2.2.1 The Perceived Appropriateness of Expression

People consider the appropriateness of emotional expression contextually, where context encompasses several ideas. Context can mean the situation precipitating the emotion, whether the level of emotion observed is appropriate given the context (e.g., Gračanin et al. Reference Gračanin, Bylsma and Vingerhoets2018; Vingerhoets Reference Vingerhoets2012). Context can also mean the environment in which an emotion occurs; some people may perceive a certain space (e.g., work) as especially inappropriate for emotional expressions (Warner & Shields Reference Warner and Shields2009a). More recently, communication channels have also shaped perceptions of appropriateness: More private communication channels (e.g., text messages) are seen as more appropriate means of transmitting negative emotions than more public channels (e.g., social media, see Choi & Toma Reference Choi and Toma2022). And, context could mean who is expressing the emotion – research has long suggested that people respond differently depending on the identity of the “emoter” (e.g., Albertson & White Reference Albertson, White and Rudolph2022; Shields Reference Shields2005). Race, for example, can profoundly shape how people respond to expressions of anger (Phoenix Reference Phoenix2019).

Another contextual factor affecting perceptions of appropriateness is agreement with the reaction (Gračanin et al. Reference Gračanin, Bylsma and Vingerhoets2018; Kottler Reference Kottler1996; Vingerhoets Reference Vingerhoets2012). Because politics is often considered in terms of “sides” – whether partisan divisions or issue disagreements (e.g., Druckman et al. Reference Druckman, Klar, Krupnikov, Levendusky and Ryan2024) – agreement with the reaction may also proxy for political agreement.

In sum, the perceived appropriateness of emotional expression hinges on people’s beliefs about the context in which the emotion is expressed. We deliberately qualify the term “appropriateness” as perceived so as not to suggest that there are universal norms about emotional expression (e.g., Shields Reference Shields2005). And, as we will discuss later, it is this perceptual nature of appropriateness that often creates tension in how people evaluate and respond to the expressed emotions of others. The way people evaluate the expressed emotions of others speaks to the assumptions and beliefs of the person observing the expressed emotion about what constitutes the “right situation” and the “right place” (Shields Reference Shields2005).

2.2.2 The Perceived Authenticity of Expression

People have some ability to control how they describe or express the way they’re feeling – that is, they have some control of their emotional expression (Grandey Reference Grandey2003; Keller & Becker Reference Keller and Becker2021; Vingerhoets Reference Vingerhoets2012; Zloteanu & Krumhuber Reference Zloteanu and Krumhuber2021). Such control is a part of emotional regulation, or “the processes by which individuals influence which emotions they have, when they have them, and how they experience and express these emotions” (Gross Reference Gross1998, p. 275). Researchers have referred to this control as “acting” (Grandey Reference Grandey2003; Keller & Becker Reference Keller and Becker2021). Acting may include a person suppressing their tears at work, for example, waiting until they are in the car after work or home alone after a stressful day to cry (Warner & Shields Reference Warner and Shields2009a). On the other hand, acting could also mean exaggerating an emotional response. A person working a service job may attempt to seem especially happy for the sake of their customers (Grandey Reference Grandey2003). Notably, these examples foreshadow a tension pivotal to our manuscript: An observer only sees the external communication of emotion – the explanation of this expression is perceptual.

The idea that people can, in theory, regulate and “act” their emotional expression means that not all emotional expressions are “authentic,” and, perhaps more importantly, not all emotional expressions are perceived as authentic (Ekman & Rosenberg Reference Ekman and Rosenberg1997). This umbrella of perceived authenticity includes a variety of considerations. Judgments about the sincerity of the expressed emotion, the trustworthiness of the person expressing the emotion (Lee & Eastin Reference Lee and Eastin2021), and the perceptions about whether “one’s emotional expression display reflects one’s current emotional experience” (Ashforth & Tomiuk Reference Ashforth, Tomiuk and Fineman2000, p. 195) all comprise evaluations of authenticity.

How, then, do people form perceptions about whether an expression of emotion is authentic? People do try to distinguish between authentic and inauthentic emotional expression – even if they are not always successful (Hideg & Van Kleef Reference Hideg and Van Kleef2017; Wang et al. Reference Wang, Nguyen, Johnson and Groth2021). Central to this is the perception of the spontaneity of the emotion – is the expression something that “emerge[d] quite spontaneously from our construal of the situation” (Salmela Reference Salmela2005, p. 214)? Is the emotional expression “automatic” because a person is “living through it” – or is a person very deliberately trying to display and express this emotion (Salmela Reference Salmela2005, p. 214–215)? People read emotional expressions as less sincere and less authentic when the display appears to them to be effortful, rather than spontaneous (Keller & Becker Reference Keller and Becker2021; Pinheiro et al. Reference Pinheiro, Anikin and Conde2021).

These perceptions of authenticity play a central role in how observers respond to the emotional expression. Perceptions of authenticity increase both “compliance [and] cooperation” (Keller & Becker Reference Keller and Becker2021, p. 407), while people respond more negatively to emotional expressions that they perceive to be inauthentic (Keller & Becker Reference Keller and Becker2021). For example, people are less likely to trust or cooperate with partners who seem emotionally inauthentic (Krumhuber et al. Reference Krumhuber, Manstead and Cosker2007). Again, however, it is important to return to the tension at the heart of these perceptions: The extent to which people can accurately evaluate authenticity varies (Lechner & Paul Reference Lechner and Paul2019; McLellan et al. Reference McLellan, Johnston, Dalrymple-Alford and Porter2010). People are, generally, poor (and sometimes, biased) evaluators of the authenticity of others (Bailey & Levy Reference Bailey and Levy2022). In sum, perceiving an emotional expression as authentic is important – but again, it is just that, a perception.

2.2.3 Intersection of Perceived Appropriateness and Perceived Authenticity

Perceptions of appropriateness and authenticity could be somewhat orthogonal, though they are likely related. Shields (Reference Shields2005) offers an example: People may perceive happiness as an appropriate response when given a candy bar, but may come to perceive an extended, days-long expression of happiness about this gift to be less authentic. Similarly, perceived authenticity is not sufficient for perceived appropriateness. A person may fully believe that someone is authentically upset but perceive that sadness is inappropriate to the context. Ultimately, however, both perceptions of appropriateness and authenticity need to be in place for people to respond with empathy – and potentially action – to an expressed emotion.

Given the perceptual nature of both appropriateness and authenticity, it would not be surprising to suggest that a multitude of contextual factors likely shape how people arrive at their perceptions. In what follows, we focus on one such set of contextual factors: The communication channels through which people encounter the expressed emotions of others.

2.3.1 What Is the Expressed Emotion About?

People tell others about important events that happen to them and how they felt when these events occurred (Rimé Reference Rimé2009). This precipitating event is important: People consider the appropriateness of the emotional expression based on their own understanding of the event that caused the emotion (MacArthur & Shields Reference MacArthur, Shields, Hess and Hareli2019; Vingerhoets Reference Vingerhoets2012). Sometimes these inciting events have a clear near-consensus about valence. For example, people generally agree that a family member falling ill is a negative event (even if they disagree on how much emotional expression is appropriate in response). Political context is more complicated. People disagree about politics and, as a result, they may also disagree about the valence of a given political event and the appropriateness of a given emotional response. If a person believes a new policy is good, that person is unlikely to perceive expressions of anger and sadness about that same policy as appropriate.

It is possible that people may also disagree about the extent to which politics generally is an appropriate reason to express emotions (e.g., Krupnikov & Ryan Reference Krupnikov and Ryan2022). In this case, a person might perceive any emotional expression about politics as inappropriate. That said, even here, disagreement with the valence of a particular political event should still lead to still lower perceptions of appropriateness.

Expectation 1: When someone disagrees about the valence of the inciting event, they will perceive the expression of emotion as less appropriate than when they agree with the valence.

2.3.2 How Did the Emotion Get on Our Screens?

Central to the question of “how” is the idea of “access.” One definition of access is ease of use – requiring a particular app, for example, may make a channel less accessible (e.g., Choi & Toma Reference Choi and Toma2022). Here, we consider access as the extent to which a gatekeeper is required for this transmission (Barzilai‐Nahon Reference Barzilai-Nahon2009). A person has more access to a social media platform relative to a newspaper, where a journalist and an editor would be the gatekeepers. When a person has their own access to transmission and makes their own decision to share an expressed emotion, to some, the expression may seem more “effortful” (Keller & Becker Reference Keller and Becker2021), and, therefore, less spontaneous. Put another way, encountering someone’s feelings because that person decided to share them via social media may seem less spontaneous than encountering their feelings because a journalist decided to broadcast them.

Underscoring this idea of access is editability (e.g., Chen & Toma Reference Chen and Toma2024) – can a person “delete or polish messages before sending them via mediated channels” (p. 435). A person-on-the-street interview or reaction to an event shared with a journalist someone likes has “one take” – they cannot ask a journalist to rerecord an interview. They also cannot delete a published article with their comments. Communication channels without gatekeepers, however, often allow for more editability; people can revise, change, and perfect what they are about to share (Leonardi & Treem Reference Leonardi and Treem2012). Not all people edit their posts even when the affordances allow them to do so (e.g., Walther Reference Walther2007), nor are all “non-gatekept” channels equally editable. Yet these ideas of editability are, ultimately, tied to broader behaviors of self-presentation: The goal of perfecting and revising content is in service of “curating” how one is seen by the audience (Duffy and Hund Reference Duffy and Hund2019). Thus, a context that seems more editable may undermine the perceived spontaneity of expressed emotion, which, in turn, undercuts the perceived authenticity.

Expectation 2: When a channel does not require a gatekeeper, the expressed emotion will seem less authentic than when a channel does require a gatekeeper.

2.3.3 Why Did Someone Share This Emotion?

Communication channels vary in their visibility (e.g., Treem & Leonardi Reference Treem and Leonardi2013) – that is, they have differential “capacity to make messages visible to large audiences” (Choi & Toma Reference Choi and Toma2022, p. 141). Some channels (e.g., text messages) have “private message visibility,” while others (e.g., social media platforms) have “public message visibility” (Choi & Toma Reference Choi and Toma2022, p. 141). A “public message visibility” channel implies the possibility of reaching an audience that moves beyond a network of close ties to a much broader set of people.

We expect that emotions transmitted via channels that seem closer to “private message visibility” (e.g., text messages) may seem more authentic than the same emotions transmitted via a channel that has more public visibility (e.g., Instagram) (Waterloo et al. Reference Waterloo, Peter and Valkenburg2018). This expectation stems from a number of theoretical mechanisms. For one, people differentiate between the intended reach of different communication channels; channels with “public” visibility will reach a broader network of weaker ties and strangers, while channels with more “private” visibility are often designed to reach close confidants (e.g., Choi & Toma Reference Choi and Toma2022; Malhotra Reference Malhotra2024). In turn, people may perceive emotions transmitted to close confidants as more sincere (e.g., Hendriks et al. Reference Hendriks, Croon and Vingerhoets2008; Warner & Shields Reference Warner and Shields2009a). Another factor is that people may generally associate authenticity with these less visible channels (Kreling et al. Reference Kreling, Meier and Reinecke2022) and “performance” with more visible ones (Enli Reference Enli, Bruns, Enli, Skogerbo, Larsson and Christensen2015). This is perhaps why social media – with its “public message visibility” – is perceived as a means of achieving “self-promotion” (Duffy & Hund Reference Duffy and Hund2019). People also seem to see less persuasive intent in messages transmitted via less visible communication channels relative to those with more public visibility (Malhotra & Shin Reference Malhotra and Shin2025).

Ultimately, it is unlikely that the relationship between perceived authenticity and communication channel visibility is a function of only one specific mechanism. More likely, these mechanisms work in tandem. Our focus here, however, is on the role of channel visibility in shaping perceptions of authenticity.

Expectation 3: When a channel is perceived as more visible, the expressed emotion will seem less authentic than when a channel is less visible.

2.3.4 An Aside: Who Is Expressing the Emotion?

We have focused on conditions specific to the content of the emotional expression (the “what”) and the way in which the expression is communicated (the “how” and the “why”). Research, however, has also shown that the “who” matters – the person expressing emotions can shape perceptions of appropriateness and authenticity (e.g., Keller & Becker Reference Keller and Becker2021; Shields Reference Shields2005). A person’s gender and race, for example, can affect how people perceive their emotional response (Phoenix Reference Phoenix2019; Shields Reference Shields2005; Warner & Shields Reference Warner and Shields2009b). And, people are more likely to perceive those they like as authentic (Bailey and Levy Reference Bailey and Levy2022) – a pattern that could extend to the partisanship of the expressor. To further complicate things, the “who” may also extend to the identity of the observer – for example, some research suggests that men and women respond differently to the emotional expression of others (Stadel et al. Reference Stadel, Daniels, Warrens and Jeronimus2019).

In this Element, our focus is on the what, how, and why. In the first study we completed (Section 4, Study 1), we did consider the gender of the person expressing emotions. We did so because gender is one of the most commonly studied identities that moderates evaluations of emotional expression (e.g., Vingerhoets et al. Reference Vingerhoets, Boelhouwer, van Tilburg, van Heck, Vingerhoets and Conelius2001; Warner & Shields Reference Warner and Shields2009b). Moreover, a body of work suggests that gender may be important in emotional expression around politics (see Boussalis et al. Reference Boussalis and Coan2021), though empirical patterns are mixed (e.g., Brooks Reference Brooks2011).

To be clear, gender is only one of many dimensions of the “who” – and one that we consider largely as an initial “check.” Tracking how the “who” intersects with the other conditions we identify here is an important direction for future research.

2.4.1 A Focus on Climate Change

Our theoretical conditions apply broadly to political communication. In our empirical approach, however, our application is to climate change. As we note in Section 1, climate change is one of the most important issues of contemporary politics, is a highly emotional issue for some and, in recent years, expressed emotions on the topic have been present across social media platforms and news articles (e.g., Basch et al. Reference Basch, Yalamanchili and Fera2022; Smirnov & Hsieh Reference Smirnov and Hsieh2022). This issue context, we argue, establishes climate change as an ideal application to consider how people respond to the expressed emotions on their screens.

2.4.2 A Focus on Negative Emotions

We focus on expressions of discrete negative emotions: fear and sadness. Expressions of negative emotions are an especially interesting consideration for several reasons. First, in the context of social media platforms, norms point to more positive expression (e.g., Chen et al. Reference Chen, Yan and Leach2022; Waterloo et al. Reference Waterloo, Peter and Valkenburg2018), but negative emotional content is more likely to be spread (e.g., Hasell Reference Hasell2021). Second, research suggests that the negative emotions of others may be especially powerful. For example, expressing sadness through crying can facilitate bonds between people and mobilize help for the crier (Balsters et al. Reference Balsters, Krahmer, Swerts and Vingerhoets2013; Gračanin et al. Reference Gračanin, Bylsma and Vingerhoets2018; Kottler Reference Kottler1996; Stadel et al. Reference Stadel, Daniels, Warrens and Jeronimus2019; Vingerhoets Reference Vingerhoets2012). Observing others’ fear motivates attention and action (e.g., Derryberry & Tucker Reference Derryberry, Tucker, Niedenthal and Kitayama1994; Neubert et al. Reference Neubert, Kacmar, Carlson, Chonko and Roberts2008).

Speaking theoretically, questions of appropriateness and authenticity also apply to the way people perceive positive emotions (and their negative emotions, like anger). In our case, expectations about the “what” and “how” questions should generalize to positive emotions. It is possible that negative emotions may draw special attention to the “why.” Negative emotions, and especially sadness, are likely to draw sympathy (e.g., Hendriks & Vingerhoets Reference Hendriks and Vingerhoets2006; Vingerhoets et al. Reference Vingerhoets, Boelhouwer, van Tilburg, van Heck, Vingerhoets and Conelius2001). In turn, the visibility platform may exacerbate perceptions that a person is expressing negative emotions as a way of seeking an audience – that is, that the emotion is not authentic.

2.4.3 A Focus on Broad Policy

In addressing the intersection between the topic and the expression, we focus on expressions of fear and sadness about the potential future impacts of climate change, rather than responses to specific disasters. This focus on broad political policy – rather than very specific events – distinguishes us from previous work on the social sharing of emotions (e.g., Toma Reference Toma, Nabi and Myrick2023). In our studies, people express sadness about the possibility of climate change rather than about direct personal harms (e.g., lost homes) due to a disaster.

Despite the centrality of disaster in the national conversation about climate change, this choice was deliberate. How observers respond to victims of disaster varies along a multitude of factors, such as the characteristics of the disaster (e.g., Friedman Reference Friedman2019), the emotional predispositions of the observer (e.g., Andrews & Smirnov Reference Andrews and Smirnov2020), evaluations of whether observers believe victims should have taken more steps to protect themselves (e.g., Friedman Reference Friedman2019), or whether observers even believe climate change causes extreme weather events. Whether a direct experience with a climate disaster or general concern about future impacts precipitated emotional expression may play a role in how such expression is evaluated – but we leave this important consideration for future work.

2.4.4 A Focus on Perceptions, Not Content Creation

We want to underscore that we do not suggest that there is some objective standard or method for producing appropriateness or authenticity when expressing emotion about climate change. Indeed, we agree with Shields (Reference Shields2005) that evaluations of emotions are inherently subjective, perceptual, and likely “political.” Our goal is not to challenge the authenticity of people who communicate their emotions about climate change on social media – indeed, our treatments are designed with the assumption that the speaker is, ultimately, sincere. Moreover, it is not our aim to produce some objective catalog of the types of emotional reactions people should be having about climate change. Our focus is entirely on the way people perceive the expression of emotions in the context of climate change.

2.4.5 A Focus on Specific Social Media Platforms

In considering visibility, we focus on two particular social media platforms: X/Twitter and TikTok. There are other social media platforms where people can and do share their emotions, which we do not consider here (Leon Reference Leon2021). X/Twitter and TikTok are platforms that may be perceived as more “public” spaces for reaching wider audiences (Malhotra Reference Malhotra2025) – which allows us to consider expectations about visibility. These are also platforms that vary in their use of visuals and text (Hase et al. Reference Hase, Boczek and Scharkow2024), an idea we address in Section 6. Still, as platforms vary in the way they facilitate expression – for example, through platform norms (Thorson et al. Reference Thorson, Vraga, Kligler-Vilenchik, Hendricks and Schill2015), audiences and affordances (Literat & Kligler-Vilenchick Reference Literat and Kligler-Vilenchick2023) – our focus on two platforms is a limitation.

Of course, many people do not use either X/Twitter or TikTok. Indeed, for this reason, in some studies, we also present results conditioned on use. At the same time, use of social media platforms is increasing among all Americans (Tomasik & Matsa Reference Tomasik and Matsa2025). Moreover, exposure need not happen in the platform itself – posts from X/Twitter and TikTok may be cross-posted to other platforms (Barta & Andalibi Reference Barta and Andalibi2024) and embedded in other outlets entirely (McGregor Reference McGregor2019; Oschatz et al. Reference Oschatz, Stier and Maier2022). In short, someone who is not a TikTok user, but does, for example, read People Magazine or visit Facebook, may still come across the expressed emotions of a stranger.

3.1.1 The Person Expressing an Emotion

While many communication channels allow people to engage with their friends directly (Walther Reference Walther, Knapp and Daly2011) and even post their emotions for an assumed audience of close friends on social media (Toma Reference Toma, Nabi and Myrick2023), people also encounter the emotional expressions of strangers. These types of emotional encounters are more difficult to predict – there is no history, no relationship, and less certainty in the interpretation of the emotion (Walther Reference Walther, Knapp and Daly2011). Our goal is to consider what happens in an informational environment that increasingly features the emotions of strangers. We note, however, that this aspect of our treatment acts as a scope condition to our findings: It is possible (and, research suggests, likely) that people would respond more positively if the same emotion were transmitted by a friend or even an influencer or celebrity they liked (e.g., Haugseth & Smeplass Reference Haugseth and Smeplass2023; Mede & Schroeder Reference Mede and Schroeder2024). Alternatively, our results would likely be more negative if the message came from a known person who was strongly disliked.

3.1.2 The Emotion Expressed

As we note in Section 2, we focus on negative emotions: sadness and fear. The studies in Section 4 include both discrete emotions; the studies in Sections 5 and 6 focus on sadness. We consider these negative emotions (e.g., rather than anger) because these are the emotions most often felt in the context of climate change (Hickman et al. Reference Hickman, Marks and Pihkala2021; Leiserowitz et al. Reference Leiserowitz, Maibach and Rosenthal2024). They are also often expressed, as illustrated by the current focus on climate anxiety and climate grief (e.g., Kligler-Vilenchik & Literat Reference Kligler-Vilenchik and Literat2024). We also focus on sadness because it is an emotion that can be most easily conveyed in a nonverbal treatment.

3.3.1 Main Measures: Perceived Appropriateness and Authenticity

As we discuss in Section 2, people’s responses to the expressed emotions of others hinge on perceptions of the authenticity and appropriateness of these emotions. Our studies rely on outcome measures that capture these ideas (Table 2).

Table 2Main outcome measures. The question wording for the primary dependent variables used throughout this element. In most experiments, the target person expressing their opinions about climate change is named “Patricia Merrill.”⁹ There are minor variations in question wording across studies to best fit the context of the study, discussed in more detail in the respective sections.

Table 2Long description

The table shows how each outcome of interest is measured. The two outcomes of interest are perceived appropriateness and authenticity. Appropriateness is measured using a single question, while authenticity is measured using four. The questions wordings are as follows: Appropriateness: How appropriate is (Target Person’s) response to climate change? Authenticity (Sincere): Please indicate whether you agree or disagree with the following statement about (target person): (Target person) is being sincere. Authenticity (Trustworthy): Please indicate whether you agree or disagree with the following statement about (target person): (Target person) is trustworthy. Authenticity (Emotion): How (emotion) is (Target person)? Authenticity (Persuasion Attempt): Please indicate whether you agree or disagree with the following statement about (target person): (Target person’s) goal is to persuade.

3.3.2 Other Outcome Measures

In Studies 1 and 2, we track measures that consider other evaluations of the person expressing the emotion – such as their perceived knowledge of climate change (Study 1) and general affinity toward the expressor (Studies 1 and 2). Further, our two initial studies also measure posttreatment opinions about the importance of climate change. Finally, in Studies 5 and 6, we also include an open-ended measure to track people’s own explanations of their responses to treatments.

3.3.3 Pretreatment Measures

Each study measured pretreatment sociodemographic characteristics and political affiliation. These include age, race/ethnicity, income, gender, education, and partisan affiliation on a seven-point scale ranging from strong Democrat to strong Republican. The exact question wording is available in Online Appendix A.1. We also include predispositions toward climate change, which allow us to consider how people feel about the “inciting situation” – addressing the “what” question from Section 2. Participants are first asked to indicate whether they believe in climate change via the same measure that is used in the American National Election Studies and in the Yale program on Climate Change Communication’s annual Climate Change in the American Mind survey (see Ballew et al. Reference Ballew, Leiserowitz, Roser-Renouf and Maibach2019). Second, participants indicated how much they worry about climate change on a 1–10 scale. While typical measures of climate change worry only include five response options ranging from “not at all” to “very” worried, the ten-point scale allows for more variation in pretreatment worry.

In some studies, we also included additional pretreatment measures for preregistered tests of other heterogeneous treatment effects. Some came from previous research on response to emotions (e.g., perspective-taking, see de Waal Reference de Waal2008). Others were specific to our experimental context (e.g., use of social media and TikTok specifically). For each study, we preregistered looking at the effect of the treatment with and without controlling for sociodemographic characteristics. Additionally, we preregistered looking at treatment effects by specific pretreatment measures – most notably, belief in climate change.

4.2.1 Study 1 Design

Participants (N = 865, CloudResearch survey platform) read a short hypothetical newspaper article that included factually correct information from a Morning Consult poll along with a mention of a Morgan Stanley report discussing climate change. Morgan Stanley has issued a climate change report in recent years, and in 2023, the report raised the issue that a growing number of people are choosing not to have children due to fears about the increasing threat of climate change. Each article also included a quote from a ninth-grade teacher about the issue.

In this first study we completed, we manipulated the gender of the person quoted (either Patrick or Patricia Merrill). We included this manipulation as, given previous research, we anticipated questions about the role of gender. The manipulation in Study 1 serves as a check; given the results, we keep the identity of the expressor constant in the remaining studies (see section 4.2.2 for full results).

In the control condition, participants read the following:

The first paragraph – factually correct information – was consistent across all conditions. In the emotional conditions, we manipulated the quote and the description of the speaker in the second paragraph. In the sadness condition, the second paragraph instead read:

The fear condition was similar, with Merrill instead noting they “felt afraid,” were “overwhelmed” by fear, and became so “fearful” they had a hard time getting through the day. In this first study, we focus on appropriateness, asking respondents to evaluate the emotional expression across conditions.Footnote ¹⁰

4.2.2 Study 1 Results

To identify the effects of the treatment, we follow our preregistered analytic approach and rely on a series of regressions (full results and robustness checks are available in Online Appendix B.1).Footnote ¹¹ We consider our results in a series of steps. First, we treat the gender of the speaker as a nuisance factor (a factor that may affect evaluations of emotional expression but is not our primary focus). We focus on the role of emotions, tracking the effects of emotional expression on our focal variable – perceived appropriateness. Then, we consider the role of gender, tracking whether people respond differently to women expressing their emotions.

The Perceived Appropriateness of Emotions

People perceive the speakers quoted in news articles as significantly less appropriate than when they express an emotion (Figure 1, sadness: p = 0.01, fear: p = 0.03). This is not to suggest that people believe it is inappropriate to express emotion on climate change: Across all conditions, the quoted person is evaluated as more appropriate than not. Rather, this suggests that people perceive the expression of emotion to be a somewhat less appropriate response than stating concerns alone. We see no statistical difference between the expression of sadness and that of fear.

Figure 1 Predicted agreement that the speaker’s response to climate change is appropriate. Full results in Online Appendix B.2.

The Role of Gender

While many different characteristics might shape evaluations of the expressor, one possibility is that people may perceive and evaluate emotions differently when they are expressed by a woman rather than a man (Shields Reference Shields2005). Yet research on the intersection of gender and evaluations of emotional expression is mixed. In the domain of the emotional expression of political candidates, for example, some studies find that men and women candidates both lose support in equal amounts if they cry (Brooks Reference Brooks2011) and are equally rewarded when they express anger (Boussalis & Coan Reference Boussalis and Coan2021). Therefore, in Study 1, we also consider whether the gender of the person in the story shapes the way people respond to their emotional expression.

We regress evaluations of evaluations of the speaker’s appropriateness on an interaction between the emotion condition and the gender of the speaker while controlling for sociodemographic characteristics of the respondent (see Online Appendix B.2 for full coefficient tables). Doing so, we find no evidence that the gender of the speaker shapes evaluations (Figure 2). Expressing emotion decreases perceptions of appropriateness regardless of whether it is Patrick or Patricia who expresses emotions.Footnote ¹²

Figure 2 Predicted evaluations of the speaker’s appropriateness across all conditions. Full results are available in Online Appendix B.1.

Another possibility is that there could be interactive gender effects: Previous research suggests women are less put off by the emotional expression of all genders; however, men may find other men’s emotional expression inappropriate while being willing to assist emotional women (e.g., Warner & Shields Reference Warner and Shields2009b). Yet in exploratory analyses, we find no moderating effects of respondent gender (Online Appendix B.2).

4.2.3 An Emotional Vox Pop

Randomly assigning whether a “person on the street” expresses emotion when quoted by a journalist, we find that people perceive emotional expression as a less appropriate response to climate change. Notably, we do not find that the gender of the speaker (nor the gender of the receiver) shapes how people perceive the emotional expression. In other words, whether the “person on the street” was Patrick or Patricia, their emotional response in the context of climate change was still perceived as less appropriate than a response that contained no emotion.

This study, however, leaves open important questions. As we suggested in Section 2, perceptions of appropriateness are connected to people’s beliefs about the reason for the emotion – the “what” question. People may find emotional expression more appropriate when they agree that the “inciting incident” is actually negative. In our case, people may respond differently when they agree with the speaker’s position on climate change. In Study 1, we did not preregister analyses by participants’ climate change positions. Moreover, the design of the study was not well-suited to considering the role of the “what” – the position of the speaker was always constant and was not randomly assigned. Therefore, in Study 2, we consider the “what” question more directly.

4.3.1 Study 2 Design

Study 2 tests the moderating effects of agreeing with the position of the speaker. Much like in Study 1, participants (N = 1,200, YouGov) were randomly assigned to read news about climate change.Footnote ¹³ The news – this time set up as a transcript of a TV program – described a plan to teach climate change in middle and high schools. Much like Study 1, the news once again had a “vox pop.” Given our results in Study 1, in all conditions here and in the remaining experiments in this Element, the speaker was a woman, Patricia Merrill.

In this 2×2 design, we first randomized whether the speaker was emotional: All participants read a quote, but some were randomly assigned to see the speaker described as very sad. Second, we randomized whether the person quoted was opposed to discussing climate change in schools or in favor.

For example, the emotional, anti-climate change condition was as follows (the full text of the treatments and questions is available in Online Appendix B.2):

We highlight the “anti” treatment because it allows us to make an additional consideration: In this study, Patricia Merrill is not directly crying about climate change, but rather about the school policy. We can consider these treatments in Study 2 as a check on our effects: Do we see different patterns when the emotion is not directly about climate change itself, but about a specific policy?

Once again, our main measure is the perceived appropriateness of the response. Participants were also asked to indicate how strongly they agreed with the speaker’s position (which can also serve as a check on our pretreatment measures). We once again included measures about climate change concern and climate change behaviors, this time asking respondents how likely they were in the next six months to: Sign a petition, attend a rally or protest, talk to their friends and neighbors about the issue, and turn out to vote in favor of climate change. These measures identify whether encountering emotional expression has downstream consequences beyond just the evaluation of the expression.

Pretreatment, we asked participants about their climate change positions. Relying on their responses, we coded participants as in a consistent or inconsistent condition. First, participants who indicated that climate change “definitely is happening” or “probably is happening” were coded as believers, and those who said climate change “probably is not happening” or “definitely is not happening” as nonbelievers. Consistent with our preregistration, those who said “don’t know” were excluded from the analysis (112 participants). Then, believers who read about how climate change should be taught in schools or deniers who read about how it shouldn’t be taught in schools were coded as in consistent conditions. The rest of the participants were in inconsistent conditions.

There is research suggesting that evaluations of others’ emotions might vary based on observers’ perspective taking: the tendency to automatically feel the emotional state of others (e.g., Davis Reference Davis1980). Therefore, we included pretreatment perspective-taking measures and preregistered analyses that track the moderating role of perspective-taking.

4.3.2 Study 2 Results

In our analyses, we regressed our outcome measures on an interaction between whether a participant was in a consistent or inconsistent condition (given their position on climate change relative to the speaker’s support for teaching climate change in schools) and whether they were in the emotional expression condition. For ease of interpretation, all dependent variables were rescaled to range from 0 to 1. Before turning to our focal relationship – emotional expression and appropriateness – we can consider whether our categorization of conditions as consistent or inconsistent reflects our participants’ positions.

We find that participants in the consistent conditions agreed with the speaker more, and rated their response as more appropriate than in the inconsistent condition, regardless of whether the speaker was emotional. Treating the emotional expression manipulation as a nuisance factor, we see that appropriateness ( $\Delta$ = 0.28, p < 0.001) and agreement with the position ( $\Delta$ = 0.45, p < 0.001) are significantly higher in the consistent – relative to the inconsistent – condition. Therefore, we now turn to the main question at hand: the role of emotional expression (Figure 3).

Figure 3 Predicted evaluations of speaker appropriateness, agreement with the speaker, worry about climate change, and willingness to take climate action across conditions. Full results are available in Online Appendix B.2.

Agreement and (In)Appropriate Emotional Expression

In the inconsistent conditions – when people encountered someone they disagreed with – emotional expression made the speaker’s reaction seem less appropriate ( $\Delta =$ −0.060, p = 0.02). This is in line with our theory; appropriateness is highly perceptual, and a person crying over an idea with which one disagrees is less likely to seem like an appropriate emotional reaction. What is somewhat more surprising, however, is that emotional expression is seen as less appropriate than a more neutral response even when the observer and emoter hold consistent opinions ( $\Delta =$ −0.095, p < 0.001).

Certainly, the relative levels of perceived appropriateness are higher in the consistent conditions; perceived appropriateness, for example, is at its lowest point when the emotional expression is from a person with whom you disagree.Footnote ¹⁴ Still, we see decreases in perceived appropriateness across conditions. Even those who believe in climate change perceive emotional expression advocating for teaching climate change in schools as less appropriate than a more neutral response. And those who deny climate change are similarly put off by emotional expression from someone who wants to keep it out of schools.

As a next step, we consider our results another way – analyzing the “con” and “pro” treatments separately (Figure 4). In the “con” treatment, the fictitious parent is upset about the possibility of a climate change program in schools, while in the “pro” condition, the speaker is in favor. Rather than considering them in relation to respondents’ belief in climate change, we consider the conditions separately because the expressed emotions are not about climate change directly, but instead a policy related to discussing the issue.

Figure 4 Predicted evaluations of speaker appropriateness, agreement with the speaker, worry about climate change, and willingness to take climate action across conditions. Results pool across the climate attitudes of the respondents. Full results are available in Online Appendix B.2.

We find that in the “con” treatment, emotional expression leads to a significant decrease in perceived appropriateness ( $\text{Δ}$ = −0.081, p = 0.002). The negative effect of emotional expression persists even when the emotion is not about climate change itself, but about climate change in the curriculum (Figure 4). Still, this negative effect persists even in the “pro” condition ( $\Delta$ = −0.08, p = 0.001). Even when we limit the analysis to participants who believe climate change is real and are in the “pro” climate education condition, the group who, arguably, should be most sympathetic to the emotional expression, we still observe a negative response ( $\Delta$ = −0.09, p = 0.002, see Online Appendix B.2).

4.3.3 (Dis)Agreement and Emotional Expression

As expected, people perceive the emotional reactions of those with whom they agree to be more appropriate. Indeed, it is the emotional expression of a person with whom they disagree that is seen as least appropriate. At the same time, emotional expression leads to negative reactions compared to more neutral responses, regardless of agreement; an emotional reaction is seen as less appropriate even among people who agree with the general position of the “emoter.”

Both studies thus far leave an open question: How do people evaluate the gatekeeper who chose to include these emotional quotes? In our treatments, emotional expression is transmitted by the journalist, who describes the speaker as crying (or afraid, in Study 1). This rhetorical strategy is common in the news (e.g., Wahl-Jorgensen Reference Wahl-Jorgensen2020). At the same time, the idea that the description of the emotion comes from the journalist intersects with another idea we consider in Section 2: How did the emotion get on the screen? We will address the “how” directly in Sections 5 and 6; for now, however, we want to address a dimension of this idea: Does including emotions in the news shape people’s perceptions of the journalist – that is, the person who made the decision to highlight the expression of emotion?

4.4.1 Study 3 Design

Building on Studies 1 and 2, we randomly assigned participants (N = 1,000, YouGov) to read an article that included either an emotional or neutral quote. Each treatment noted the (hypothetical) journalist, Casey Smith.Footnote ¹⁵ In the control condition, the article included a neutral quote in which an interviewee mentions regularly thinking about bringing kids into an uncertain future brought about by climate change. In the sadness condition, the person quoted instead emphasizes how sad she is thinking about this uncertain future, and in the fear condition, she emphasizes being overwhelmed by fear. For example, the fear condition was as follows:Footnote ¹⁶

Much as would happen outside of the lab environment (Peters Reference Peters2011; Stenvall Reference Stenvall2014), the article included emotionality in two ways. First, the speaker being quoted expressed her emotion by saying she was “overwhelmed by fear.” Second, the journalist described the emotionality of the speaker, stating she “becomes so fearful she has a difficult time getting through the day.”

Participants were first asked whether they agreed that “Casey Smith’s article on climate change was appropriate.” Unlike Studies 1 and 2, the measure shifts perspective: Study 3 does not capture whether people believe the emotional expression itself is appropriate, rather the study considers whether the decision to include the emotional expression is appropriate.

Given our focus on the journalist, we also consider whether reporting on emotion shapes other perceptions of the journalist. Again, our goal here is limited to tracking people’s immediate responses to the journalist, rather than to offer a test of perceived media credibility (e.g., Jahng & Littau Reference Jahng and Littau2019). Therefore, we ask direct questions about the journalist’s perceived credibility and whether the participants would read an article by this journalist again. We also ask whether people perceive that the journalist is attempting to persuade them (Coleman et al. Reference Coleman, Thorson and Chen2025).

As we did in Study 2, we measured participant beliefs about climate change pretreatment. Participants who indicated climate change is “definitely happening” or “probably happening” were categorized as believers (831 respondents), while those who indicated climate change “probably is not happening” or “definitely is not happening” were categorized as deniers (169 respondents). Consistent with the preregistration, those who indicated they “don’t know” were excluded from the analyses (only 72 respondents).Footnote ¹⁷

4.4.2 Study 3 Results: Emotional Expression and Perceptions of the Journalist

Evaluations of the journalist do not change depending on whether or not they include emotional expression in their article (all variables in Online Appendix B.3). We see no evidence, for example, that including either a sad quote ( $\Delta$ = −0.002, p = 0.934) or a fearful quote ( $\Delta$ = −0.005, p = 0.830) changes perceived appropriateness. Although we see the largest effect sizes for perceptions of persuasion, the effects still do not meet conventional levels of significance for either sadness ( $\Delta$ = 0.026, p = 0.254) or fear ( $\Delta$ = 0.033, p = 0.142).

There are significant and largely unsurprising differences between how those who believe in and deny climate change evaluate the journalist across treatments. Holding the inclusion of the emotional quote as a nuisance factor, those who believe in climate change, for example, are significantly more likely to think the article is appropriate ( $\Delta$ = 0.29, p < 0.001) and think the journalist is more credible ( $\Delta$ = 0.30, p < 0.001). We note, however, that there are no differences by climate change beliefs in evaluations of whether the goal of the journalist is to persuade ( $\Delta$ = −0.010, p = 0.746); across climate attitudes, respondents believe the journalist is at least somewhat trying to persuade their audience. We see similar differences in climate change attitudes within conditions as well (full results in Online Appendix B.3).

We also see similar treatment effects within climate change positions: There are no significant differences in any of our other outcome measures. The null effects are also robust to combining both the sadness and fear conditions into a single emotional condition, compared to the control (see Online Appendix B.3). We underscore, however, the difference in sample size between the number of participants who believe in climate change and those who do not. Therefore, in the Online Appendix, we conduct a test of a negligible effect (e.g., Rainey Reference Rainey2014). We note that this test still leaves uncertainty due to the low sample size.

5.1.1 Study 4 Design

In Study 4, all conditions centered on the same Morgan Stanley report as in Study 1, which raised the issue that a growing number of people are choosing not to have children due to fears about the increasing threat of climate change. All participants were first told: “Morgan Stanley recently released a report that the decision to not have children owing to fears over climate change is growing and impacting fertility rates quicker than any preceding trend in the field of fertility decline.”

They were then shown a screenshot of someone’s response to this factual information. Across every condition, the response to this information came from someone named “Patricia Merrill,” who wrote: “People don’t want to have kids because of climate change. That makes sense! I read so much about climate change and get so sad that I honestly have trouble getting through the day sometimes.”

While the content remained the same across all conditions, we manipulated the communication channel through which this message was transmitted. Study participants were randomly assigned to see the quoted response as a private text sent to a friend, a comment on a news article, or a Twitter/X post (see Figure 5).Footnote ¹⁹ After exposure to the treatment, participants (N = 1,171, Prolific) evaluated Merrill’s authenticity using the set of measures discussed in Section 3: Sincerity, the extent to which Merrill is trying to persuade, the extent to which they are sad, and their trustworthiness. As we did in our other studies, we also measured climate change beliefs pretreatment.

Figure 5 Images of each condition. Participants were randomly assigned to read the response as a text message, comment on a news article, or a Twitter/X post.

Study 4 also included a fourth condition in which participants encountered the same content, but instead included as a quote in a news article. This fourth condition reflects the treatments in the studies in Section 4: It is the journalist who made their quote public. In this first empirical section of Section 5, our focus is on non-gatekept channels; therefore, we begin with a focus on the first three conditions. In the second half of Section 5, we will return to this fourth condition to consider how the presence of a gatekeeper shapes responses to the expressed emotion. All analyses, including this two-part analytic approach, were preregistered.Footnote ²⁰

5.1.2 Study 4 Results: Different Communication Channels

Contrary to our theoretical expectations in Section 2, we find limited evidence that these different communication channels undermine perceptions of the authenticity of emotional expression (Figure 6). Here we present results regressing each dependent variable on an indicator for each treatment condition, and in Online Appendix C.1, we present preregistered robustness checks with sociodemographic controls as well as results by belief in climate change. Results are substantively the same across each robustness check.

There are no significant differences in perceptions of authenticity on any of our measures between the text message and comment conditions. A comment on a news article is more visible than a text, yet participants did not think Patricia Merrill was consistently significantly less authentic on any dimensions: Trustworthiness ( $\Delta =-0.028, p=0.110)$ , sincerity ( $\Delta =-0.032, p=0.111)$ , or actually reflecting the stated emotion ( $\Delta =0.014, p=0.551)$ . On average, participants did lean toward thinking both were persuasion attempts – the average respondent fell somewhere between “agreeing” and “neither agreeing nor disagreeing” that Merrill’s goal was to persuade – but this response did not vary significantly between the text and comment conditions ( $\Delta =-0.006, p=0.780)$ .

Figure 6 Predicted evaluations of whether the message is a persuasion attempt, trustworthy, authentic, and whether the speaker is actually sad by the treatment. Results are from linear regressions, regressing each dependent variable on indicators for the treatment condition. Error bars are 95% confidence intervals. Full model results and robustness checks are available in Online Appendix C.1.

Compared to a text, posting to Twitter/X is substantially more visible (e.g., Choi & Toma Reference Choi and Toma2022). Still, we observe a few consistent differences in evaluations of authenticity between the text and social media conditions. Participants do perceive Merrill as somewhat less trustworthy ( $\Delta =-0.035, p=0.045)$ , and slightly less sincere in the Twitter/X condition compared to the text condition ( $\Delta =-0.032, p=0.104)$ , but these effects do not consistently reach conventional levels of significance. Effects on evaluations of whether Merrill is sad ( $\Delta =0.024, p=0.286)$ , and engaging in a persuasion attempt ( $\Delta =0.033, p=0.145)$ , are in the expected direction, but also do not reach statistical significance.

We also do not see any significant differences between the Twitter/X and the comment conditions. Still, comparing our conditions to the least visible communication channel, texting, we see that even as visibility increases, there is no consistent evidence of decreased evaluations of authenticity.

5.1.3 Communication Channels and Limitations

In Section 2, we argued that visibility would lead people to perceive expressed emotions as less authentic. Although we see suggestive patterns, we see little consistent evidence to this end. Further, we do not see that pretreatment characteristics – like our participants’ positions on climate change – shape their evaluations. There are different ways to consider these findings. One possibility is a limitation: These communication channels may vary beyond visibility alone – perhaps these other differences diluted effects. It is also possible, however, that visibility matters much less than we expected – in short, our theoretical expectations were incorrect. Moreover, although we showed the treatments as “screenshots,” the experimental context could not fully transport people into the communication channel, even as they understood that these communication channels had different levels of visibility. And, it is possible that the affordances that may most fundamentally alter perceptions of authenticity were not present in this study. Perhaps it is images and videos that matter most – for example, someone sharing an image of themselves crying (as news articles about crying on social media would suggest – e.g., Kaur Reference Kaur2025). We return to this idea in Section 6.

For now, however, we underscore that to this point our results do not follow our theoretical expectations in Section 2. As a next step, then, we consider the second part of our analysis – the extent to which the “how” matters. Specifically, we next turn to the fourth condition in our study: What happens when the emotional expression is transmitted via the news.

5.2.1 Revisiting Study 4: Access to the Gate

In the conditions discussed thus far, there was no gatekeeper, but the visibility differed. We now turn to our fourth condition: The same emotion is expressed via the news. This condition allows us to compare a case in which the decision to share the emotion is made by a gatekeeper, rather than the person expressing the emotion. The first three conditions are shown in Figure 5, and the new condition is shown in Figure 7. If assigned to this condition, participants were told that Patricia Merrill made the comment to a journalist. All outcome measures remain the same.

Figure 7 Images shown to participants in the news condition.

5.2.2 Does the “Gate” Matter?

We see limited evidence that “gating” affects how our participants perceived this expression of emotion. Following the preregistration, in the main text, we present results by condition without additional control variables. In Online Appendix C.1, all analyses are replicated with sociodemographic controls. In Figure 8, we focus on the comparison between the news and the Twitter/X condition – the conditions with, arguably, the largest audiences. When we compare the news condition to the other conditions – text messages and comments on a news article – we see similarly null results.

Figure 8 Mean predicted evaluation of the speaker on each dimension. Differences are as follows: Persuasion = −0.03 (p = 0.145); trustworthy = 0.001 (p = 0.942); sincere = −0.03 (p = 0.122); sad = −0.02 (p = 0.360). Point estimates are predicted values from OLS regression, regressing each dependent variable on an indicator for the treatment condition, and error bars are 95% confidence intervals. Full results are available in Online Appendix C.1.

There are no significant differences between the news quote to a post on Twitter/X and on any of our outcome variables. Regardless of whether the decision to share the expressed emotion was made by Patricia Merrill or a journalist, the emotional expression appears just as sincere ( $\Delta =0.03, p=0.132)$ and sad ( $\Delta =0.02, p=0.356)$ . Regardless of the communication channel, Merrill herself seems just as trustworthy ( $\Delta =-0.001, p=0.941)$ and aiming to persuade ( $\Delta =0.03, p=0.132)$ . Substantively, we see that participants are mostly middling in their perceptions of authenticity. For example, the average respondent falls somewhere between disagreeing and neither agreeing nor disagreeing that Merrill is trustworthy. They are similarly neutral, though lean toward agreeing, that Merrill is sincere and sad. We do not see evidence that the presence of a “gatekeeper” significantly shapes these perceptions.Footnote ²¹

5.2.3 Limited Role of the Gatekeeper

Ultimately, we do not find that the presence of a gatekeeper matters. The patterns point to a question. In Section 2, we note that it matters how the emotional expression made it to our screens; in particular, that when the decision to share the expressed emotion was made by another person (e.g., a journalist) and could not be edited by the “emoter” before sharing, people might view the emotion as more authentic. This idea, however, hinges on people viewing the journalist as the gatekeeper. Is it possible that people do not view the journalist in this way?

6.2.1 Study 5 Results: Appropriateness and Authenticity

Consistent with our preregistration, results in the main text regress each dependent variable on an indicator for the treatment condition, and in Online Appendix D.1, we replicate the results with sociodemographic control variables.Footnote ²⁸ Results are consistent across both sets of models. All variables are rescaled from 0 to 1.

Our results suggest that seeing the emotional expression can shape perceptions. The sad TikTok is perceived as significantly less appropriate than the neutral TikTok ( $\Delta =-0.075,$ p < 0.001) (see Figure 10). The sad TikTok is also perceived as less sincere than the neutral TikTok ( $\Delta =-0.11, p<0.001)$ , though we see no significant differences between these conditions on perceptions of whether Merrill is actually sad or is engaging in persuasion. Finally, the sad TikTok also leads Merrill to be seen as significantly less trustworthy relative to the neutral TikTok ( $\Delta =-0.061,$ p < 0.001, see Figure 11).

Although complicated by cross-platform comparisons, we do consistently see differences between the TikTok conditions and Twitter/X condition. Compared to the tweet, the sad TikTok seems less appropriate ( $\Delta =-0.059,$ p = 0.001), though there are no differences between the tweet and the neutral TikTok ( $\Delta =0.016,$ p = 0.364). Relative to the tweet, both TikTok conditions are perceived as less sincere (neutral: $\Delta =-0.06, p=0.001$ ; sad: $\Delta =-0.17, p<0.001)$ , less sad (neutral: $\Delta =-0.052, p=0.011$ ; sad: $\Delta =-0.067, p=0.001),$ and more likely to be engaging in a persuasion attempt (neutral: $\Delta =+0.083, p<0.001$ ; sad: $\Delta =+0.09, p<0.001)$ . The sad TikTok also makes Merrill seem significantly less trustworthy than the tweet ( $\Delta =-0.075,$ p < 0.001), though there is no difference in perceived trustworthiness between the tweet and neutral TikTok.

Figure 10 Predicted evaluations of speaker appropriateness, sincerity, and whether they are actually sad by treatment. Results are from linear regressions, regressing each dependent variable on indicators for the treatment condition. Error bars are 95% confidence intervals. Full model results and robustness checks are available in Online Appendix D.1.

Figure 11 Predicted evaluations of whether the speaker is engaging in a persuasion attempt and is trustworthy. Results are from linear regressions, regressing each dependent variable on indicators for the treatment condition. Error bars are 95% confidence intervals. Full model results and robustness checks are available in Online Appendix D.1.

Taken together, these results suggest that including a facial expression of emotion may not enhance the perceived appropriateness or authenticity of the expressed emotion – in fact, it seems to do the opposite. All our treatments included the same textual expression of emotion, but the facial expression of emotion alongside made the post seem less sincere and trustworthy.

6.2.2 Feelings about Feelings

At the end of Study 5, participants were asked whether they had any final comments. This question was open-ended and did not prompt our participants to write about anything in particular. This open-ended box was designed as a means of giving people an opportunity to comment on the study. Some participants left comments about the study itself – for example, noting that it was “fun.” Other participants left notes about how climate change is real, and we should all be worried (and the opposite, that climate change is a hoax, and they can’t believe people are posting about it on social media). Still other participants commented on the post they saw – and social media generally. It was not our a priori preregistered plan to consider these comments substantively,Footnote ²⁹ but looking at them at the conclusion of our study proved instructive.

Across all conditions, participants were equally likely to leave substantive comments about the treatments.Footnote ³⁰ What shifted by condition was the proportion of the substantive comments that focused on climate change versus the proportion focusing on the post in the treatment, or social media in general. In the Twitter/X conditions, among those who made comments, 82.8% focused on climate change. In the neutral TikTok condition, the majority still focused on climate change, but the proportion declined to 67.2%. Instead, nearly 30% of participants in this group focused on social media. We see the same pattern in the sad TikTok condition, where 31.1% of comments are about the post or social media.

In addressing the post itself, participants often turned to authenticity – and the authenticity of sharing emotions on social media generally. “I personally think that Patricia seemed very insincere and manipulative,” wrote one participant. Other participants agreed: “I believe that a person who posts things like this may only be trying to ‘present’ themselves as a vehemently concerned individual. I would also assume that she is hoping to get ‘sympathy’ from other people by stating that this ruining her days.”

Some of these comments, we note, focused on social media beyond the specific post in the treatment. One participant wrote that the “TikTok platform, robs the message of some of its sincerity to me.” Another noted: “I feel sometimes that people say things like that for social media clout. If they are sincere about climate change, they should not over exaggerate [sic].”

Some participants in the Twitter/X condition also raised doubts about social media; “I think most people put up a facade on social media, they won’t actually donate or contribute in any meaningful way,” wrote one participant. Still, these comments were more frequent when participants encountered TikTok.

To be clear, these comments are not about the quality of the treatment – that is, we see no pattern suggesting that the post seemed faked or generated by us as researchers.Footnote ³¹ Further, no one complained that the emotion expressed in the TikTok wasn’t sadness. As described above, the face images have been rigorously pre-evaluated to ensure they convey the target emotion (in this case, neutrality or sadness; van der Schalk et al. Reference van der Schalk, Hawk, Fischer and Doosje2011). Instead, people questioned the sincerity and appropriateness underlying the emotion. As we suggest in Section 2, people can identify the emotion as sadness but believe that it is “performance” rather than a sincere reaction.

6.3.1 Study 6 Results: Appropriateness and Authenticity

The patterns in Study 6 are similar to those in Study 5 (Figure 12). For ease of discussion, we term our treatments “no face,” “neutral face,” and “sad face.” Compared to both the no face and the neutral face condition, Merrill seems less appropriate in the sad face condition (no face: $\Delta =-0.08,$ p < 0.001, neutral: $\Delta =-0.07,$ p < 0.001). Moreover, she also seems less sincere (no face: $\Delta =-0.07,$ p < 0.001, neutral: $\Delta =-0.07,$ p < 0.001) and less trustworthy (no face: $\Delta =-0.06,$ p < 0.001, neutral: $\Delta =-0.05,$ p < 0.001). There are no differences in these measures between the no face and neutral face conditions. Like in Study 5, the inclusion of Merrill’s face increases perceptions that she is engaging in a persuasion attempt, regardless of whether her expression is neutral or sad (neutral: $\Delta =0.05,$ p < 0.001, sad: $\Delta =0.04,$ p = 0.007).

Figure 12 Predicted evaluations of the speaker. Results are from linear regressions, regressing each dependent variable on indicators for the treatment condition. Error bars are 95% confidence intervals. Full model results and robustness checks are available in Online Appendix D.2.

Merrill is seen as significantly more sad when the post includes a sad face compared to a neutral face ( $\Delta =0.20,$ p < 0.001) or no face whatsoever ( $\Delta =0.33,$ p < 0.001). Surprisingly, Merrill is also seen as sadder when the post includes a neutral face compared to no face ( $\Delta =0.13,$ p < 0.001). In both conditions, she is seen as somewhere between “a little sad” and “moderately sad,” though closer to only “a little sad” without any face included. This was unexpected, given that the images included in this study were pretested to ensure the neutral face indeed seems neutral (van der Schalk et al. Reference van der Schalk, Hawk, Fischer and Doosje2011).

6.3.2 Familiarity with the Platform

One possibility is that our effects may be due to the fact that the treatment is unusual or jarring for people who are unfamiliar with TikTok. People who are frequent users, however, may be more responsive and perceive them as more authentic. While we did not measure platform familiarity in Study 5, we did ask respondents whether they were familiar with TikTok in Study 6. Relying on our pretreatment measure of TikTok use, we see that, if anything, treatment effects are stronger among those who are more familiar with the platform. Although those who use TikTok do evaluate the no face and neutral face posts much more positively than those who do not use TikTok, all participants are equally skeptical of the sad face TikTok. For example, people who use TikTok believe Merrill is more trustworthy than those who do not use TikTok, as long as her post does not include her face (No face $\Delta =0.06,$ p = 0.02; Neutral $\Delta =0.09,$ p < 0.001). Once the post has a sad face, platform use has no effect on perceptions that Merrill is trustworthy ( $\Delta =0.03,$ p < 0.14). We observe a similar pattern in evaluations of Merrill’s appropriateness and sincerity. Platform familiarity does not moderate evaluations of whether Merrill is sad or engaging in a persuasion attempt (see Online Appendix D.2).

6.3.3 Even More Feelings about Feelings

After seeing their assigned social media post, everyone was asked: “Do you have any comments about the TikTok screenshot you just saw?” Like in Study 5, people left comments at similar rates across conditions (no face: 36%, neutral: 39%, sad: 41%). And, like in Study 5, people were more likely to leave comments about the post itself when it included a facial expression: (no face: 62%, neutral: 68%, sad: 73%). Unlike in Study 5, this was a preregistered open-ended question about our treatment (rather than an open-ended at the end of the study). We emphasize that this question was about the treatment, as this leads to substantively different types of responses from participants – for many, our question inspired comments on the structure/design of the TikTok screenshot (e.g., the font, background image, etc.).Footnote ³³

That said, the sad facial expression evoked a substantial amount of skepticism. Commenters noted it seemed like “virtue signalling” or “click-baity”[sic]. One noted that “She’s probably as fake as climate change is.” Negative evaluations of the post also spurred negative comments about posting on social media in general. For example, one commenter said: “I think the fact that people pretend to be organically sad and film, when we all know that they’re faking an emotion on purpose for attention is absolutely ridiculous.”

What is notable, however, is that some 13% of the comments either requested additional information about the IPCC report or complained that Merrill (or we, the experimenters) did not include information about the report. One commenter said: “It would have been better if I could have read the report from the Intergovernmental Panel.” Another complained, “Wish I had more information with the post.” Among those who left comments, people were less likely to request additional information in response to either the sad or neutral conditions (no face: 19%, neutral: 13%, sad: 8%). This is an unexpected result emerging from the comments – we did not preregister this possibility. Still, it is suggestive of a change in willingness to engage with the information (even when that engagement was complaining about the lack of information).

7.1.1 A Challenge in Identifying Visibility Distinctions

We theorized that the visibility of the communication channel – the “why” of sharing the emotion – would matter for the way people perceived expressed emotions. In Study 4, where we test these ideas directly, we find no evidence to this point. Results are similar across treatments, regardless of whether the emotion is expressed through a low-visibility (text message) or high-visibility (Twitter/X post) channel. In an additional analysis, we used Rainey’s (Reference Rainey2014) test of a “negligible effect” to more closely consider these results; across all measures, we find platform effects are not just null, but they are also “negligible” when compared to the effects in Study 5 (see Online Appendix B.3).

Given these unexpected results, we conducted an additional check with a different sample. We find that, as we had anticipated, people do believe that social media has a larger audience than other channels.Footnote ³⁶ Moreover, open-ended responses in Studies 5 and 6 suggest that people perceive social media to be more self-promotional than other channels – just as we argued in Section 2. The “medium, TikTok platform, robs the message of some of its sincerity to me,” wrote one participant in Study 5. The person in the screenshot in the treatments “seemed to really be trying to attract clicks,” wrote a different participant in Study 6.

We are left with the challenge of reconciling these results. One possibility is that visibility may not undermine perceptions of appropriateness or authenticity to the degree we expected. It is also possible that it is especially challenging to manipulate channel visibility in a study. A text message in an experiment may feel less personal and private than in real life, where someone knows their text is going to a closed set of people. And, a social media post in an experiment may, ironically, seem less public. A single social media post may also not have the same effect alone in an experiment, compared to outside the lab, where it would be nested in a richer information environment (e.g., Andersen & Ditonto Reference Andersen and Ditonto2018). Moreover, all treatments were read by strangers, given that the expressor is a hypothetical person. In the experimental context, the audience for a text message and the audience for a social media post may seem less starkly different.

It is also possible that, given the results in Section 6, modality matters more than we anticipated; evaluations of appropriateness and authenticity may be more sensitive to the inclusion of a visual depiction of the emotion. The inclusion of an image – especially an image that is, ostensibly, the poster themselves – could heighten considerations of authenticity (e.g., Lobinger & Brantner Reference 75Lobinger and Brantner2015). Emotions expressed through images may be perceived as more intense than those expressed through text. Indeed, research underscores people’s sensitivity to visual displays of emotion (e.g., Balsters et al. Reference Balsters, Krahmer, Swerts and Vingerhoets2013; Hendriks & Vingerhoets Reference Hendriks and Vingerhoets2006).

7.1.2 A Questionable Role for the Gatekeeper

We theorized that whether the emotion was shared by a gatekeeper (“how”) may also shape response. Yet we found that participants did not distinguish between channels requiring gatekeepers (newspapers) and those that do not (Twitter/X; Study 4). Including emotional quotes also did not seem to reflect on the gatekeeper (the journalist; Study 3).

Again, we turned to additional checks with a separate sample, finding that people do view journalists as gatekeepers. Attempting to reconcile these results, we again return to the same possibilities as in Section 7.1.1 – perhaps our delivery of the treatments “flattened” the types of differences that are so clear in people’s information environments. People may also have more general views toward “the media” and its credibility, making it difficult for them to distinguish individual actors – as they were asked to do in Study 3.

7.1.3 A Lack of Heterogeneous Effects

We preregistered tests of heterogeneous treatment effects. Given our focal issue, we measured people’s own positions on climate change. Other pretreatment measures captured ideas and suggestions that emerged as we worked on the project. We measured perspective-taking, for example, because research suggests that people vary in their ability to consider the emotional states of others (e.g., Davis Reference Davis1980; de Waal Reference de Waal2008). Early on in the project, people frequently asked about the role of age – would younger people respond differently, given their different relationship to social media? From the same perspective, we also considered whether a person’s social media use and familiarity matter.

Across our studies, we see little evidence that these individual differences shaped results. In some cases – like perspective-taking – we find no evidence that the measure played any role in the way participants viewed the treatments. In other cases, while these factors shaped the levels of the outcome variable, they did not shape the response to treatment. Put another way, people’s own positions on climate change (for example) correlated with whether they believed any comment on climate change was appropriate, but seeing emotional expression still shifted those perceptions in the same direction across the whole sample. In short, climate change positions mattered – but they mattered in people’s overall willingness to accept the general issue goal rather than in their response to emotional expression.Footnote ³⁷

Of course, these tests focus on only a small set of the potential characteristics that may moderate evaluations of emotional expression across platforms – there may be other factors that exert greater effects. Vulnerability to climate change, for example, could lead people to be more sympathetic to emotional expression; in other domains, people believe emotional expression is more sincere and appropriate when done over events that are more severe (e.g., Vingerhoets Reference Vingerhoets2012). Moreover, perhaps more fine-grained measures of social media use may uncover differential responses to emotional expression.

7.3.1 Whose Emotions Do We Expect on Our Screens?

At the start of 2025, The Atlantic published an article titled “Beware the Weepy Influencers,” written by Maytal Eyal, a practicing psychologist. The article focused on influencers, and contrasted the “vulnerability” of crying in a face-to-face setting with what Eyal termed “McVulnerability” – influencers “toggl[ing] between montages of sadness and sponsored videos that show them cozily sipping fancy tea” (Eyal Reference Eyal2025).

This focus on influencers is reasonable. Research, for example, has suggested that posting emotions can be a strategic signal of authenticity for influencers (Audrezet et al. Reference Audrezet, de Kerviler and Moulard2020). The realm of influencers and content creators has often addressed these questions of self-disclosure, emotion, and authenticity (Duffy and Hund Reference Duffy and Hund2019; Duffy, Ononye, & Sawey Reference Duffy, Ononye and Sawey2024). It is not a coincidence that our measures of authenticity are based on a scale developed to consider the perceived authenticity of influencers (Lee & Eastin Reference Lee and Eastin2021).

Does this connection between social media, influencers, and emotions pose a tension for understanding how people respond to the emotions they see on their screens? When The Atlantic shared Eyal’s article on its own Instagram account, the comments reflected research on the skepticism about the motives of others on social media (Luo et al. Reference Luo, Hancock and Markowitz2022). “If you choose to go on social and cry for the world we all know you’re just doing it for attention,” read one comment; “sharing emotions with friends and family used to be enough, now we live in an era where oversharing on social media is monetized for clicks and likes,” wrote another. If people have long been “pre-treated” (e.g., Druckman & Leeper Reference Druckman and Leeper2012) by the idea that influencers might strategically share emotions, that emotions are “for attention” and that social media is a realm of attempts to influence, is it possible for any emotion on our screen to seem authentic?

7.3.2 What Does This Mean for Climate Change Communication?

Climate change is an increasingly emotional issue for the American public. It’s perhaps unsurprising, then, that people are taking more time to express their feelings online. Doing so can help people find a community that also cares about the issue (e.g., Berryman & Kavka Reference Berryman and Kavka2018). Indeed, in writing about how young people express themselves on social media, Kligler-Vilenchik and Literat (Reference Kligler-Vilenchik and Literat2024) find that discussions of climate change are closer to “insider dialogue” than attempts to “convince others of the significance of climate change” (p. 117). Even without instrumental goals, expressing emotions makes people feel better. Whether telling a close friend, posting a comment on a newspaper article, or broadcasting across social media sites, sharing one’s emotional pain is cathartic (e.g., Vingerhoets Reference Vingerhoets2012).

But where instrumental goals do exist, our work offers a note of caution: Observers perceive expressions of emotions as less appropriate and authentic than more neutral calls to messages about climate change. Perhaps surprisingly, this pattern holds even among observers who themselves are very worried about the issue. To be clear, in most of our studies, people rate emotional posts as more appropriate than inappropriate, just less appropriate than more neutral posts. The takeaway then is not that people should keep their feelings to themselves, but instead that such expression won’t be universally well-received.

While we have considered perceptions of appropriateness and authenticity, we have left open the question of persuasion. If more emotional posts are received with more skepticism, does this mean they are also persuasive? Given that people operate in a media environment where they are inundated by information, but have limited attention and time to navigate this information, they are drawn to posts and messages with more emotional content (e.g., Hasell Reference Hasell2021). So, even if an emotional post is less persuasive than an unemotional post, if it better captures people’s attention, is there a cumulative persuasive effect that outweighs the less emotional content?

Within these expressions of emotion, there may also be gradations. Across contexts, people are much more sympathetic to negative emotional expression when the event that caused the expression was more severe (e.g., MacArthur & Shields Reference MacArthur, Shields, Hess and Hareli2019; Vingerhoets Reference Vingerhoets2012). For example, people are more sympathetic if someone cries after a divorce compared to after a relationship of only a few days’ ending. In some extreme instances, observers even think it’s inappropriate to not get emotional (e.g., Van Kleef Reference Van Kleef2016)! Our goal is not to argue that emotional expression is always seen as less appropriate and authentic, but instead to highlight that we can’t assume such expression will be taken at face value.