Hear You in Silence: Designing for Active Listening in Human Interaction with Conversational Agents Using Context-Aware Pacing

In human conversation, empathic dialogue requires nuanced temporal cues indicating whether the conversational partner is paying attention. This type of "active listening" is overlooked in the design of Conversational Agents (CAs), which use the same pacing for one conversation. To model the temporal cues in human conversation, we need CAs that dynamically adjust response pacing according to user input. We qualitatively analyzed ten cases of active listening to distill five context-aware pacing strategies: Reflective Silence, Facilitative Silence, Empathic Silence, Holding Space, and Immediate Response. In a between-subjects study (N=50) with two conversational scenarios (relationship and career-support), the context-aware agent scored higher than static-pacing control on perceived human-likeness, smoothness, and interactivity, supporting deeper self-disclosure and higher engagement. In the careersupport scenario, the CA yielded higher perceived listening quality and affective trust. This work1 shows how insights from human conversation like context-aware pacing can empower the design of more empathic human-AI communication.

Introduction. Human connection relies on more than just words; it is profoundly shaped by the subtle, temporal rhythms of communication [17]. In supportive conversations, “active listening” requires the listener to use nuanced pacing to signal empathy, attention, and shared understanding [117]. Silence, in this context, is not an empty void but a powerful communicative tool to show attention and interest [17]. It can be used strategically to hold space for a speaker, encourage deeper self-disclosure, or convey empathy [8, 61]. Understanding the function of this pacing is fundamental to understanding how people build trust and rapport. Despite the increasing integration of Conversational Agents (CAs) into socially significant roles like emotional support, their design largely ignores subtle but crucial forms of human communication such as pacing [30]. Current systems typically adopt a static and mechanistic interaction style, prioritizing efficiency and immediate responses [77, 81]. This efficiency-first paradigm creates a critical disconnect, resulting in interactions that feel robotic and superficial [30]. The core problem is not just technological, but a failure to model the interpersonal functions of conversational pacing that are essential for human-centric communication [105, 106]. However, historically in the field of CA design, dynamic pacing with respect to textual content has been largely ignored. Previous research attempting to bridge this gap has faced two major limitations, stemming from an incomplete understanding of the dynamic pacing in human communication:

• Superficial Models of Timing: Attempts to manage conversation timing in CAs have often treated pauses as simple, static delays—technical signals to simulate “thinking” or reduce perceived automation [83, 130]. This perspective overlooks the potential for silence to serve as a dynamic, relational tool embedded within the conversation’s context. • Over-emphasis on Verbal Content: Research on implementing active listening in CAs has focused almost exclusively on verbal strategies, such as paraphrasing and asking follow-up questions [24]. While valuable, these efforts ignore the non-verbal dimension of pacing, which communication psychology has long established as an intentional communicative act essential for building rapport [8].

What existing works fundamentally overlook is that human communication is highly context-dependent [106]. The appropriateness of a particular pacing strategy shifts based on relational goals [63]. For instance, while immediate responses suit task-oriented goals, supportive contexts demand more patient and reflective pacing to foster connection [50]. Therefore, to design more human-centric and supportive CAs, there is a need to deepen our understanding of how people use pacing strategies in different social contexts. To address this gap, our study first seeks to understand and systematize human pacing strategies before translating them into computational design. We conducted a qualitative analysis of ten exemplary video recordings of human-to-human active listening to distill the functional uses of silence and response timing. This analysis yielded a taxonomy of five context-aware pacing strategies: Reflective Silence, Facilitative Silence, Empathic Silence, Holding Space, and Immediate Response, corresponding to eight specific pacing strategies (Recognize, Reconfirm, Re-engage, Reposition, Reconsider, Resonate, Holding, and Resolve). Based on these findings, we designed and built an LLM-based CA [124, 132] that uses context-aware pacing embodying these strategies. With this CA, our work investigates the value of operationalizing the relational functions of human silence into a mechanistic, implementable framework of context-aware pacing. We hypothesized that users would perceive this pacing as empathetic social cues, facilitating more empathic human-AI communication. The CA dynamically adjusts the response pacing strategies according to user input. More specifically, this work explores these two research questions:

• RQ1: How do CAs using context-aware pacing strategies impact users’ perceived quality of interaction and experience, specifically in terms of listening quality, affective trust, cognitive trust, human-likeness, smoothness, and interactivity? • RQ2: How does context-aware pacing influence user interaction behaviors in text-based supportive conversations with CAs, specifically in terms of depth of self-disclosure and level of engagement?

To answer these questions, we conducted a between-subject study (N=50) comparing our context-aware CA against a staticpacing baseline CA across two supportive scenarios (career and relationship advice). These contexts were designed to contrast a scenario blending practical advice with emotional needs (career support) against one centered on interpersonal and affective processing (relationship support). Our work makes the following contributions:

Related work. Active listening is a dynamic process that combines attentiveness, understanding, and constructive intention [52]. It conveys unconditional acceptance and unbiased attitudes towards speakers’ experiences [94]. Rooted in the principle of empathic listening, psychotherapists have developed several active listening strategies, such as eye contact, head-nodding, backchanneling, paraphrasing, and summarizing [66, 116]. Despite being subtle, these cues for listening could produce positive interaction outcomes [117], such as reducing loneliness [41], improving trustworthiness [90], responsiveness [40], and friendliness [14]. One of the primary benefits of active listening is its power to encourage deeper self-disclosure. Guiding individuals to express themselves more deeply has been a central topic in HCI studies [125]. In human-to-human communication, people tend to share more about themselves when they perceive clear benefits or feel emotionally safe [26, 33, 129]. One key barrier to deeper disclosure lies in the anxiety of social evaluation, which refers to the distress or fear experienced in situations where an individual anticipates being judged by others [25]. In contrast, active listening demonstrates a non-judgmental and open attitude towards speakers, thereby encouraging more expression. Inspired by these benefits, HCI researchers have explored integrating active listening capabilities into CAs. In voice-based CAs, backchannel cues (“hmm,” “yeah”) could significantly promote a user’s sense of being “heard” [21, 84, 120]. Similarly, verbal strategies, such as paraphrasing, verbalizing emotions, summarizing, and encouraging disclosure, also have positive effects in conveying attentiveness and improving user experience with CAs [120]. Social Penetration Theory suggests that the exchange of personal information catalyzes relationship development through reciprocity [3]. Similar to interpersonal communication [78], text-based CAs that disclose their feelings, emotions or preferences could create an equal and affable conversational atmosphere [74], thereby eliciting a sense of being “heard” and enhancing people’s perceived intimacy [60]. More recently, researchers have also proposed displaying the agent’s “internal empathic resonance” via a secondary text channel to explicitly visualize its understanding of the user [97]. Despite this progress, existing studies primarily focus on what an agent says, overlooking a fundamental component of human communication: how and when it responds. In human dialogue, the non-verbal, temporal dynamics of a conversation (pacing) are powerful communicative signals that convey empathy and presence [17, 117].

Method. Most research on active listening in CAs focuses on verbal strategies like paraphrasing and questioning [109, 120], leaving the crucial role of pacing within conversations largely unaddressed. This overlooks how silence functions as a powerful communicative resource in human interaction for emotional articulation and self-awareness [32]. We reframe silence as context-aware pacing, i.e., the strategic use of silence to support a speaker’s needs. To understand how to operationalize this concept, we conducted a formative study analyzing real-world active listening cases, focusing specifically on classifying the function of pacing strategies in response to the conversational context.

3.1.2 Data Analysis. We employed an open coding approach to analyze the function of pacing within active listening strategies employed by the counselors [75]. Three researchers independently coded the data, focusing on “strategic use of silence segments” as the unit of analysis. Rather than treating silence as an isolated phenomenon, we analyzed how it was embedded in context, responded to prior utterances, and guided the dialogue. As shown in Figure 2, our coding scheme captured dimensions such as the strategy type, contextual trigger, silence duration, timing, and surrounding verbal responses. Any disagreements were resolved through discussion to reach a consensus. During the coding process, we reached saturation, meaning that no new categories emerged as we continued analyzing the data. The categories identified effectively covered all instances of silence and its strategic functions, indicating that we had fully captured the relevant behaviors within this context. To ensure consistency and reliability in the coding, we adopted a multi-coder approach. After each researcher independently coded the data, we compared their results and resolved any discrepancies through discussion. This discussion process helped us maintain consistency while ensuring the accuracy of the coding scheme.

3.1.3 Pacing-Based Listening Strategies. Ultimately, we identified five key types of pacing-based strategies, each serving distinct communicative and psychological support functions within conversations: Reflective Silence, Facilitative Silence, Empathic Silence, Holding Space, and Immediate Response. Reflective Silence conveys the listener’s thoughtful processing and confirmation of the speaker’s message. Through brief silence and reflective responses, the listener signals that they are seriously attending to the content, thereby reinforcing trust and engagement. Facilitative Silence is employed in contexts of ambiguity, disruption, or topic transition. Here, silence acts as a subtle prompt, encouraging the speaker to clarify thoughts or re-engage in narration. Empathic Silence broadly applies when speakers articulate deep feelings, beliefs, or values. By remaining silent, the listener expresses acceptance and empathy without judgment, fostering self-expression and emotional integration. Holding Space emphasizes creating a non-directive, emotionally safe environment during moments of intense vulnerability. The listener uses silence to allow the speaker to fully experience and process their internal state without interruption or pressure. Finally, Immediate Response, which applies no deliberate silence, addresses explicit information-seeking needs efficiently. These strategies represent distinct modes of support in active listening that help respond to emotional disclosures, clarify narratives, and facilitate deeper conversational progression. More specifically, they are operationalized by eight specific, implementable strategies (e.g., Recognize, Reconfirm, etc.), which are detailed in Table 1.

3.2.1 Reflective Silence. Reflective Silence is mainly applicable when people share their experiences or feelings, or when they indirectly seek advice. For example, words like “I’ve been unsure if I’m...” or “I just don’t know if I can keep going with...” often indicate such situations. In these cases, the listener should focus on providing emotional support and validation, helping people feel understood 3.2.2 Facilitative Silence.

Discussion. To understand conversational pacing in human-CA interaction, we implemented and evaluated a system featuring context-aware pacing. We designed the agent to operationalize different pacing strategies ranging from Immediate Response for task-oriented queries to deliberate silence (e.g., Holding Space) for emotional vulnerability, assessing effects on perceived interaction quality and user experience. Our study found that context-aware pacing improved perceived listening quality, aligning with previous interpersonal communication works where appropriate response pacing is suggested to enhance responsiveness and high-quality listening [106]. Unlike prior HCI studies focusing on verbal content [21, 46, 56, 60, 66, 120], our results suggested that compared to a static fast response, CAs’ adaptively slow response or even silence can also lead to higherquality listening perception, even when the content of the response is similar. Crucially, positive user perception relied on the CA’s high accuracy (86.2%) in selecting suitable strategies. This reliability ensured appropriate pacing for distinct contexts like factual queries (Immediate Response) versus intensive distress (Holding Space), preventing frustration from mismatched pacing (e.g., a long, awkward silence during a simple information exchange). While precision was slightly lower for subtle emotional nuances, these were 6.2 Design Implications for Embodying Pacing in CAs 6.3 Pitfalls of Current CA Response Timing • The fallacy of uniform immediacy. The prevailing design philosophy often treats any delay as system latency to be eliminated. Although contemporary generative models exhibit varied response delays based on computational complexity (“thinking” efforts), this technical latency is often decontextualized from the user’s relational needs [87]. In other words, the system delays are due to complex queries, not the users’ emotional space. While speed is crucial for purely task-oriented queries, our findings confirm that in supportive contexts where both advice and emotional comforting are needed, consistently immediate responses can be perceived as perfunctory, emotionally detached, and affectively untrustworthy. By responding instantly, the agent fails to perform the social function of “deliberation”, signaling to the user that their input was merely processed rather than truly considered. • Conversational dominance through continuous output. The current paradigm often encourages CAs to “talk too much” by providing lengthy, comprehensive answers as quickly as possible. This approach fails to recognize that silence and pauses are essential for user reflection and turntaking. When an agent floods the interaction with continuous text, it creates a high cognitive load for the user and shifts the dynamic from a collaborative dialogue to a passive information dump. This disempowers the user and discourages self-disclosure. In contrast, our findings show that strategies like Holding Space and Facilitative Silence, where the agent strategically “shuts up and listens”, empower users by giving them room to process emotions and elaborate on their thoughts, leading to deeper engagement.

The findings from this study have several implications for generalizability, which must be considered alongside the study’s specific limitations (discussed in Section 6.6).

6.4.3 Generalizability Across User Populations and Experiences. While the core principle of using pacing to convey social cues in CAs is likely generalizable, we caution readers against generalizing our specific results or five-strategy framework to all populations. Our findings suggest that users’ backgrounds and traits are critical factors.

Conclusion. This paper underscores the often-overlooked importance of temporal cues in designing CAs for active listening. Based on a qualitative analysis of human conversations, we derived a framework of five context-aware pacing strategies: Reflective Silence, Facilitative Silence, Empathic Silence, Holding Space, and Immediate Response. We implemented these into a prototype and conducted a betweensubjects study (N= 50) comparing it against a static baseline across career and relationship scenarios. Results demonstrate that contextaware pacing significantly enhances perceived human-likeness, smoothness, and interactivity, while fostering deeper self-disclosure and engagement (evidenced by increased emotional language, firstperson pronouns, and conversation turns). This work provides empirical evidence encouraging designers to move beyond optimizing response content and strategically incorporate the communicative power of silence and pacing. We hope this research serves as an exploratory step toward promoting more human-centric and emotionally attuned CAs by encouraging a design paradigm that values strategic listening as highly as articulate responding.

Limitations. 6.6 Limitations and Future Work 6.6.1 Limitations in Controlling for Generative Content. A limitation of our study is the inherent variability of generative content. Because participant conversations were free-flowing and agent responses were dynamically generated, it was infeasible to strictly control the turn-by-turn text. While the semantic content analysis (Section 5.2) confirmed strong semantic equivalence, this analysis cannot fully discount subtle, unmeasured differences. Furthermore, as we discussed in Section 6.2.2, response quality can influence the user’s perception of pacing.