Listen for a change? A longitudinal field experiment on listening’s potential to enhance persuasion

Significance

High-quality, nonjudgmental listening is widely believed to enhance persuasion in interpersonal conversations by reducing defensiveness, improving perceptions of the persuader, and increasing cognitive processing. However, there has been little rigorous empirical testing of this assumption. Through a field experiment involving nearly 1,500 video conversations about immigration policy with trained canvassers acting as confederates, we found that while sharing a persuasive narrative durably changed attitudes, adding high-quality listening did not enhance these persuasive effects—even though listening improved interpersonal perceptions and reduced defensiveness. These findings challenge widespread assumptions about listening’s role in persuasion, though listening may serve other valuable interpersonal functions. This has important implications for how we understand attitude change and approach difficult conversations across divides.

Abstract

Scholars and practitioners widely posit that listening to other people enhances efforts to persuade them. Listening may enhance persuasion by promoting cognitive processing, reducing defensiveness, and improving perceptions of the persuader. However, empirical tests of this widely theorized hypothesis are surprisingly scarce. We review the case for and against this hypothesis, arguing previous research has not sufficiently attended to reasons why listening may not enhance persuasion. We test this hypothesis using a preregistered, well-powered field experiment in which trained professional canvassers, acting as confederates, had ∼10 min video conversations with U.S. participants (N = 1,485) about unauthorized immigration, a salient topic of disagreement. We independently randomized whether confederates shared a persuasive narrative about an undocumented immigrant and whether they practiced high-quality nonjudgmental listening to participants’ opinions. We measured outcomes immediately after the conversation and again five weeks later. Sharing a persuasive narrative meaningfully and durably reduced prejudice and changed policy attitudes. The listening manipulation also successfully improved perceptions of the persuader and increased processing. Surprisingly, however, the listening manipulation did not enhance persuasion: Sharing a persuasive narrative was just as effective in the absence of high-quality listening. We discuss theoretical and practical implications.

For centuries, robust interpersonal debate between citizens has been regarded as a cornerstone of democratic societies (e.g., ref. 1): People frequently engage in efforts to persuade one another about a variety of matters, such as endorsing particular candidates, advocating for specific causes, or shifting perspectives on contentious issues.

How can people most effectively persuade one another on such matters? Recently, several bodies of scholarship and a growing number of practitioners have posited that high-quality, nonjudgmental listening can enhance^* attempts to persuade them. For instance, research on door-to-door canvassing interventions has posited that when canvassers nonjudgmentally listen to voters, they are more persuasive (2). Relatedly, research on receptiveness suggests that high-quality listening behaviors communicate receptiveness (i.e., openness to others’ ideas), and may thus enhance persuasion by improving perceptions of the persuader and increasing elaboration (for review, see ref. 3)—an idea which dates to Aristotle’s concept of “ethos” (4, 5). As we discuss, practitioners and civil society leaders echo these expectations.

Despite this optimism about the potential for listening to enhance persuasion, there is surprisingly little empirical evidence on the matter. Indeed, as we review below and in SI Appendix, although there are widespread claims and theoretical reasons to believe that high-quality listening enhances persuasion, few studies have directly tested this hypothesis, and those studies’ results are less promising than one might expect. We also theorize that listening may be less important for facilitating persuasion than some prior research suggests. In particular, recent research suggests that people still learn from and change their attitudes in response to messages from intensely disliked sources offering sharply divergent perspectives (e.g., refs. 6–9). Such findings suggest that the mechanisms typically associated with listening may be less important for facilitating persuasion than commonly thought.

To investigate the potential of high quality, nonjudgmental listening to enhance persuasion, we conducted a large, well-powered field experiment. We hired professional canvassers as confederates and intensively trained them in high-quality listening (for shorthand, in the remainder of the paper we simply refer to this as “listening”). Over the course of four months, eight confederates worked an average of eight hours per week, ultimately having 1,526 one-on-one conversations with participants over Zoom. Confederates and participants discussed disagreements about a policy toward unauthorized immigrants, a salient and contentious topic. We manipulated whether confederates listened, and whether they made a persuasive appeal. Confederates’ persuasive appeal was in the form of a narrative, which past work finds can durably change attitudes (10, 11). We then measured attitudes both immediately after the conversation and five weeks later in a long-term follow-up, allowing us to test whether listening enhances persuasion in both the short and long term.

As expected, we found that making a persuasive appeal without listening meaningfully and durably reduced prejudice and changed policy attitudes. Adding listening to these persuasive conversations also improved perceptions of the persuader.

Surprisingly, however, adding listening did not enhance persuasion. This may be because participants processed the persuasive appeal even when they were not listened to: They learned just as much from the conversations when they were not listened to and reported elaborating just as much, even though they liked the persuader less and felt more defensive. Underscoring the surprising nature of these results, the confederates who implemented the study themselves did not expect them. In Discussion, we discuss limitations of these findings, potential future research, and the theoretical and practical implications of our work for efforts to persuade across divides.

Listening and Persuasion.

A consistent theme in research on persuasion is that people often do not accept messages contrary to their prior viewpoints or from sources they distrust or dislike (e.g., ref. 12).

In contrast, listening is widely expected to increase acceptance of persuasive messages, even from disliked sources, and therefore to enhance the effects of persuasive attempts. High-quality, nonjudgmental listening behaviors are thought to be particularly effective for that goal (e.g., refs. 2, 13, and 14). For example, field studies on the persuasive power of interpersonal conversations have speculated that high-quality, nonjudgmental listening is a crucial facilitator of persuasion in those conversations, although this work did not manipulate its presence (e.g., ref. 2). Theoretical work on receptiveness—or behaviors like listening that signal openness to a counterpart’s views—has similarly speculated that such “acts of receptiveness can promote persuasion” (3, p. 230). More generally, other recent scholarship has posited that listening is a way of bridging divides, in part through persuasion (15, 16).

High-quality, nonjudgmental listening consists of paying attention and comprehending what a counterpart communicates, and in turn engaging in both nonverbal and verbal observable behaviors that demonstrate understanding and affirmation of the other’s perspective (13, 14, see also SI Appendix, Table S1 for other definitions of listening), such as asking open-ended, follow-up questions and avoiding interrupting and judgmental language.

Why might listening behaviors enhance persuasion? First, by acknowledging the other’s viewpoint, listening may satisfy basic psychological needs of acceptance, thereby reducing defensiveness and increasing openness to a counterattitudinal arguments (17, 18). Second, listening may increase interlocutors’ self-reflection. This could directly lead to attitude change (14) by increasing cognitive engagement with the message (3, 19). Third, targets who are listened to by a persuader have more favorable impressions of the persuader, as listening increases liking (20, 21). In turn, the more someone likes and trusts their counterpart, the more they may be persuaded—either due to source cues (under low elaboration) or biased processing (under high elaboration) (3, 19).

These arguments are supported by the receptiveness literature (22–24). Receptiveness is defined as “behaviors or actions that signal a person’s openness to ideas, arguments, and attitudes that are new or opposing to their own” (3, p. 229). Nonjudgmental and high-quality listening both explicitly and implicitly communicate openness to the other’s ideas—i.e., receptiveness—by asking questions and affirming the other (a supposition we confirm in our study). In turn, this may enhance persuasion for these same reasons of increased processing and improved interpersonal liking (3). (See SI Appendix, Table S2 for review.) These ideas partially date back to Aristotle’s concept of “ethos”—that, among other things, liking a persuader more may increase openness to their persuasive message (4).

Political practitioners also espouse the power of listening. Outlets such as the New York Times have released features encouraging people to deploy listening in political conversations to be maximally persuasive. Nonprofits such as the Listen First Project and Braver Angels echo similar sentiments, and listening has likewise been used to attempt to facilitate persuasion at scale in ballot measure elections and presidential campaigns. Politicians and civic leaders also endorse the power of listening to produce persuasion: In his 2024 Democratic National Committee speech, President Barack Obama said “if we want to win over those who aren’t yet ready to support our candidates, we need to listen to their concerns.”^†

But despite the widespread assertion that high-quality nonjudgmental listening and receptiveness enhance attitude change, there is remarkably little research testing this hypothesis in general, and especially in the context of persuasion. Existing work does not speak directly to whether these behaviors enhance persuasion in interpersonal conversation. Some research is concerned with the direct effects of these behaviors but does not investigate whether they enhance persuasion (e.g., refs. 14 and 25). Other research examines these behaviors in contexts that are outside of live interactions, such as asynchronous conversations (see SI Appendix for further review).

Why might listening fail to enhance persuasion? Other literature offers reasons for doubt. In particular, in contrast to Zaller’s (12) model’s emphasis on whether individuals “Accept” a persuasive message, recent research suggests that people learn from and change their attitudes in response to messages even from intensely disliked sources offering markedly diverging perspectives (e.g., refs. 7–9). This research draws a sharp distinction between a) attitudes toward the sources articulating a viewpoint on an issue, and b) attitudes toward that issue. Regardless of the source of the message—even if that source is intensely disliked—this body of research suggests a message itself can still be persuasive. For instance, Broockman and Kalla (6) find that extremely conservative Fox News viewers incentivized to watch CNN were meaningfully persuaded by its content, even though they reported intensely disliking and distrusting it. Such findings suggest that one of the key mechanisms listening is expected to engage—increasing liking of the source of a message—may be less important for facilitating persuasion than commonly thought.^‡ However, this work neither studies interpersonal conversations nor has it manipulated affect toward sources, and so it remains possible persuasion could be facilitated by increasing positive sentiment toward a source in conversation.

Listening may also fail to enhance persuasion in cases when targets are sufficiently motivated to process a persuasive message anyway. This may be particularly likely to be the case when discussing topics that people care about (such as when discussing a politically sensitive topic like immigration, e.g., ref. 19) or when the persuasive appeal communicates novel information in a captivating way, such as an engaging narrative might (e.g., ref. 11). (See SI Appendix, Table S2 for review.)

Three important studies that are most directly related to our work in the literature—all of which focus on the short-term effects of text-based conversations—reinforce these doubts. First, a small study ($N=24$ per cell) with American undergraduates by Chen et al. (26, Study 1) found that in a text-based online chat with a confederate who argued for an opposing view, a confederate asking an elaboration question after an argument increased perceptions of the confederate’s receptiveness but did not affect views on the policy. Minson et al. (26, Study 3) examine the effect of being more receptive in text-based persuasive conversations about vaccination. They found that receptiveness improved attitudes toward the persuader (e.g., trust) but did not result in different attitudes toward vaccines. Finally, Argyle et al. (28) studied text-based discussions between people who disagreed about gun regulation. Some participants received AI-assisted rephrasing of their messages to convey greater understanding and politeness. Though this improved their counterparts’ perceptions of and respect for them, it did not change their counterparts’ attitudes toward gun policy. In short, only three text-based conversation studies have studied whether persuasion can be enhanced through listening-related behaviors (such as conveying receptiveness), and none have found evidence for this. To our knowledge, no studies have used live conversations to investigate this. As we describe in SI Appendix, we build on these and other studies by studying live face-to-face (video-based) conversations, using confederates extensively trained in listening behaviors, and measuring long-term outcomes in a large-N field trial.

Data and Methods

This research was approved by the University of California, Berkeley Committee for the Protection of Human Subjects, the Columbia University IRB, and the Yale University Human Subjects Committee. Informed consent was obtained from all participants. The data and code is publicly available, see ref. 29.

In our study, 1,485 participants recruited through social media advertisements had approximately 10-min Zoom conversations with highly trained confederates and completed post-conversation measures. Participants discussed a disagreement about a policy toward unauthorized immigrants: whether or not unauthorized immigrants should get access to in-state college tuition at state colleges and universities. Attitudes about this issue and prejudice toward undocumented immigrants served as our primary dependent variables, following previous work by Kalla and Broockman (30).

Procedures.

We next provide an overview of our procedures. As this study was logistically complex, we review further details in SI Appendix, including details on confederate recruitment and training materials, participant recruitment materials, the technology we developed to allow participants to book times with confederates and implement their random assignment, participant incentives, and additional demographic details on participants.

Recruitment.

We recruited participants via Meta advertisements (see SI Appendix, Fig. S1) to take an initial baseline survey during the spring of 2024. Those who expressed anti-immigrant views on the baseline survey ($N=3,944$) were invited to sign up for a 20-min time slot to have a conversation with a “fellow participant” ($N=2,667$). In reality, this person was a trained confederate with previous professional experience as a door-to-door canvasser. When participants returned at their scheduled time ($N=1,872$), each was directed to their own Zoom room with a confederate.

In the Zoom room, the confederate announced that they were randomly selected as the person who was given instructions about how the conversation should unfold, and read instructions according to the assigned experimental condition. These instructions introduced the topic: whether or not undocumented immigrants should receive access to in-state college tuition at state colleges and universities. This was the first time the participant learned of the conversation topic. In all conditions, the confederate shared that they supported the policy and asked the participant to very briefly state their opinion on a 0 to 10 scale where 10 $=$ strongly support and 0 $=$ strongly opposed. In other words, in every condition, both people stated their position, establishing that they disagreed on the topic. The average rating participants gave at the start of the conversation was 3.8 out of 10, suggesting that participants communicated their disagreement; this also did not differ by condition ($P=0.551$). The participant and confederate then had a conversation ($N=1,526$ participants who began a conversation, of whom 1,485 completed post-conversation measures), with the confederate following the random assignment laid out in the instructions (SI Appendix).

Experimental conditions.

For the experimental conditions, we independently randomized whether the confederate attempted to persuade the participant and whether the confederate nonjudgmentally listened to the participants’ views. This produced the four conditions summarized in Table 1.

Table 1.

Experimental conditions

		Listen?
		No	Yes
Persuasive	No	Placebo	Listening only
Appeal?	Yes	Persuasive appeal only	Listening + persuasion

The first factor of the design was whether or not the confederate engaged in high-quality, nonjudgmental listening to the participant’s opinion about the immigration policy. When they did so, confederates asked multiple probing follow-up questions, nodded their head, made eye contact, and avoided expressing judgment (13, 14, 31, 32). This treatment changed the behavior not only of the confederates, who listened, but also naturally changed participant behavior, leading them to talk and share more about their perspective (see SI Appendix, Table S9 for average speaking times of both participants and the confederate by condition). In the conditions without listening, the conversation moved on after participants briefly stated their opinion on a 0 to 10 scale. We chose this design to manipulate high-quality listening (and, by extension, receptiveness) as strongly as possible; for instance, relative to previous studies which manipulate receptiveness through subtle changes such as through pronoun use (SI Appendix), confederates in our study quite explicitly signaled openness to participants’ opposing views.^§

The second factor of the design was whether confederates made a persuasive appeal. When they did so, they argued why they supported the policy by sharing a detailed, persuasive narrative about an undocumented immigrant (2, 10, 30). For an example of a narrative used, see SI Appendix.

Immediately following the conversation, as well as five weeks later, we measured participant’s policy attitudes and prejudice toward unauthorized immigrants ($N=$ 1,485 and 1,406, respectively). These constitute our primary persuasion outcome measures. As described in further detail in SI Appendix, we successfully reinterviewed over 92% of participants who began a conversation in the long-term follow-up survey. The immediate postconversation survey also measured several manipulation checks and potential mechanisms.

This design enabled us to test whether adding listening to a persuasive appeal enhanced immediate and long-term attitude change among participants. As stated in our preregistration, our primary analysis (hypothesis) concerns the difference between the Persuasive Appeal Only and the Listening $+$ Persuasion condition on our primary outcome measures of policy support and prejudice (detailed below). The Listening Only condition also allowed us to test whether listening itself had any direct persuasive effects, which could have represented an alternative explanation for why adding listening to a persuasive conversation might increase its persuasive effects.

SI Appendix provides a detailed review of the noteworthy features of our design, participant experience, including advertising creatives and the scripts the confederates followed in each condition. This experiment was preregistered, and we follow our preregistered analysis plan unless explicitly noted.^¶

Measures.

To ensure that confederates successfully communicated their perspective, nonjudgmentally listened, and shared stories according to the random assignment, we administered several manipulation checks. Participants rated the extent to which they believed that their conversation partner supported the immigration policy, the extent to which their conversation partner shared a story, and the extent to which their partner listened.

Our main research question was whether listening enhances persuasion. Since confederates were trying to change participant attitudes toward both the policy and toward undocumented immigrants more generally, we operationalized persuasion as a reduction in “exclusionary attitudes”—prejudice toward undocumented immigrants and support of anti-undocumented immigrant policies (2, 10, 30, 33). We assessed these outcomes at baseline, immediately after the conversation, and in the long-term follow-up survey.

To assess anti-undocumented immigrant prejudice, participants rated how warmly they felt toward “illegal immigrants” (we used the term “illegal” to be in the vernacular of the target participant), as well as their agreement with four items describing attitudes toward undocumented immigrants. To assess anti-undocumented immigrant policy preferences, participants rated their agreement with six policy questions. Both measures were scored such that higher scores reflected greater support of proimmigration policy or more tolerance toward undocumented immigrants.

We also administered several measures assessing three different possible mechanistic explanations of why listening might or might not enhance persuasion. These include self-reflection (elaboration), interpersonal favorability, and learning.

See SI Appendix for details and question wording. We also report the secondary and exploratory measures and additional baseline measures in SI Appendix.

For all measures involving more than one item, we created a standardized index. To construct our indices, we first standardized each item to have a mean of zero and a SD of 1. Then, we averaged all items that the participant answered. Finally, we standardized the index to have a mean of 0 and a SD of 1. Our estimates are thus interpretable as SD changes (i.e., $d$s).

SI Appendix also discusses our sample size selection, random assignment procedure, exclusions, and analytical approach.

Results

Manipulation Checks.

We first test whether the confederates’ behavior had the intended effects. We find that it did: Participants recognized that their partner supported the in-state policy position, they felt listened to when the confederate was instructed to listen, and they felt that they received a story when the confederate was instructed to make a persuasive appeal. SI Appendix, Table S8 summarizes these results.

First, participants in the Listening $+$ Persuasion and Listening Only conditions felt more listened to compared to those in the Placebo condition; those in the Persuasive Appeal Only condition did not differ in how listened to they felt compared to those in the Placebo condition. This suggests that confederates successfully deployed nonjudgmental listening in the appropriate conditions (i.e., in the Listening $+$ Persuasion and Listening Only conditions).

Participants in the Listening $+$ Persuasion condition and the Persuasive Appeal Only condition also more strongly agreed that they received a story compared to those in the Placebo condition; those in the Listening Only and the Placebo conditions did not differ in their perceptions of being told a story.

In addition, participant perceptions of their partner’s policy support did not differ between the Placebo condition and either the Listening $+$ Persuasion condition or the Persuasive Appeal Only condition. Participants in the Listening Only condition perceived the confederates as less supportive of granting access to in-state college tuition, although this was driven largely by differences between 4 and 5 on the 1 to 5 scale (intensity of perceived support).

Objective statistics from our video recordings of the conversations themselves reinforce our interpretation that the experimental conditions manipulated what we intended.^# First, as shown in SI Appendix, Table S9, confederates used language that communicated receptiveness in the Listening Only and Listening $+$ Persuasion conditions. To measure this, we used the politeness package (34) to code transcripts of the conversations. According to this analysis, confederates used more receptive language in the Listening $+$ Persuasion and Listening Only conditions (means of 1.4 and 1.3, respectively) than they did in the Persuasion Only or Placebo conditions (means of 0.7 and 0.5, respectively).^‖ We also find that the participant said a similar number of words in the Placebo and Persuasive Appeal Only conditions, consistent with them not providing their point of view in either condition. Participants also said a similar number of words in both conditions where confederates were instructed to listen (Listening Only and Listening $+$ Persuasion), and confederates were similarly receptive in both conditions.

Overall, then, we conclude that our random assignment successfully manipulated the constructs that we intended.

Primary Outcomes: Exclusionary Attitudes.

We next test for effects on exclusionary attitudes, our primary outcome measures. Fig. 1 and Table 2 summarize these results.

Fig. 1.

Means of primary outcome measures by condition and time. Notes: Covariate-adjusted means are surrounded by 95% CIs. Higher scores reflect greater support of proimmigration policy or more tolerance toward undocumented immigrants.

Table 2.

Statistics for exclusionary attitude outcomes (primary outcomes)

	Descriptive statistics M (SD)				Inferential statistics β (SE)
Measure	Placebo	Listening	Persuasive appeal only	Listening + Persuasion	Listening vs. Placebo	Persuasive appeal vs. Placebo	Listening + Persuasion vs. Placebo	Listening + Persuasion vs. Persuasion	Persuasion × Listening interaction
Immediately post-conversation
Policy	−0.19	−0.10	0.08	0.15	0.09+	0.26***	0.33***	0.07	−0.02
	(0.60)	(0.60)	(0.62)	(0.65)	(0.05)	(0.05)	(0.05)	(0.04)	(0.07)
Prejudice	−0.17	−0.10	0.07	0.13	0.08+	0.25***	0.30***	0.06	−0.02
	(0.53)	(0.56)	(0.58)	(0.54)	(0.05)	(0.04)	(0.04)	(0.04)	(0.06)
N	283	267	466	469
Long-term follow-up
Policy	−0.11	−0.10	0.10	0.13	0.01	0.21***	0.23***	0.02	0.01
	(0.65)	(0.66)	(0.64)	(0.61)	(0.06)	(0.05)	(0.05)	(0.04)	(0.07)
Prejudice	−0.07	−0.07	0.11	0.11	−0.01	0.17***	0.18***	0.01	0.02
	(0.62)	(0.60)	(0.58)	(0.61)	(0.05)	(0.05)	(0.05)	(0.04)	(0.07)
N	264	258	442	441

Notes: $+$$=$ adjusted P-value $<$ .10, * $=$ adjusted P-value $<$ 0.05, ** $=$ adjusted P-value $<$ 0.01, and *** $=$ adjusted P-value $<$ 0.001. Means and effect estimates are shown on outcomes rescaled to SD one and mean zero across the entire sample. Coefficients can therefore be interpreted as SD differences. Inferential statistics for the simple effects show the estimated effect of the first condition listed relative to the second; i.e. the second condition listed is the reference condition for comparisons shown. Means are covariate adjusted. Based on the number of participants completing each measure, sample size may differ slightly across measures from the N given in the last row. More positive numbers indicate more pro-undocumented immigrant attitudes and affect.

First, both the Listening $+$ Persuasion condition and the Persuasive Appeal Only condition had large and statistically significant immediate effects relative to the Placebo condition on exclusionary attitudes, between 0.25 and 0.33 SDs. These effects largely persisted in the long-term follow-up, with estimates of approximately 0.20 SDs. This effect size is comparable in size to past research on persuasive canvassing (2, 10, 30). These effects are also substantively meaningful: Immediately after the conversation, whereas only 33.6% of people supported the policy in the Placebo condition, 34.5% of people supported it in the Listening Only condition, 45.9% supported it in the Persuasive Appeal Only condition, and 48.8% supported it in the Listening $+$ Persuasion condition, for average differences over 10 percentage points; five weeks later, these differences were 5 to 10 percentage points.

To test our primary research question of whether listening enhances persuasion, we next focus on the comparison between the Listening $+$ Persuasion and the Persuasive Appeal Only conditions. We preregistered this as our primary comparison of interest. If listening enhances persuasion, then we should expect to see in the Listening $+$ Persuasion condition more tolerance and policy support compared with the Persuasive Appeal Only condition, both immediately after the conversation and in the long-term follow-up. Yet when comparing these conditions, we find a very small and statistically insignificant difference in the immediate postconversation survey ($d$’s $=0.06$ and $.07$; $P$’s $>0.10$). This effect decayed to zero 5 wk later ($d$’s $=-0.02$; $P$’s $>0.617$).^**

When allowing for the possibility that listening may have some direct effect, our results suggest even greater pessimism about whether listening enhances persuasion. As noted above, we included a Listening Only condition in case listening led to attitude change independent of persuasive appeals. Consistent with this possibility, although not our main hypothesis of interest, we found that the Listening Only condition also may have had a small (approx. 0.09 SDs) direct effect on attitude change ($P_{\text{adj}}=0.062$) compared to the Placebo condition, consistent with refs. 14 and 25. However, these effects appear to have decayed in the long term.^†† To take account of any potential direct effect of listening, we therefore preregistered that we would also estimate a regression with an interaction term between receiving listening and receiving a persuasive appeal as well as prespecified covariates. These interaction terms were null: There was no evidence for an interaction of persuasion and listening for either policy attitudes ($P_{\text{adj}}=0.566$) or prejudice attitudes ($P_{\text{adj}}=0.566$) immediately. There was also no evidence for an interaction in our long-term follow-up.

The Top half of Fig. 2 summarizes our findings: Although the persuasive appeal effectively changed prejudice and policy attitudes (Left panel), adding listening to the persuasive appeal did not change attitudes any further (Right panel).

Fig. 2.

Estimated Effects of Persuasive Appeal vs. Placebo conditions, and Listening $+$ Persuasive Appeal vs. Persuasive Appeal Only conditions. Notes: Each point shows the estimated effect of either the Persuasive Appeal Only condition relative to the Placebo condition (Left panel) or the Listening $+$ Persuasion condition relative to the Persuasive Appeal Only condition (Right panel, our main preregistered comparison of interest), on the outcomes listed at Left. The points are point estimates from the same linear regressions reported in Tables 2 and 3, surrounded by SEs (thick lines) and 95% CIs (thin lines).

SI Appendix, Tables S21 and S22 show that the results are similar when we reweight the sample to match the population that started our initial baseline survey, including those who were uninterested in participating in a conversation.

We hasten to note that, although our SEs are small relative to estimates of the effects of listening from previous studies, given statistical uncertainty we cannot rule out that listening could slightly enhance persuasion. Moreover, in a post hoc power calculation, we found that we were well powered (80% power) to detect a difference between the Listening $+$ Persuasion and Persuasive Appeal Only conditions equivalent to listening increasing persuasiveness by approximately 40% on a combined index of policy and prejudice items (i.e., an increase of $d=0.102$ over the effect of the Persuasive Appeal Only condition, which was $d=0.26$).

In summary, then, we do not find evidence that adding listening to a persuasive intervention meaningfully enhances persuasion.

Notably, although this was not our primary research question, we also did not find much evidence that listening alone changed attitudes: There were only marginal differences between the Listening Only and Placebo conditions in the immediate follow-up survey (policy attitudes: $d=0.09,P_{\text{adj}}=0.062$; prejudice: $d=0.08,P_{\text{adj}}=0.062$), and these decayed to zero in the long term. This contrasts with recent work finding that listening alone can change attitudes (14, 25).

Mechanisms.

Why did listening fail to enhance persuasion? Here, we test three mechanisms by which listening might enhance persuasion: reduced defensiveness, interpersonal favorability, and self-reflection (elaboration). We also test a mechanism by which the persuasive appeal might enhance persuasion: learning. Rather than present mediation tests, which would not reliably show causality in our setting (35)^‡‡, we estimate the effects on each of the posited mediators. Table 3 gives the full results. The Bottom half of Fig. 2 also visually summarizes our key results.

Table 3.

Statistics for mechanism measures

	Descriptive statistics M (SD)				Inferential statistics β (SE)
Measure	Placebo	Listening	Persuasive appeal only	Listening + Persuasion	Listening vs. Placebo	Persuasive appeal vs. Placebo	Listening + Persuasion vs. Placebo	Listening + Persuasion vs. Persuasive appeal only
Immediately postconversation
Reflected on attitudes	−0.34	0.20	0.08	0.15	0.54***	0.42***	0.49***	0.07
	(0.89)	(0.91)	(0.94)	(0.88)	(0.08)	(0.07)	(0.07)	(0.06)
Liked partner	−0.28	0.08	0.06	0.26	0.37***	0.36***	0.55***	0.19**
	(0.83)	(1.01)	(0.83)	(0.87)	(0.08)	(0.07)	(0.07)	(0.06)
Felt mutual respect with partner	−0.30	0.04	0.01	0.30	0.35***	0.32***	0.60***	0.28***
	(0.98)	(0.87)	(0.91)	(0.87)	(0.08)	(0.07)	(0.07)	(0.06)
Felt understood and valued by partner	−0.32	0.31	−0.34	0.45	0.64***	0.00	0.78***	0.79***
	(0.93)	(0.90)	(0.88)	(0.85)	(0.08)	(0.07)	(0.07)	(0.06)
Felt defensive during conversation	−0.04	−0.10	0.08	−0.15	−0.06	0.11	−0.12	−0.23***
	(0.92)	(0.97)	(0.87)	(0.95)	(0.08)	(0.07)	(0.07)	(0.06)
Have a lot in common with undoc. imms	−0.38	−0.20	0.13	0.19	0.18*	0.51***	0.57***	0.06
	(0.85)	(0.84)	(0.87)	(0.85)	(0.07)	(0.07)	(0.07)	(0.06)
Amount learned in conversation	−0.91	−0.79	0.48	0.53	0.12+	1.39***	1.43***	0.04
	(0.69)	(0.73)	(0.73)	(0.72)	(0.06)	(0.06)	(0.06)	(0.05)
N	282	267	465	469

Notes: $+$$=$ adjusted P-value $<$ 0.10, * $=$ adjusted P-value $<$ 0.05, ** $=$ adjusted P-value $<$ 0.01, and *** $=$ adjusted P-value $<$ 0.001. Means and effect estimates are shown on outcomes rescaled to SD one and mean zero across the entire sample. Coefficients can therefore be interpreted as SD differences. Inferential statistics show the estimated effect of the first condition listed relative to the second; i.e. the second condition listed is the reference condition for comparisons shown. Means are covariate adjusted. Based on the number of participants completing each measure, sample size may differ slightly across measures from the N given in the last row.

Table 3’s first row shows the results for the elaboration outcome (i.e., “reflected on attitudes”). Although the Listening Only condition increased elaboration, consistent with recent work (14), we found that adding listening on top of a persuasive appeal did not increase elaboration. This is perhaps because, in the context of an interpersonal conversation, people may have chosen to closely engage anyway or because narratives themselves are highly engaging (11).

The next four rows of Table 3 examine perceptions of the source, such as whether people like their partner, as well as feelings of defensiveness. We consistently find that the Listening $+$ Persuasion condition led to improved interpersonal perceptions (i.e., increased liking, trust, feeling understood) and reduced defensiveness compared to the Persuasive Appeal Only condition. Yet recall that earlier we found that the Listening $+$ Persuasion condition did not lead to additional attitude change above and beyond the Persuasive Appeal Only condition. Together, these results suggest that improved interpersonal perceptions and reduced defensiveness^§§ do not appear to enhance persuasion, in line with our skeptical line of reasoning earlier and other recent studies finding that people change their attitudes in response to persuasive appeals even from disliked sources (e.g., ref. 6).^¶¶

The last row reinforces this interpretation. While extant theories suggest that adding listening to a persuasive conversation might increase openness to learning new information during it, we found that participants in the Listening $+$ Persuasion condition were no more likely to say they had learned from the conversation than participants in the Persuasive Appeal Only condition.

In summary, although our findings with respect to mechanisms are not definitive, they do suggest amendments to extant theoretical models of high-quality listening and receptiveness. Contrary to the literature’s expectations, the Listening $+$ Persuasion treatment did not increase elaboration relative to the Persuasive Appeal Only condition, perhaps because conversations were already engaging to begin with (because they were face-to-face over video instead of using text only). Furthermore, although we confirmed the literature’s expectations that listening would increase liking, we did not find that this enhanced persuasion. Finally, consistent with our interpretation that people appear willing to learn from sources they dislike, we found that participants said they learned a similar amount from the conversations in the Persuasive Appeal Only and Listening $+$ Persuasion conditions.

Confederate Expectations.

At the conclusion of the study, we surveyed the confederates as well as the research assistants (RA’s) who coded the videos, for a total of 17 responses. The confederates and RA’s rated their expectations about the persuasiveness of each condition. As demonstrated in SI Appendix, Fig. S6, confederates and research assistants expected the Listening $+$ Persuasion condition to be more persuasive than the Persuasive Appeal Only condition. In fact, all 17 people chose the Listening $+$ Persuasion condition as being most likely to be persuasive. This suggests that even those who partook in or observed these conversations thought that the way listening was implemented in our study should have enhanced persuasion.

Discussion

Does high-quality, nonjudgmental listening enhance persuasion? We shed light on this question using a longitudinal field experiment involving nearly 1,500 one-on-one conversations. Participants opposed to a policy providing government benefits to undocumented immigrants were recruited from across the United States, and they had conversations with another “participant” who was a supporter of the policy. In reality, this person was a trained confederate, who either actively listened to the participant discuss their views or not, and either made a persuasive appeal by sharing a narrative about an undocumented immigrant or not. Relative to those in the placebo condition (in which confederates neither nonjudgmentally listened nor made a persuasive appeal but in which both the confederate and participant each stated their position), participants who received just a persuasive appeal had large ($\sim$0.25 SDs) reductions in exclusionary attitudes—prejudice toward undocumented immigrants and opposition to policies benefiting undocumented immigrants. These attitude changes persisted over a month later. Importantly, participants who received both listening and a persuasive appeal showed nearly identical attitude change. The small and insignificant difference between these conditions decayed to zero in a long-term follow-up, even though both remained persuasive relative to placebo. Although listening improved interpersonal attitudes (e.g., liking of the confederate), listening did not increase reflection or learning, which may explain its failure to enhance attitude change. This research suggests that, at least in conversations about policy disagreements, listening may not reliably enhance persuasion efforts. On a practical level, this suggests that adding listening to persuasive appeals may not be worth the added costs if persuasion is the goal.

This finding suggests several theoretical implications. First, our study provides one of the first empirical tests of whether behaviors that signal receptiveness enhance persuasion in live face-to-face conversations. Much literature has speculated that behaviors like listening that communicate receptiveness—openness to others’ ideas—should enhance persuasion (3, 16). However, little work has tested this in actual conversations (see SI Appendix, Table S5). In our work, we used high-quality nonjudgmental listening (2, 14) as one example of a behavior that would increase receptiveness. We find that, contrary to expectations, listening did not enhance long-term persuasion. This result was surprising to even our confederates, all of whom predicted that the Listening $+$ Persuasion condition would be more persuasive than the Persuasive Appeal Only condition (SI Appendix, Fig. S6). Our findings thus resonate with the recent trend in research on receptiveness to move “beyond persuasion” in investigating the effects of receptiveness (27)—and indeed, in our study, we show that listening increases liking, trust, and respect, and reduces defensiveness. These findings complement recent research suggesting that political disagreement with strangers is often more positive than expected (37).

Second, our work adds to a long tradition of scholarship on attitude change (19). Past theorizing suggests that reducing defensiveness, increasing reflection, improving source perceptions, or just communicating novel information should enhance attitude change (3, 14, 38). Although we find that listening improved liking and reduced defensiveness above and beyond a persuasive appeal, we do not find that it increased either reflection or the amount of information learned—though receiving just a persuasive appeal increased both. This suggests that attitude change may have occurred in conditions containing a persuasive appeal because the narratives communicated novel information to participants. This supports other findings showing that attitudes are changed from exposure to ideas even from disliked sources or when those ideas are contrary to prior attitudes (e.g., refs. 6–9). Our findings contribute to that literature by manipulating affect toward the source of a message and finding persuasion is not enhanced, thus raising doubts about the role of source credibility in willingness to accept persuasive political messages (e.g., ref. 12). Of course, there may be limits to how disliked a source can be before people become less willing to learn from it (e.g., ref. 9).

Third, this study clarifies the mechanisms likely responsible for the well-replicated success of door-to-door conversational interventions involving the nonjudgmental exchange of narratives (2, 10, 30, 39). Past work on this topic speculated that canvasser listening behaviors contributed to the effects it found, but has not separately manipulated these behaviors. By testing the separate influence of persuasive narratives and listening behaviors and showing that narratives alone can have durable persuasive effects, this study suggests that narratives, not listening behaviors, are likely responsible for these findings (see also ref. 11). Relatedly, our findings corroborate past research showing that narratives improve interpersonal perceptions (21, 40).

This study has several important limitations. First, we only administered one type of persuasive appeal—using detailed narratives to change attitudes. We used this method because previous research finds it more persuasive than other methods, such as presenting facts or reasoning (e.g., refs. 11 and 41), and it is commonly used by practitioners. However, even if listening does not enhance the persuasiveness of narratives, it is possible it may enhance the persuasiveness of facts or other persuasive methods. Second, we recruited participants interested in having a video conversation with a stranger. Although participants did not know the topic of conversation ahead of time, our findings may not generalize to people who are not interested in talking to strangers. Third, we studied only online video conversations, and so it is not clear whether this finding would generalize to other conversational media, such as text-based conversations (42) or in-person conversations, where listening could be easier to signal. Fourth, although we expected listening to be most helpful in the context of conversations about contentious topics, future research should probe whether our findings would generalize to disagreements about other topics that are less contentious or political—or where persuasion may be easier or harder—than unauthorized immigration. Fifth, we only recruited participants who were opposed to this policy; future work could explore the reverse. Finally, we did not gather objective measures of learning; given the results on this potential mediator, future research should do so.

Our findings also suggest several directions for future research. First, future work would benefit from studying more persuasive forms of listening. In this study, we tested whether listening would change a counterpart’s evaluation of the same persuasive appeal delivered without listening. However, a persuader could listen to understand a counterpart’s needs or concerns, and then in turn tailor the persuasive message accordingly (39). We explicitly asked our confederates to avoid engaging in listening of this type in order to not confound any effects of listening with changes in the persuasive appeal made, but future research should investigate these questions. Second, future work should study more naturalistic conversations. Our study involved scripted one-off conversations with a stranger—who was in actuality a professional canvasser acting as confederate. Third, future work should also study people’s choices to opt into or out of conversations. In Zaller et al. (12) framework, listening may increase the chances people Receive conversations with persuasive messages—either because people stay in the conversation longer, or are willing to have it in the first place. In the context of a political campaign, for instance, listening could thus be worth its “cost” (in lengthier conversations) if it leads voters to be more willing to stay in persuasive conversations, even if, per our results, listening does not increase their odds of Accepting (i.e., being persuaded by) the messages they hear.^## Relatedly, failing to listen well may also be uncomfortable for the speaker, reducing their interest in having such conversations. Finally, the limited direct effects of listening we found (when comparing the Placebo and Listening Only conditions) also merit further research.

Taken together, our work suggests that high-quality, nonjudgmental listening may not enhance persuasion to the extent many hope. That said, our research confirms that high-quality listening can make others feel respected, less defensive, and increasing their liking of their conversational counterpart.

Supplementary Material

Appendix 01 (PDF)

Data, Materials, and Software Availability

Anonymized survey data have been deposited in OSF (29).