavior and account for users’ treatment of computers as social actors. The literature emphasizes damage control for conflict and failure, alongside thoroughness that supports human-likeness and believability while requiring contextual adaptation.
- Social Intelligence: Social intelligence is the ability to produce adequate social behavior to achieve desired goals, while users may respond to computers as social actors.This framing motivates designing chatbots with social characteristics that match human interaction expectations.
- Damage Control: Damage control helps chatbots handle conflict, abuse, testing, and failures caused by limited linguistic or world knowledge.Strategies include responding to harassment, establishing social limits, and admitting or cleverly covering knowledge gaps.
- Damage Control: 4% of hotel-chatbot conversations contained vulgar, indecent, or insulting vocabulary, while 2.8% of all statements were abusive and 1.8% were sexual expressions.Longer conversations encouraged users to move beyond the chatbot’s main functions.
- Damage Control: 10% of exchanged messages in a task-management chatbot were unanswered, motivating designs that handle novel scenarios and acknowledge limited knowledge.Chatbots cannot eliminate conflict because humans also misunderstand and lack assumed common knowledge; damage control instead aims to prevent escalation.
- Damage Control: Damage-control strategies face limits with unfriendly or mischievous users and must fit both the social situation and conflict intensity.Cooperative users were more likely to give higher ratings for overall evaluation and decision efficiency, whereas curiosity and mischief caused some interaction failures.
- Thoroughness: Thoroughness can increase human-likeness and believability, but chatbots should vary detail with the task and maintain a consistent language style.Simple questions may need concise answers, important transactions more information, and mismatched combinations such as formal language with emojis can seem strange.
3.3. Personification
Personification attributes identity and personality traits to chatbots, shaping users’ expectations and interaction outcomes. The surveyed literature links these characteristics to engagement, human-likeness, believability, and richer interpersonal relationships, while emphasizing consistency with user expectations and careful management of stereotypes, capabilities, humor, and trait balance.
- Identity: Attributing identity helps chatbots build common ground in open-domain interactions and manifest credibility and trust in customer service.Identity was primarily studied in open-domain and customer service chatbots, with four studies in each domain.
- Identity: Identity-related design can increase engagement and perceived human-likeness through agent-oriented conversations, human-like language, names, and greetings.Agent-oriented conversations supported anthropomorphic engagement, while human-like language style, names, and greetings produced higher naturalness scores.
- Challenges: Designers must avoid negative stereotypes, balance identity with technical capabilities, adapt humor to users’ cultures, and avoid extreme or conflicting personality traits.Fully human representations produced contradictory outcomes across studies, culture-specific puns had low portability, and extreme personalities sometimes sounded unnatural.
- Personality: Personality can increase believability and enrich interpersonal relationships by making chatbot interactions more coherent, enjoyable, and entertaining.Entertainment was the second most frequent chatbot-use motivation for 20% of participants, and consistent personality helped users relate better to chatbots.
- Design strategies: Personality should consistently influence chatbot language through persona-based models that encode traits and guide response generation.The reviewed strategies recommend architectures in which personality representations influence language and response generation.
4. Discussion
The discussion shows that social characteristics align with human-human interaction expectations and benefit chatbot design, while excessive humanization can create contrasts. Their relevance varies by conversational domain, and the characteristics form complex relationships in which some influence or manifest others.
- Perceived humanness: Social characteristics align with human-human interaction patterns and can benefit chatbot development, but overly humanized agents may produce contrasting outcomes.Chatbots’ perceived social roles may approach human profiles as their communication and social skills become richer.
- Conversational domains: 10 out of 11 social characteristics were found in open-domain conversation studies, where manners, moral agency, emotional intelligence, damage control, identity, personality, and thoroughness were especially relevant.These characteristics address personal content, testing or flaming, and the need to increase chatbot capabilities in unrestricted conversations.
- Conversational domains: Education and customer services reported the greatest number of characteristics among task-oriented domains, although this may reflect greater research maturity and coverage.Customer services use emotional intelligence and manners to manage frustration, communicability to convey services, and personalization to align services with customers.
- Relationships among characteristics: The proposed framework makes explicit literature-based relationships among social characteristics while emphasizing the complexity of developing chatbots with appropriate social capabilities.It is intended to describe relationships found in the surveyed literature rather than provide a comprehensive map.
- Relationships among characteristics: Proactivity, conscientiousness, emotional intelligence, personalization, communicability, thoroughness, and manners influence or manifest other characteristics through context management, adaptation, empathy, and damage recovery.Examples include proactivity influencing perceived personality and conscientiousness, personalization improving emotional intelligence, and manners facilitating damage control.
5. Related Surveys
Previous chatbot surveys have examined urgency, domain-specific applications, and design technologies, but few have addressed social characteristics as their central theme. The reviewed surveys therefore provide related insights without directly focusing on social characteristics.
- Related Surveys: Previous surveys cover chatbot urgency and applications in domains including education, business, and health.They also address information retrieval and e-commerce, among other areas.
- Related Surveys: Three surveys discuss chatbots’ social characteristics, although none focuses specifically on this theme.They include work on conversation-mimicking chatbots, script-based chatbot development practices, and related design insights.
6. Limitations
The survey’s scope excluded embodied and speech-based agents, potentially omitting social characteristics shaped by physical representation and vocal delivery. Because chatbot definitions vary across domains, relevant studies may also have been missed despite broadened search terms and Google Scholar indexing.
- Scope: The survey excluded embodied and speech-based conversational agents, limiting coverage of social characteristics influenced by physical representations, tone, and accent.Examples include identity, politeness, and thoroughness.
- Search coverage: Because chatbot definitions are not consolidated and research spans domains, some studies involving chatbot social aspects may not have been found.The authors used several synonyms in the research string to address this limitation.
- Search coverage: Google Scholar was used as the search engine because it provides fairly comprehensive literature indexing across more domains.This choice complemented the use of several synonyms in the research string.
7. Conclusion
The survey proposes a conceptual model of social characteristics for disembodied, text-based chatbots and identifies challenges, benefits, and strategies for designing appropriate social behaviors. It also highlights domain influences and complex relationships among characteristics as opportunities for further research.
- 7. Conclusion: The survey’s main contribution is a conceptual model of social characteristics for disembodied, text-based chatbots.The model addresses which social characteristics benefit human interactions and the challenges and strategies associated with them.
- 7. Conclusion: Design research should address challenges in manifesting each characteristic, assess users’ perceived benefits and satisfaction, and develop new manifestation strategies.These are identified as three research opportunities arising from the conceptual model.
- 7. Conclusion: Some social characteristics are strongly domain-influenced, whereas manners and damage control are more generally applied.Moral agency and communicability are examples of domain-influenced characteristics.
- 7. Conclusion: 22 propositions from the surveyed literature describe relationships among characteristics and underscore the complexity of developing chatbots with appropriate social behaviors.The survey also reports the domains in which social characteristics were primarily investigated.
About the Authors
The paper is authored by researchers based at Northern Arizona University, with expertise spanning human-chatbot interaction, tourism technologies, software engineering, and CSCW.
- About the Authors: Ana Paula Chaves is a Ph.D. Candidate at Northern Arizona University and a faculty member at the Federal University of Technology–Paraná, researching social aspects of human-chatbot interactions and tourism technologies.She is based in Brazil and provides further information at anachaves.pro.br.
- About the Authors: Marco Aurelio Gerosa is an Associate Professor at Northern Arizona University whose research focuses on software engineering and CSCW.He has served on program committees for FSE, CSCW, SANER, and MSR and is an ACM and IEEE senior member.