Abstract
Podcasts elicit strong affective reactions that shape how listeners participate and evaluate. Drawing on Emotional Engagement Theory, we conceptualize emotion as a behavioural activation process that translates emotional expression into observable participation. We specify how this operates via the Elaboration Likelihood Model: utilitarian shows tend to invite more central-route processing (argument quality, informational utility), whereas hedonic shows more often engage peripheral-route cues (narrative tone, affective heuristics). We analysed a dataset of 108,464 Apple podcasts with 2,017,209 reviews using a transformer-based artificial intelligence emotion detection model. We found that low levels of anger, disgust, fear, sadness, and surprise, next to a dominant joy emotion, positively influenced the number of reviews a podcast receives for both utilitarian and hedonic podcasts. In contrast, achieving higher ratings requires minimizing these alternative emotions relative to joy. Some differences emerged between utilitarian and hedonic podcasts, with hedonic podcasts benefitting from a broader range of emotions, while balancing the effects of the mixed emotions in terms of increasing the number of reviews and reducing the average rating score.
Keywords
Introduction
Podcasts have surged in popularity as a medium for content consumption, covering diverse topics and audiences. They often evoke strong emotional responses, which can influence their success. This study examines the role of different listener emotions in driving engagement with podcasts. We focus on two primary podcast categories – utilitarian (informative) and hedonic (entertaining) – to determine whether emotional impacts vary by content type. Drawing on Emotional Engagement Theory, we conceptualize emotional engagement not merely as an affective reaction, but as a behavioural activation process that translates emotional arousal expressed in reviews into observable participation outcomes. We leverage a transformer-based emotion detection model to analyse 2,017,209 written podcast reviews. We investigate how mixed emotions expressed by listeners relate to key engagement outcomes, namely participatory engagement (number of reviews, reflecting expressive action) and evaluative favourability (ratings, reflecting consolidated judgement). Previous research assessed the critical role of emotions in more general media consumption. For instance, Schreiner et al. (2021) applied Emotional Engagement Theory for evaluating consumer reactions to podcasts, which posits that emotional involvement influences content perception and satisfaction.
Dolan et al. (2019) emphasized that emotional content in social media facilitates engagement more effectively than purely informational content, highlighting the importance of emotional appeal in content strategy. Rational appeals in social media are reported to be more effective at encouraging both active and passive engagement among users, while emotional appeals tend to promote passive engagement. This distinction arises from the nature of how these appeals interact with users’ cognitive and emotional responses. A study on social media advertising (P. Yang et al., 2024) found that rational appeals often include informative content, which leads to higher perceived enjoyment and responsiveness, fostering greater engagement with ads and messages.
In contrast, emotional appeals, while capable of building empathy and influencing attitudes, primarily enhance passive engagement, as for example seen in studies on water conservation and social marketing (Abu Bakar et al., 2024). Nevertheless, Abu Bakar et al. (2024) and Balderjahn and Hoffmann (2023) found that positive ratings on platforms like Reddit foster positive emotional expressions and higher engagement, while negative ratings result in the opposite (Davis & Graham, 2021). Hollebeek and Macky (2019) pointed out that consumer motivations are based on functional (utilitarian) and hedonic aspects and that this drives consumer interactions with digital content marketing, implying a relevance for engagement.
However, prior podcast and digital media research typically conceptualises engagement as a multidimensional construct comprising cognitive, emotional, and behavioural components (e.g., Brodie et al., 2011; Levesque & Pons, 2023), without explicitly distinguishing behavioural activation (e.g., review participation) from evaluative judgement (e.g., ratings). As a result, it remains unclear how discrete and mixed emotions differentially translate into participatory engagement versus favourability, particularly across distinct processing regimes.
We investigate the relationship between reviews of utilitarian and hedonic podcasts, ratings, and listener emotions. In contrast to prior podcast studies that analyse content characteristics, our approach centres on the emotions expressed in audience reviews and their impact on engagement. We utilize a previously developed transformer-based emotion detection model (Lee et al., 2023) to identify six basic emotions in a large corpus of Apple Podcast reviews (Stuart Axelbrooke, 2023). We assess how emotions are related to consumer engagement; that is, the number of reviews and ratings, while distinguishing between more utilitarian and hedonic podcasts.
We address the effect of the six specific emotions from the model of Ekman (1972) – joy, anger, disgust, sadness, surprise, and fear – on engagement. Ekman’s (1972) model has been used extensively in applications of deep learning for emotion detection (see Lee et al. [2023, 2024] for overviews). Our first research question, RQ1, is formulated as: “How does the mix of emotions evoked by podcasts influence consumers’ engagement with podcasts?” RQ2 is formulated as: “Does the mix of emotions evoked by podcasts influence engagement for utilitarian versus hedonic podcasts differently?”
Our contribution is threefold. First, we advance theory by specifying an Elaboration Likelihood Model (ELM)-based account (Petty & Cacioppo, 1986) in which podcast motive (utilitarian vs. hedonic) conditions how discrete emotions (Ekman, 1972) expressed in reviews map onto two outcomes: participatory engagement (reviewing) and evaluative favourability (ratings). This motive-conditioned mechanism clarifies why a binary operationalization is analytically useful: it represents distinct processing regimes (central vs. peripheral) and yields testable implications that move beyond global sentiment to the role of specific emotions and their mixed configurations. Second, we contribute methodologically by adapting a transformer-based emotion detection model to the podcast domain, scaling prior applications from Tweets and donations (Lee et al., 2024) and online book reviews (Lee & de Villiers, 2025) to a large corpus of podcast reviews. Third, we offer empirical insight by showing that discrete and mixed emotions relate differently to participation versus favourability, and that these relationships vary systematically by motive, thereby enriching work on utilitarian-hedonic media effects without relying on detailed feature engineering or hand coding.
Next, we report the literature review that contextualizes emotional engagement theory and its relevance to engagement with podcasts. Subsequent sections present our methodology, analyse the data, and discuss the findings, leading to conclusions about how different emotions affect consumer engagement with hedonic and utilitarian podcasts.
Literature Review
Basic Emotions and Their Impact
Ekman (1972) found that six basic human emotions (e.g., anger, disgust, fear, joy, sadness, and surprise) play a crucial role in human experiences. These six emotions are universally recognized, have distinct facial expressions, and unique physiological responses. In this section we discuss how these emotions may relate to listener engagement with podcasts.
Anger is a strong feeling of displeasure or hostility in response to a perceived unfair or threatening situation. Anger often arises in reactions to negative events or actions, triggering behaviours aimed at confronting or addressing the negative stimulus (Ekman, 1992). Guzsvinecz and Szűcs (2023) observed that negative reviews, often longer and more detailed, may be driven by anger. Another extant study examines how anger influences the perceived helpfulness of product reviews, suggesting that anger likely correlates with lower podcast ratings as consumers express dissatisfaction with content that triggers this emotion (Craciun et al., 2020).
Disgust is an emotion that arises in response to something perceived as repulsive or offensive, whether it be physical, moral, or social stimuli that are considered unpleasant or unacceptable (Ekman & Friesen, 1975). Sreeja and Mahalakshmi (2017) argued that disgust leads to negative behavioural responses, which in turn can result in critical reviews. In the context of podcast reviews, disgust is likely to associate with strong negative reactions and lower ratings.
Fear arises in response to a real or perceived threat and is related to survival instincts, prompting behaviours aimed at avoiding or escaping danger (Ekman, 1992). While fear generally leads to lower ratings due to the stress and anxiety it induces, there are contexts where it can engage audiences positively, such as in thriller or horror genres (Lee & De Villiers, 2025). Kwon and Park (2023) found that emotional responses to content varied significantly across different social media platforms, suggesting that the context of fear in podcasts might also influence its impact on reviews. Therefore, given that emotional responses to content can vary significantly depending on the context, fear might enhance audience engagement and, under specific circumstances, lead to higher ratings.
Joy is a feeling of pleasure and satisfaction that arises from positive events or achievements. It enhances social bonds and improves the quality of life (Ekman et al., 1990). Lyubomirsky et al. (2005) found that frequent positive affect, such as joy, leads to success across multiple life domains, including enhanced social relationships and overall well-being. This suggests that podcasts eliciting joy are likely to receive high ratings.
Surprise is an emotion that arises in response to unexpected events or information, focusing attention and prompting a reevaluation of the situation (Ekman, 1992). Surprise can be either positive or negative depending on the context, and it can lead to joy or fear based on the situation (Noordewier & Breugelmans, 2013). According to Sreeja and Mahalakshmi (2017), positive surprises can lead to favourable reviews by creating memorable and engaging content. Chan-Olmsted and Wang (2020) also highlighted that entertainment, often driven by surprising content, is a strong predictor of positive attitudes towards podcasts.
Sadness arises from negative events such as loss, failure, or disappointment. It provides time for emotional recovery and reorganization and can foster emotional connections and support from others (Ekman & Friesen, 1978). Chan-Olmsted and Wang (2020) found that negative emotions such as sadness detract from listener satisfaction, resulting in lower ratings. Fredrickson (1998) also suggested that while sadness can have short-term adaptive benefits, it generally leads to negative evaluations in the context of entertainment.
Having reviewed prior studies on how basic emotions influence listener responses, we turn to the broader concept of podcast engagement, which encompasses cognitive, social, and parasocial forms of involvement before focusing on emotional engagement.
Podcasts’ Dual Engagement
Across emerging media research, emotional engagement stands out as the decisive mechanism shaping intensity, loyalty, and expressive outcomes such as listener reviews (Chan-Olmsted & Wang, 2020). Recent podcast-specific work shows that intimacy and affect are actively constructed through host voice, first-person storytelling, and conversational delivery, which listeners experience as authentic and relational (Lindgren, 2023; Wang et al., 2025). Authenticity itself functions as an engagement cue that strengthens trust and parasocial bonds with hosts (Maloney Yorganci & McMurtry, 2024), while compatibility and satisfaction with podcast features that fit flexible, “anytime” use predict continued usage and word-of-mouth (F.-C. Yang et al., 2024). Notably, parasocial intimacy is observable directly in listener reviews, where changes in hosts or content trigger nostalgic and relational responses (Vilceanu, 2025). Building on work in consumer emotion, narrative processing, and electronic word-of-mouth (eWOM), current evidence suggests that emotional engagement is the central integrator through which other forms of engagement are amplified and translated into favourable and unfavourable evaluative behaviours (Berger & Milkman, 2012).
Emotional engagement is typically triggered by three antecedents: narrative quality, authenticity, and affective cues. Vivid, coherent storytelling elicits emotion and enhances attention and memory (Van Laer et al., 2014). Authentic performance by podcast hosts further intensifies affective connection by signalling sincerity and lowering perceived psychological distance, consistent with research on brand authenticity and emotional attachment (Moulard et al., 2015). Finally, affective cues – such as vocal warmth, humour, and vulnerability operate as emotional stimuli that strengthen resonance and empathy, aligning with foundational work on affect-as-information in consumer settings (Jeong, 2022; Pham, 2009).
Within the context of consumer podcast engagement, cognitive engagement reflects focused attention and elaboration, often facilitated by narrative structure and informational relevance (Van Laer et al., 2014). Social engagement manifests in community participation and conversational sharing, consistent with research showing that emotionally stimulating content increases social transmission (Berger & Milkman, 2012). Parasocial relationships are emotionally laden, one-sided relational bonds with media figures. Parasocial theory shows that repeated mediated exposure fosters intimacy, affinity, and trust, which in turn heighten emotional investment and loyalty outcomes (Horton & Wohl, 1956; Tukachinsky & Stever, 2019).
Affective responses exert disproportionate influence on downstream behaviours. That is, emotion acts as the amplifying force, shaping depth and durability of engagement. Such affective amplification may stimulate behavioural expression (e.g., review writing) even when it does not enhance evaluative favourability, suggesting a structural distinction between activation and judgement. In line with hedonic consumption research, emotionally resonant experiences create meaning, self-relevance, and attachment, increasing commitment to the media object (Holbrook & Hirschman, 1982; Thomson et al., 2005). In podcasting, this suggests that emotional engagement is not only a response, but a relational mechanism that transforms listening into a meaningful, identity-linked experience – especially when reinforced by parasocial intimacy.
An overt engagement outcome that this study investigates concerns review comments and review ratings by podcast listeners. Highly engaged listeners commonly externalise their emotional responses through expressive eWOM, including platform reviews. eWOM research in prior research shows that emotions, rather than cognition alone, drive both positive advocacy and negative backlash (Kim et al., 2016; Kuo & Nakhata, 2019). Favourable reviews arise when emotionally engaged listeners experience admiration, gratitude, or inspiration, often seeking to reinforce identity and signal affiliation to the host community (Berger, 2014). Conversely, unfavourable reviews are frequently rooted in emotional disconfirmation – anger, disappointment, or a sense of violated expectations – consistent with work on negative affect in eWOM (Kim et al., 2016; Kuo & Nakhata, 2019). This aligns with parasocial theory: when listeners feel relationally connected, breaches of emotional expectation provoke stronger negative reactions, making emotional engagement a double-edged antecedent of review behaviour.
In summary, evidence on basic emotions and podcast reviews underscores that emotions are pivotal in shaping listener engagement. Prior work finds that favourable ratings correlate with emotional impact, that is, listeners who feel emotionally moved are more likely to leave positive evaluations (Tobin & Guadagno, 2022). Negative emotions (e.g., sadness, anger) can also drive participation as controversial or emotionally charged topics attract attention, although excessive negativity can depress ratings (Qahri-Saremi & Montazemi, 2022). These insights motivate a closer look at how specific basic emotions map onto participation (review incidence/length) versus favourability (ratings).
Emotional Engagement in Podcasts and Other Domains
Emotional Engagement Theory has been applied in diverse fields such as ethics (Dunbar, 2005), work psychology (Xanthopoulou et al., 2013), and education (Park et al., 2012), all highlighting that emotional involvement can significantly influence decisions and performance. In the social media domain, Schreiner et al. (2021) reported the mediating effect of emotional responses, finding that arousing content is related to increased engagement behaviour. Their study applied the Circumplex Model of Affect proposed by Russell (1980), which uses the two dimensions of valence (positive vs. negative) and arousal (arousing vs. relaxing) to classify emotional response states. The model further classifies content engagement in three ways: (i) liking, (ii) sharing, and (iii) commenting on media content. These types of engagement are akin to three of the four classes discussed by Dolan et al. (2019). Schreiner et al.’s (2021) findings for social media reports show that arousal seems to enhance engagement behaviour whereas emotional valence interacts as a facilitator.
In a different online context, namely YouTube, it was found that emotional appeals influence consumer engagement positively. Emotions such as contentment and happiness were found to enhance engagement, while negative emotions decreased engagement (Kujur & Singh, 2018). Emotions in online product reviews influence customer engagement, revealing a bimodal distribution, with extreme reviews containing more emotional content impacting e-WOM generation and product evaluation (Ullah et al., 2016). Similarly, Tellis et al. (2019) found that online content rich in emotional cues (both positive and negative) was more likely to be shared widely, underscoring the link between emotional appeal and virality in digital marketing.
Emotional engagement affects memory, attention, and overall enjoyment, making it a vital component in evaluating media effectiveness. Illouz and Alaluf (2018) showed that the intertwining of rational and emotional aspects in consumer practices is a result of the increasing co-production of consumer actions, commodities, and emotional experiences. They emphasized how rationalization, commodification, and emotionalization have become intertwined in consumer decisions and choices.
Mixed Emotions in Podcasts
Mixed emotions refer to the simultaneous experience of two or more emotions with opposite valences (Larsen et al., 2001). Previous studies have compared cases where happiness and sadness coexist with cases where only happiness is present (Hong & Lee, 2010; Quach et al., 2021; Septianto, 2021). Other studies have examined mixed emotions involving hope and fear (Bee & Madrigal, 2013), joy and anger (Bee & Madrigal, 2013), pleasure and guilt (Ki et al., 2017), or fear and happiness (Andrade & Cohen, 2007).
It was found that mixed emotions can influence consumer behaviour such as satisfaction and loyalty (e.g., Olsen et al., 2005) by creating a more dynamic and compelling listener experience. According to Quach et al. (2021), mixed emotional appeals in advertising can lead to higher levels of positive word-of-mouth compared to purely positive emotional appeals. Similarly, Lee and de Villiers’ (2025) study on online book reviews found distinct emotional differences between fiction and nonfiction genres, with fiction evoking more intense negative emotions such as anger and sadness, whereas nonfiction displayed more joy.
Podcasts may also evoke a mix of emotions, leading to complex listener reactions. The combination of positive and negative emotions can create engaging content that prompts consumers to write reviews. Chan-Olmsted and Wang (2020) highlighted that the multidimensional nature of podcast consumption requires measuring emotional engagement from multiple aspects. This multidimensional approach is further supported by research indicating that factors such as vocabulary diversity, emotion, and syntax in podcast descriptions and transcripts are predictive of engagement (Reddy et al., 2021).
Others found that listeners’ engagement with podcasts can be influenced by their emotional responses to the content. A study by Hamilton and Barber (2022) on music podcasts found that listener reviews often reflect a complex interplay of motivations, including entertainment and emotional connection, which drive engagement. Reddy et al. (2021) also found that podcasts with positive sentiments and suspense are associated with high engagement: on the whole, high engagement is associated with more positive and less negative emotions and sentiment.
Overall, mixed emotional engagement in podcast content potentially plays a critical role in how listeners perceive, evaluate, and engage with content. The interplay of positive and negative emotions may create a richer and more nuanced engagement experience, leading to more detailed and reflective reviews. Beyond co-occurrence, mixed emotions may function as intensity modulators that shape whether emotional arousal reaches a participation-triggering threshold without escalating into avoidance. Next, we discuss this interplay of different emotions for utilitarian and hedonic podcasts.
Emotions in Utilitarian and Hedonic Podcasts
The importance of segmenting podcasts by motive, specifically utilitarian and hedonic, is critical for understanding listener engagement and the related emotions. According to the framework of Hollebeek and Macky (2019), consumer motivations based on utilitarian and hedonic aspects drive consumer interactions with digital content marketing. An earlier study by Scarpi (2012) focused on the interplay between utilitarian and hedonic motives in online environments, particularly how these two dimensions influence user behaviour and experiences on the internet. Scarpi’s (2012) findings align with those of other studies (Hu et al., 2022; Taufique et al., 2024) and emphasize the importance of both utilitarian and hedonic attributes in shaping consumer behaviour in web-based retail contexts and online services.
Another stream of research focused on marketing messages and online advertising appeals, as well as e-retail consumer appeals, for donations and online purchases (Luan, 2023; Rahman & Pial, 2019). It was found that a relationship between marketing communication and consumer cognitive and affective (emotional) responses can be linked to utilitarian and hedonic themes of marketing communication(s). Rational appeals, which emphasize product benefits and features, are more effective in influencing consumers’ purchasing decisions for utilitarian products (Rahman & Pial, 2019).
These prior research findings suggest that utilitarian podcasts that focus on practical information and benefits could effectively use rational appeals to engage consumers. In contrast, emotional appeals, which resonate more with social and relational aspects of podcasts, are relevant in contexts that evoke feelings of sociality (Rahman & Pial, 2019). We suggest that utilitarian podcasts provide functional value, such as educational content, news, or professional advice (Childers et al., 2001). Hedonic podcasts are designed to provide pleasure and entertainment. They include genres such as comedy, storytelling, and music, which are consumed for enjoyment and emotional gratification (Voss et al., 2003).
Podcasts that fall into the utilitarian category are designed to fulfil listeners’ needs for knowledge, professional development, or staying informed about current events. For instance, educational podcasts often use a structured format to deliver content effectively. Drew (2019) highlighted the importance of choosing the right podcast genre for educational purposes, noting that different genres such as chat shows, narratives, tutorials, and quick bursts have unique strengths and can be employed for various learning tasks. Furthermore, news podcasts, which are utilitarian in nature, play a crucial role in informing the public. Bird (2023) examined how news podcasts contribute to democratic participation in Australia, emphasizing that these podcasts enable listeners to engage in political actions and stay informed about important issues. The utility of such podcasts is evident in their ability to provide valuable, timely information that listeners can use in their daily lives.
Hedonic podcasts focus on entertainment and emotional engagement. These podcasts aim to provide pleasure, amusement, and emotional experiences to their listeners. The emotional engagement theory plays a significant role in understanding why listeners are drawn to hedonic content. Lindgren (2023) discussed how award-winning journalism podcasts leverage emotional storytelling to build intimate relationships between journalists and listeners, thus enhancing the emotional impact and engagement of the content. True crime podcasts are a prime example of hedonic content that combines entertainment with elements of suspense and ethical storytelling. Boling (2019) explored how true crime podcasts impact public opinion and the criminal justice system, demonstrating the genre’s ability to engage listeners through compelling narratives and ethical considerations. This genre’s engagement is driven by its ability to evoke strong emotional responses and keep listeners captivated.
Since utilitarian podcasts primarily fulfil practical needs, joy tends to be the dominant emotion that drives engagement, as it aligns with the satisfaction of acquiring useful knowledge or solving a problem. In contrast, hedonic podcasts, which focus on entertainment and emotional experiences, appeal to listeners who seek leisure and emotional connection. Hamilton and Barber (2022) explored how emotional engagement and entertainment value in music podcasts contribute significantly to listener motivation. Unlike utilitarian podcasts, hedonic podcasts evoke a wider spectrum of emotions, such as surprise, anger, or sadness, which enhance the entertainment experience. This aligns with Oliver and Raney’s (2011) expansion of hedonic motivations to include eudaimonic motivations, as the former embraces a diverse emotional range that deepens engagement. For listeners of hedonic podcasts, the variety of emotions beyond just joy creates a richer, more dynamic experience, keeping them more deeply connected and engaged with the content.
The Elaboration Likelihood Model (Petty & Cacioppo, 1986) can be applied to explain how emotional reactions influence listeners differently depending on podcast type (utilitarian, and hedonic podcasts). Utilitarian podcasts (focused on function and utility) encourage central route processing, where listeners scrutinize argument quality, structure, and factual clarity. On the other hand, hedonic podcasts encourage peripheral route processing, where cues such as voice tone, emotional expressiveness, and narrative pacing guide judgement. Emotional resonance (e.g., sadness, joy) may substitute for rational elaboration in forming judgements of podcast favourability. Petty and Cacioppo (1986) use the Elaboration Likelihood Model (ELM) to explain the mechanism through which emotions act as either facilitators or distractors of cognitive elaboration, depending on the content type and listener motivation (Petty & Cacioppo, 1986).
Prior literature frequently distinguishes between utilitarian (functional, informative) and hedonic (entertaining, affective) consumption motives (Batra & Ahtola, 1991). In podcast contexts, this distinction is increasingly blurred, as many formats combine information delivery with immersive storytelling. Podcasts often blend utilitarian and hedonic elements, making the binary classification overly simplistic. Importantly, this concern pertains to a strictly mutually exclusive (either-or) view of motives; it does not preclude categorizing content by its dominant motivational framing for empirical testing. While prior work often dichotomizes podcast content into utilitarian or hedonic, emerging literature (Oliver & Raney, 2011; Scarpi, 2012) suggests a more nuanced continuum of listener motives – ranging from instrumental knowledge-seeking to immersive, affect-laden experiences that serve eudaimonic needs. These studies do not reject categorical distinctions per se, but rather emphasize that motivational orientations vary in intensity and can coexist within media experiences.
Building on this insight, we recognize that utilitarian and hedonic motives may co-exist, but classify each podcast by the motive that is most salient (dominant) in its primary value proposition and genre conventions for analytic purposes.
Accordingly, this study classifies podcast listening motives into utilitarian and hedonic categories using a dominant-motive operationalization, consistent with the discussions of Scarpi (2012) and Oliver and Raney (2011). For instance, podcasts that provide news, education, or current-affairs information are categorized as utilitarian, whereas those centred on entertainment content – such as comedy or film reviews – are classified as hedonic. Although certain podcasts may incorporate elements of both motives, this study adopts a binary operationalization to represent analytically distinct processing regimes (central vs. peripheral) as specified by the ELM (Petty & Cacioppo, 1986). Thus, our categorization does not assume that motives are mutually exclusive in practice, but treats utilitarian and hedonic orientations as dominant processing frames for analytical clarity. This dominant-motive approach allows us to reconcile the continuum view of listener motivation with the need for theoretically grounded empirical testing. This distinction enables a theory-driven examination of how emotional signals are differentially weighted across motivational contexts. This distinction provides an empirical basis for clearly examining differences in listeners’ emotional responses.
Conceptual Model
The distinction in responses between utilitarian and hedonic podcasts not only helps in understanding listener preferences but also aids content creators in tailoring their podcasts to better meet the needs of their audience. Figure 1 presents a research framework tested in this study.
Influence of emotions on engagement and favourability by motive.
The model proposes that discrete emotions expressed in podcast review text (anger, disgust, fear, sadness, and surprise; with joy as the reference emotion) are associated with two distinct engagement outcomes: Participatory Engagement (review incidence/number of reviews) and Evaluative Favourability (ratings). This structure addresses RQ1 by examining how emotional composition relates to engagement outcomes.
Furthermore, podcast motive (utilitarian vs. hedonic) is theorized as a moderating condition that shapes how emotional expressions translate into these outcomes, pertaining to RQ2. In the model, the emotional composition of review text serves as the independent variable, podcast motive operates as a moderator, and engagement outcomes serve as the dependent variables.
This study grounds its theoretical framing in the Elaboration Likelihood Model (ELM; Petty & Cacioppo, 1986) to explain how emotions expressed in review text translate into engagement outcomes (e.g., review writing, star ratings). We posit that utilitarian podcasts primarily evoke central-route processing, where message clarity, argument quality, and informational utility dominate evaluation; in this route, emotions function as diagnostic cues but carry relatively less weight. By contrast, hedonic podcasts more often trigger peripheral-route processing, where affective heuristics, narrative tone, and experiential appeal heighten the influence of specific emotions (e.g., joy, sadness, anger) on participation and evaluations. Although the broader podcast landscape can span a continuum of motives, from information delivery to immersive storytelling, our empirical design operationalizes a binary distinction (utilitarian vs. hedonic) to maintain measurement clarity and interpretability. This ELM-based account specifies why emotional signals in text may differentially shape engagement across the two podcast types. Importantly, the model distinguishes between participatory engagement (number of reviews) and evaluative favourability (ratings), reflecting two related but structurally distinct outcomes. Consistent with ELM, we propose that emotional expressions in reviews may carry different behavioural implications depending on the processing regime associated with podcast motive.
Method
For emotion coding, we employ the Transformer Transfer Learning (TTL) model that was previously developed by Lee et al. (2023). Trained on more than 3.6 million sentences across 11 benchmark datasets, the TTL model achieves an average accuracy of 84 % and has demonstrated robustness across diverse short‑text domains, ensuring reliable emotion classification for the present dataset.
The relationship between emotions and two overt listener responses to podcasts is assessed using logistic regression analysis. We distinguish listener response along two complementary dimensions: behavioural engagement – operationalized as the number of written reviews per podcast – and attitudinal favourability – captured by the average star rating. Separating these dimensions enables us to disentangle how emotions drive participation versus evaluation.
Research Context: Rating Podcasts
Marketing practitioners regard “podcast ratings and reviews [as] an important part of your [brand’s] product packaging, and a form of social proof” (Misener, 2020, p. 1). Although several measures exist to rate social media and podcasts (subscriptions, review comments, star ratings, helpfulness ratings, number of committed fans), the most common and well-regarded measures are quantity of review comments and the star ratings offered by consumers of the product/brand and feedback on the helpfulness of the review comments. Apple reports that “Many listeners look at ratings and reviews before choosing to listen to or follow a show, and quality reviews can also help convey that your show has a community of committed fans” (Apple, 2024). This practitioner’s perspective is consistent with the academic literature on engagement (Dolan et al., 2019; Schreiner et al., 2021), which we discussed previously.
Concerning star ratings of podcasts, a study of 20 million podcasts by Misener (2020) found that the average star rating across all Apple Podcasts was 4.6 out of 5. In Misener’s study the average star rating count per podcast was found to range from 353 for True Crime, followed by Wilderness with 98 star ratings, to an average of only a single rating for Chemistry and Courses. Cricket podcasts received a surprisingly low average star rating of two alongside Mathematics (Misener, 2024). News Commentary scored an average of 64 in the US Analysis and Fantasy Sports 63 star ratings. These results suggest a wide spread of star ratings, and no particular pattern related to listener motive of hedonic-focused podcasts versus utilitarian-based podcasts.
Data
The dataset that we analysed has been made publicly available on Kaggle (Stuart Axelbrooke, 2023) and included 108,464 Apple podcasts with 2,017,209 reviews. The review period was December 9, 2005, to February 16, 2023. To assess engagement with the podcasts, we applied the following dichotomy for the number of reviews: <100 versus ⩾100, and for the star ratings: <5 versus having the perfect score of 5.
The distribution of podcast review counts shows that 32.8% of podcasts had only a single review, while 57.5% of podcasts had at least three reviews. Notably, the dataset includes only podcast episodes that receive at least one review. The top 10% of podcasts received 26 or more reviews. Only 3,331 podcasts had more than 100 reviews, representing 3.1% of the total. In a practical online article, Misener (2024) found an extreme upper tail of the distribution for the number of reviews that Apple podcasts receive, the top 5%, corresponding to 103 reviews in his sample. Consistent with this finding, we define high engagement as podcasts with at least 100 reviews. In our data, this threshold captures the extreme upper tail, 3.1% of the podcasts, distinguishing podcasts with exceptional audience traction from the majority with no traction or limited to moderate review activity.
The ratings for the podcasts ranged from 1 to 5, with an average score of 4.81, and 76.2% of the podcasts had a perfect score of 5. This is summarized in Table 1. Table 1 also provides statistics for utilitarian and hedonic podcasts, separately. Note that a perfect rating may be easier to achieve for niche podcasts with limited feedback, whereas podcasts with many reviewers rarely maintain a perfect score. We acknowledge this potential bias and apply corrections in the conducted logistic regression analysis in later sections of the paper.
Descriptive Statistics of Podcast Reviews.
As discussed previously, utilitarian podcasts provide information, education, or practical insights. They fulfil listeners’ needs for knowledge, professional development, or staying informed about specific topics. In our empirical study, categories classified as utilitarian included education, business, news, health and fitness, religion and spirituality, kids and family, technology, science, history, government, arts (books), and various subcategories within business and health and fitness. Additionally, podcasts about society and culture (documentary, relationships, philosophy, personal journals) and educational podcasts (how-to, self-improvement, language learning, courses) were included. News podcasts covering politics, tech news, sports news, business news, daily news, and entertainment news also fall into this category. Furthermore, podcasts focused on natural sciences, social sciences, life sciences, earth sciences, astronomy, physics, chemistry, mathematics, nature, society and culture, and religions (e.g., Razzaq et al., 2018) were considered utilitarian.
Hedonic podcasts are designed for entertainment, providing pleasure and enjoyment to listeners. These include categories such as comedy, sports (including football, basketball, wrestling, hockey, tennis, rugby, swimming, volleyball, cricket, and wilderness adventures), TV and film (after-shows, film reviews, TV reviews, film history, film interviews), and true crime. Arts-related podcasts (performing arts, visual arts, design, food, fashion, and beauty) and those related to leisure activities (hobbies, automotive, video games, aviation, animation, and manga, crafts) were also classified as hedonic in our empirical study. Additionally, music-related podcasts (music interviews, music commentary, music history), fiction (drama, science fiction, comedic fiction), and comedy (improv, stand-up) were included. Fantasy sports, home and garden, and other leisure topics were also grouped under hedonic. When distinguishing between hedonic and utilitarian activities, gardening is viewed as hedonic for its aesthetic pleasure, unlike farming, which is utilitarian due to its focus on productivity (Lu et al., 2016).
Some categories could contain both utilitarian and hedonic elements. For example, “news” podcasts often aim to inform (utilitarian) but can also entertain, and true crime podcasts, while rooted in journalism, are produced to captivate audiences emotionally. In our classification, we categorized each podcast by its primary intent (e.g., factual information vs. entertainment) based on genre descriptions and prior research (e.g., Boling, 2019; Lindgren, 2023).
In comparing the patterns of Utilitarian and Hedonic podcasts, several differences emerged (see Table 1). For Utilitarian podcasts with fewer than 100 reviews, the average rating was 4.83, with an average word count per review of 37.80, totalling 430,073 reviews across 58,349 podcasts. In contrast, Utilitarian podcasts with 100 or more reviews had a lower average rating of 4.65 and a higher average word count of 49.67, with 615,450 reviews across 1,724 podcasts. Similarly, Hedonic podcasts with fewer than 100 reviews had an average rating of 4.80 and an average word count of 35.98, amounting to 355,437 reviews across 46,784 podcasts. Hedonic podcasts with 100 or more reviews also showed a lower average rating of 4.62 and a higher average word count of 46.97, with 616,249 reviews across 1,607 podcasts. These patterns suggest that while Utilitarian and Hedonic podcasts with fewer reviews tend to receive higher ratings, those with more reviews tend to have more detailed feedback but slightly lower ratings overall.
Validation of the AI-Based Motive Classification (Hedonic vs. Utilitarian)
Because our theorizing treats podcast motive as a key conditioning mechanism (ELM-based central vs. peripheral processing), we verified the reliability of the AI-generated hedonic-utilitarian labels before hypothesis testing. We compared the AI motive label with four independent research assistants (RAs), who coded each podcast as hedonic or utilitarian based on (i) the platform category information and (ii) the top three podcast titles. Reliability was assessed using percent agreement, pairwise Cohen’s κ for AI-RA and RA-RA comparisons (Cohen, 1960), and Fleiss’ κ for multi-rater agreement across the four RAs (Fleiss, 1971), interpreted using the benchmarks of Landis and Koch (1977).
Inter-RA reliability was high (Fleiss’ κ = 0.839; N = 108 podcasts with complete RA labels). Mean pairwise RA agreement was 92.0% with mean Cohen’s κ = 0.839 (ranges: agreement 87.0%–94.4%, κ 0.739–0.889). Comparing the AI label to each RA (pairwise non-missing; N = 108-110), AI-human agreement averaged 93.2% (range 90.0%-96.3%) with mean κ = 0.863 (range 0.800-0.926). Using a stricter benchmark based on the RA majority vote (excluding 2-2 ties; N = 104), the AI achieved 97.1% accuracy with κ = 0.942. Together, these results indicate substantial-to-almost-perfect alignment between the AI labels and human coding, supporting the use of the AI-based motive classification in subsequent analyses. (The full category-to-motive mapping is reported in Web Appendix A.)
Emotion Detection in the Podcast Reviews
The consumer reviews of the podcasts were analysed using the Transformer Transfer Learning (TTL) model that was previously developed by Lee et al. (2023) for the purpose of detecting the six emotions of Ekman (1972) in short texts such as Tweets or online reviews. The TTL model used a novel two-stage learning process. In the first phase, Lee et al. (2023) trained the model on over 3.6 million sentences from four different datasets of self-reported emotions. This phase captured a broad spectrum of emotional expressions directly articulated by individuals. In the second phase, the model underwent additional training with more than 60,000 sentences from seven different datasets, where emotions had been rated by annotators, thus aligning it with socially recognized emotional expressions (Lee et al., 2023).
The resulting TTL model achieved a classification accuracy of 84% across 11 datasets, which are available online for academic research (see Lee et al. [2023] for further information about the development of the TTL model). Here we note that the 11 datasets span a wide scale of text types, ranging from headlines from Google News and CNN, to randomly generated sentences, to Tweets that were labelled by annotators or contained hashtags with Ekman emotions, and Reddit comments. The applicability of the TTL model to such a wide range of data source types supports the use of this model for also detecting emotions in podcast reviews.
We applied the TTL model and found that in the 2,017,209 reviews of the 108,464 podcasts, joy was the dominant emotion, 89.9%, with on average an additional 2.9% anger, 0.2% disgust, 1.9% fear, 3.2% sadness, and 1.8% surprise. Each podcast evoked a mix of emotions that added up to 100%. For example, “Honestly, this is just okay. I don’t feel like the hosts are particularly extra knowledgeable. And the mics one of the hosts uses is atrociously low quality,” where the emotion scores are as follows: Anger = 0.562, Disgust = 0.042, Fear = 0.003, Joy = 0.005, Sadness = 0.374, and Surprise = 0.013.
The TTL model was previously applied to detect emotions in Tweets about charity organizations (Lee et al., 2024) and online book reviews (Lee & de Villiers, 2025). In these applications joy was also dominant but less prominent than in the current podcast application. The relatively strong dominance of joy may be because Apple Podcasts feature engaging and uplifting content. For example, popular shows like “The Daily” and “This American Life” often blend informative and entertaining storytelling that enhances listener satisfaction, leading to positive reviews (Buzzsprout, 2024; Podcast Insights, 2024).
We found distinct patterns when comparing Utilitarian and Hedonic podcasts (see Table 2). Utilitarian podcasts with fewer than 100 reviews showed higher joy in the podcast reviews and less negative emotions expressed by listeners, as compared to those with 100 or more reviews. Similarly, Hedonic podcasts with fewer than 100 reviews displayed higher levels of joy and less negative emotions, while those with 100 or more reviews had reviews with relatively higher levels of anger, disgust, fear, sadness, and surprise. These descriptors highlight the variability in emotional content based on podcast type and review count and indicate that both Utilitarian and Hedonic podcasts with more reviews tend to have higher levels of negative emotions such as anger and sadness, and slightly lower levels of joy.
Emotional Distribution of Reviews by Motive and Engagement Level.
Five of the six emotions identified by Ekman (1972), that is, anger, disgust, fear, sadness, and surprise, were used as independent variables to predict two outcomes: (1) whether a podcast received at least 100 reviews, and (2) whether it achieved a perfect rating of 5 compared to lower ratings. The estimated logistic regression models provide insights into the effects of the less prominent emotions that may occur in different magnitudes next to the dominant emotion of joy. Note that including joy in the model would have resulted in multicollinearity issues, as the emotion scores add up to 100%, hence we used joy as the reference category in the conducted logistic regression analyses.
We estimated binary logistic regression models for each outcome. For the engagement outcome (whether a podcast had ⩾100 reviews), the model takes the form
Human Validation of the Emotion Detection Model
To ensure that the TTL AI model generalizes to the podcast-review domain, we conducted a human verification study on a stratified random sample of 600 reviews, selecting 100 reviews for each AI-predicted dominant emotion (anger, disgust, fear, joy, sadness, surprise). Four independent coders assigned a single dominant emotion using the same six-category scheme. A small number of coder entries violated the single-label rule (multiple or missing selections in 5/600 cases across coders); these cases were excluded only where required, e.g., for calculating the values on multi-rater statistics, preserving the maximum valid sample for each comparison.
Agreement between the AI model and individual coders was consistent. Across coders, AI-human agreement ranged from 74.8% to 79.0%, with Cohen’s κ = 0.70-0.75; average AI-coder agreement was 77.2% (mean κ = 0.73). Human coders also demonstrated strong internal reliability (coder–coder agreement 83.7%–92.0%, κ 0.80–0.90), with overall multi-rater reliability of Fleiss’ κ = 0.83 across 595 reviews with complete valid labels. When comparing the AI predictions with a stricter 3-of-4 human consensus label (excluding non-consensus cases; n = 543), agreement increased to 84.3% (κ = 0.81). Emotion-level agreement with consensus was highest for joy (99.0%) and high for anger (90.7%), disgust (89.6%), and sadness (88.8%), moderate for surprise (84.3%), and lower for fear (48.8%). Misclassifications were concentrated in semantically adjacent expressions where fear-related language overlaps with high-arousal affect in short evaluative texts. Overall, the verification indicates no anomalies that threaten inference and supports the use of the emotion detector for large-scale modelling of review volume and ratings.
Results
Analyses Results on Engagement and Favourability
Binary logistic regression and supporting tree-based CHAID analyses were conducted to investigate the impact of emotions on engagement. Note that we also conducted a Chi-square Automatic Interaction Detection (CHAID) tree analysis to complement the regression, which identified threshold values (optimal ranges) of each emotion that differentiate high- vs. low-engagement outcomes.
The first two logistic regression models (m1 and m2) predicted whether a podcast episode would receive at least 100 reviews. The explained variance for m1 in Table 3, with only main effects of anger, disgust, fear, sadness, and surprise, was only .010. Adding quadratic terms, m2, led to an explained variance of .279 with positive main effects occurring, and the quadratic terms had negative beta-coefficients. Thus, an inverted U-shaped relationship applies when predicting the occurrence of at least 100 reviews, as displayed in Figure 2(a). 1 The optimal ranges of alternative emotion intensity in this inverted U-shaped relationship were identified through tree-based analysis. We find that low levels of these emotions were associated with the highest engagement. The optimal range for anger was between .012 and .053, for fear between .010 and .034, for sadness between .023 and .068, and for surprise between .018 and .043. Disgust had a positive impact when it was greater than .001.
Logit and Tree Analysis Results on Engagement and Favourability.
Note. All reported beta-coefficients are from the models including the quadratic terms. We used the Nagelkerke R² to assess model fit (***p < .001).
Relationship types in the logistic regression results. (a) inverted U-shaped relationship and (b) declining negative impact.
The other two logistic regression models in Table 3, m3 and m4, predict whether a podcast will receive a perfect rating of 5. These models have one additional variable, namely the number of reviews that a podcast has received. This variable takes on the role of a control variable, as it is easier for podcasts with fewer reviews to receive exclusively five-star ratings. The explained variance of m3, with only main effects, was .438. Including quadratic terms in m4 increased the explained variance to .479, which is a much smaller increase than for the number of reviews in models m1 and m2 of Table 3.
The main effects in m3 and m4 were negative, while quadratic terms had positive effects, which does not support an inverted U-shape. Although, linearity approaches the relationship between the alternative emotions to joy and the rating scores better in m3 and m4 as compared to m1 and m2, the quadratic terms do significantly affect the probability of attaining the perfect podcast rating of five stars. Thus, diminishing impact applies as the emotions alternative to joy increase, as displayed by the relationship in Figure 2(b).
In summary, low levels of various emotions can increase the number of reviews a podcast receives, while achieving a perfect rating requires minimizing negative emotions. This suggests a managerial trade-off between generating a high volume of reviews and achieving high ratings.
Effects of Emotions for Utilitarian and Hedonic Podcasts
The models reported in Table 3 were run separately for hedonic and utilitarian podcasts. The logistic regression models with the dependent variable of receiving at least 100 reviews are presented in Table 4. For utilitarian podcasts, the explained variance for the model including only the main effects of the five emotions, m5 in Table 4, was .010, while the inclusion of quadratic terms, as in m6, increased the explained variance to .300. This again supports inverted U-shaped relationships. For example, the optimal range for anger in the utilitarian motive was between .007 and 0.040. Similarly, for hedonic podcasts, the main effects model m7 resulted in an explained variance of .010, increasing to .276 with the quadratic terms in model m8. Nevertheless, hedonic podcasts showed slightly broader optimal ranges for all alternative emotions to joy, except for fear. For ratings, both categories of podcasts indicated that lower levels of negative emotions are crucial for achieving a perfect rating.
Logit and Tree Analysis Results on Number of Reviews.
Note. Beta-coefficients are from models with the quadratic terms. We used the Nagelkerke R2 (***p < .001).
The four logistic regression models predicting whether a podcast receives a perfect star rating of five stars are reported in Table 5. For utilitarian podcasts, the explained variance was .431 when only main effects were included, see m9 in Table 5, increasing to .473 with quadratic terms, as in m10. For the hedonic podcasts, the explained variance was .453 for the main effects model, m11, and increased to .490 with the quadratic terms in m12. For both types of podcasts, the negative main effects and the quadratic terms with positive effects of a similar magnitude again imply a negative relationships emotion variable and ratings, that are stronger at lower levels of the independent variable (see Figure 2(b)).
Logit and Tree Analysis Results on rating by Motive.
Note. Beta-coefficients are from the models with quadratic terms. We used the Nagelkerke R2 (***p < .001).
In summary, the analyses reveal different patterns of emotional impact on engagement between utilitarian and hedonic podcasts. For the number of reviews, both types of podcasts benefited from low levels of negative emotions, in addition to a predominant joy emotion.
Conclusion
This study explored the nuanced roles of various emotions in influencing podcast engagement in terms of the number of reviews and rating, focusing on two primary categories: utilitarian and hedonic. The six Ekman (1972) emotions, fear, joy, disgust, surprise, anger, and sadness, were detected in podcast reviews using the TTL model that was previously developed by Lee et al. (2023, 2024). The emotional loadings of viewers’ podcast reviews were included as independent variables to explain scores on the two indicators of engagement in each of the 2,017,209 analysed podcast reviews.
Binary logistic regression and univariate tree-based analyses were used to examine the effects of different emotions on the number of reviews and the ratings. The findings indicated that moderate levels of various alternative emotions to joy positively influenced the number of reviews a podcast received. For both utilitarian and hedonic motives, emotions such as anger, disgust, fear, sadness, and surprise significantly increased the likelihood of a podcast receiving at least 100 reviews. The optimal ranges for these emotions varied somewhat between the two motives. For example, the optimal range for anger was between .007 and .040 for utilitarian motives, while it extended from.019 to 0.069 for hedonic motives. Fear had an optimal range of .012 to 0.036 for utilitarian and .008 to .030 for hedonic motives. Sadness and surprise also showed varying optimal ranges, indicating that moderate levels of these emotions can enhance engagement, with hedonic podcasts showing slightly broader optimal ranges.
In contrast, when examining ratings, the results revealed that achieving a perfect five-star rating required minimizing negative emotions for both utilitarian and hedonic podcasts. Negative emotions such as anger, disgust, fear, and sadness had significant negative impacts on the likelihood of a podcast receiving a perfect rating. Supporting evidence for these findings with regard to ratings can be found in the work of Felbermayr and Nanopoulos (2016), who demonstrated that emotions such as trust and joy played a significant role in the perceived usefulness of online reviews, with negative emotions leading to less favourable outcomes. Similarly, Ullah et al. (2016) emphasized the emotional content of extreme reviews, showing that reviews with more positive emotional content led to higher ratings, while negative emotions were less prevalent in positive reviews.
Beyond these empirical findings, this study contributes to the marketing literature by clarifying how emotional engagement operates as a behavioural driver in digital media contexts. In this study, emotional engagement is conceptualized not as a multidimensional engagement outcome, but as an affective state reflected in emotional expressions within reviews that is associated with and may stimulate observable participation behaviours. This distinction is important because engagement in marketing research is often treated as a broad construct encompassing cognitive, emotional, and behavioural dimensions. Our findings show that emotional activation can increase behavioural participation (number of reviews) even when it does not enhance, and may even reduce, evaluative favourability (ratings).
The inverted U-shaped relationship identified for number of reviews suggests that emotional intensity functions through a threshold mechanism. Consistent with optimal stimulation level theory (Berlyne, 1960; Steenkamp & Baumgartner, 1992; Steenkamp et al., 1996), modest levels of negative affect may increase arousal and perceived relevance, thereby prompting expressive action such as reviewing. However, as emotional intensity increases further, discomfort and avoidance tendencies may reduce participation. This finding extends prior work that often assumes monotonic valence effects by demonstrating that low, nonzero levels of negative emotion can maximize participatory engagement in podcast contexts.
Importantly, the divergence between number of reviews and ratings highlights a structural distinction between participatory engagement and evaluative favourability. Participation reflects behavioural expression that can be amplified by emotional arousal regardless of valence. Ratings, in contrast, represent consolidated evaluative judgements that are more sensitive to emotional coherence and the dominance of positive affect. This distinction helps explain why emotional volatility can stimulate discussion while simultaneously undermining perfect ratings.
Second, we highlight the importance of content context (utilitarian vs. hedonic) for emotional effects. ELM provides a useful interpretive lens for this difference: utilitarian shows more often trigger central-route processing, where argument quality dominates and non-joy affect is more likely to hinder evaluation; hedonic shows rely more on peripheral cues (tone, storytelling, affect), so a broader emotional range can boost participation without equally large penalties to favourability – up to an optimal level. While prior studies (e.g., Lee & de Villiers, 2025; Lee et al., 2024) have examined emotional drivers in other domains, our results specifically show that entertainment-oriented (hedonic) content can tolerate more emotional volatility without sacrificing audience involvement, compared to information-oriented (utilitarian) content. This insight contributes to the literature on hedonic vs. utilitarian consumption by demonstrating how emotional responses operate differently across these motives (Andrade & Cohen, 2007).
From a broader podcast marketing ecosystem perspective, these findings suggest that number of reviews and ratings serve different strategic functions on digital platforms. Review counts contribute to social proof, perceived popularity, and potentially algorithmic visibility, whereas ratings function as quality signals for prospective listeners. As a result, creators and marketers face a trade-off: emotional calibration that increases visibility through higher review activity may simultaneously threaten rating optimization. Recognizing this trade-off is particularly relevant in competitive podcast markets where discoverability and reputation jointly shape audience growth.
From a managerial perspective, our findings offer strategic recommendations for podcast creators and marketers. Practitioners can use these insights to make evidence-based creative and marketing decisions. For example, creators can calibrate emotional intensity when scripting or editing episodes, adding moderate tension or surprise to stimulate review activity while avoiding excessive negativity that could lower ratings. Marketers can segment campaigns by listener motive (utilitarian vs. hedonic) and monitor emotion-based KPIs (e.g., review rate, sentiment ratio) as part of performance dashboards. These applications translate abstract emotion-engagement patterns into actionable levers for audience growth and brand trust.
For podcasts aiming to increase the number of reviews, content that evokes moderate levels of negative emotions, such as anger or fear, can stimulate listener engagement. This is particularly relevant for hedonic podcasts, where a mix of emotions might foster deeper audience involvement. However, if the goal is to maximize ratings, particularly to achieve perfect scores, podcast creators must focus on reducing negative emotional content. Podcast marketers can use these insights to tailor content strategies based on their target audience and the nature of the podcast.
The finding that emotions exhibit an inverted U-shaped relationship for the number of reviews but have a negative effect on ratings has broader implications beyond the podcast domain. In digital media and online platforms, this pattern may explain the dual role of emotions in driving the number of reviews and podcast ratings. For example, in online product reviews, blog comments, or social media posts, low levels of negative emotions may prompt users to comment or share more frequently, driving up engagement metrics.
Theoretically, our findings extend the understanding of consumer emotional engagement in two ways. First, we empirically demonstrate the “too much of a good thing” effect in a new context: low, nonzero levels of negative emotions maximize engagement (number of reviews), supporting the notion that an optimal level of emotional intensity drives participation. Importantly, in this study emotional engagement is conceptualized as an affective driver associated with observable participation behaviour, rather than as an evaluative state. This pattern is consistent with optimal stimulation level theory (Berlyne, 1960): modest negative emotions affect can serve as a cue that draws attention and invites response, whereas excessive negativity adds noise that reduces motivation to engage. In consumer research, Steenkamp and Baumgartner (1992) and Steenkamp et al. (1996) found that responses are maximized at intermediate levels of stimulation. Our findings suggest that for podcast reviews the optimum occurs at low levels of negative affect, indicating a minimal activation threshold sufficient to prompt participation without inducing avoidance. Thus, low levels of anger, disgust, fear, sadness, and surprise may signal relevance and stimulate reflective processing. However, as the emotional intensity of these alternative emotions to joy increases, negative affect may induce discomfort and consequently avoidance tendencies, reducing the likelihood of engagement.
Second, we highlight the importance of content context (utilitarian vs. hedonic) for emotional effects. ELM helps explain this difference: utilitarian shows more often trigger central-route processing, where argument quality dominates and non-joy affect is more likely to hinder evaluation; hedonic shows rely more on peripheral cues (tone, storytelling, affect), so a broader emotional range can boost participation without equally large penalties to favourability - up to an optimal level. This finding demonstrates that responses are differentially weighted across processing regimes: in hedonic contexts, affective variability can sustain interaction, whereas in utilitarian contexts, emotional deviation from positivity more directly undermines evaluative judgements.
While prior studies (e.g., Lee & de Villiers, 2025; Lee et al., 2024) have examined emotional drivers in other domains, our results specifically show that entertainment-oriented (hedonic) content can tolerate more emotional volatility without sacrificing audience involvement, compared to information-oriented (utilitarian) content. This insight contributes to the literature on hedonic vs. utilitarian consumption by demonstrating how emotional responses operate differently across these motives (Andrade & Cohen, 2007).
This study has several limitations. First, although extensive, the database is limited to Apple Podcasts, which may not represent the full spectrum of podcast listeners. This focus could lead to biases based on user demographics and content preferences specific to Apple Podcasts. Second, the cross-sectional nature of the data per podcast limits the ability to infer causality between emotional engagement and podcast ratings. Future research could address this limitation by employing complementary methodological approaches such as controlled experiments or listener surveys to capture emotional engagement more directly. These methods would allow for triangulation of findings across behavioural, perceptual, and affective measures, thereby providing a richer understanding of the multidimensional motives underlying podcast participation. Third, demographic attributes (e.g., age, gender, cultural background) and other background variables (e.g., personality styles) that could shape both the valence and intensity of emotional engagement are not present in the publicly released dataset we analyse, which contains no personally identifiable information. As prior work suggests, these factors can moderate emotional expression and participation on social platforms (Fosch-Villaronga et al., 2021; Thakur et al., 2023). Because such fields are unavailable in the source data, we could not examine demographic or diversity-based heterogeneity in mixed emotions or engagement. Future research that combines demographic information could provide a more nuanced account of emotional diversity in podcast reviews. Fourth, we have no insights into the number of views that each podcast received. Including this and calculating the percentage of viewers leaving a review can provide additional insights into listener engagement for each podcast. Fifth, there is a possibility of endogeneity or reverse causality in our analysis. For instance, podcasts with highly engaged fan bases or particular audience demographics might both evoke certain emotional tones in reviews and receive more reviews or higher ratings independent of those emotions. Similarly, positive ratings might create a positive atmosphere that affects emotions in subsequent reviews. Our cross-sectional design cannot fully rule out these alternative explanations, so the observed relationships should be interpreted as associations rather than definitive causation.
Future research could address these limitations by incorporating data from multiple podcast platforms to ensure a more representative sample of podcast listeners. In addition, prior research indicates that review length (e.g., Ghasemaghaei et al., 2018) and various other emotions (e.g., happiness, nostalgia, or excitement) in review comments (e.g., Luan, 2023) could be meaningful areas for future research. Longitudinal studies could be conducted to observe changes in emotional engagement and ratings over time, providing insights into causal relationships. Additionally, integrating behavioural telemetry (e.g., listen duration, skip rates) and social media interactions would also provide a more comprehensive view of listener behaviour and help test whether the ELM-consistent optimal ranges we observe replicate across contexts.
Supplemental Material
sj-csv-1-anz-10.1177_14413582261442613 – Supplemental material for The Impact of Emotional Engagement on Podcast Reviews: An Analysis of Listener Responses Across Utilitarian and Hedonic Motives
Supplemental material, sj-csv-1-anz-10.1177_14413582261442613 for The Impact of Emotional Engagement on Podcast Reviews: An Analysis of Listener Responses Across Utilitarian and Hedonic Motives by Sanghyub John Lee, Leo Paas, Hyeyeon Yuk, Rouxelle de Villers and Pornchanoke Tipgomut in Australasian Marketing Journal
Footnotes
Acknowledgements
Hyeyeon Yuk was supported in part by the BK21 FOUR program at the Yonsei University School of Business.
Ethical Considerations
We retrieved the data from Kaggle, who made this publicly available for research purposes. All data available are anonymized.
Consent to Participate
Not applicable.
Author Contributions
Sanghyub John Lee drafted the manuscript and designed the study. Hyeyeon Yuk & Rouxelle de Villiers were responsible for revising the manuscript’s theoretical model development, clarifying its core contributions, and strengthening the managerial implications. All authors contributed to various conceptual discussions, reviewing and refining the results, revised the manuscript, and approved the final version.
Funding
The authors received no financial support for the research, authorship, and/or publication of this article.
Declaration of Conflicting Interests
The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Data Availability Statement
Data are available upon reasonable request.
Supplemental Material
Supplemental material for this article is available online.
Notes
References
Supplementary Material
Please find the following supplemental material available below.
For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.
For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.
