Tell don’t ask: how to use social media to mobilise local collective climate action

Public awareness, anxiety, and concern about climate change is growing^1,2,3. In the United Kingdom (UK), climate change is consistently ranked as one of the greatest concerns, alongside the cost of living, healthcare, and economy^4,5. Personal experiences of the adverse local impacts of climate change are also on the rise, such as the increased frequency and severity of extreme weather events like flooding, heat waves, and droughts⁶. To address climate change, therefore, the majority of people (64%) report trying to make individual-level lifestyle changes⁵. But many are anxious that these changes are ineffective or too costly^4,5. Others feel big polluters and governments need to act^3,4,5. Given concern, anxiety, and anger at the perceived inaction amongst businesses and government, people are increasingly turning to collective action through joining climate movements^{3,7,8,9,10,11}.

Collective climate action–performed as a part of a group with the common goal of mitigating climate change¹²–is an important avenue for the public to channel their motivation and concern, and articulate demands for government and business. The importance of collective action to address climate change (as opposed to merely individual behaviour change) has long been recognised^13,14,15. Even if individuals initially change their lifestyles, they may feel anxious and demotivated if they feel their actions have no discernible impact². If others are not perceived to do their fair share, individuals making the effort may feel “suckered”¹⁶ and stop acting, or worse, license themselves to harm the environment¹⁷. Conversely, emerging evidence finds that those undertaking collective climate actions—such as striking—report making more lifestyle changes compared to those who don’t¹¹. Rather than acting alone, acting alongside others can maintain motivation, increase collective efficacy and even help address climate anxiety³.

Furthermore, the Intergovernmental Panel on Climate Change (IPCC) has noted that large-scale transformations in governance and economic systems are required through changing policy and regulation, infrastructure, production, and energy systems (i.e., supply-side changes), apart from lifestyles (i.e., demand-side changes)¹⁸. The IPCC also notes (with ‘high confidence’) that collective action connected to social movements plays a substantial role in pressuring governments to create new laws and policy^9,18. For instance, Extinction Rebellion (XR), the largest grassroots environmental movement in the UK, pressured the government to declare climate change an emergency in 2021 through non-violent civil disobedience¹⁹.

However, there are many psychological barriers to collective climate action. Emerging work based in the UK and with movements like XR shows that perceptions of lower personal and collective efficacy, social norms against protest and pro-environmental behaviour, and lower social identification are important barriers^{10,20,21,22,23,24,25,26}. Lower social identification with climate movements and their activists are negatively associated with collective climate action intentions^22,26,27. This challenge may be rooted in the idea that taking climate actions are not simply behavioural responses to a stimulus but can instead be conceptualised as a facet of an individual’s changing critical engagement with climate change, a result of sociopolitical processes. From this ‘critical consciousness’ perspective, understanding the systems and structures that create and perpetuate climate inequities is a key motivator for action.

To illustrate how these perspectives connect, the language used by XR activists (e.g., “opinionated”, “argumentative”), social image and stereotypes (white middle class) and actions (disruptive public protest) can influence social identification^22,27,28. When a movement’s messaging and imagery appear disconnected from the social justice dimensions or even lived experiences of climate change, it may alienate individuals, particularly those from marginalized groups whose engagement may be contingent on this critical awareness²⁹. More broadly, Morris (2025) also discusses how apocalyptic language and narrative may contribute to marginalised groups feeling excluded from environmental movements and causes³⁰.

Effective communication strategies are crucial for overcoming these barriers and galvanizing public engagement. The field of climate change communication has significantly evolved, moving beyond simply conveying scientific facts to understanding how different approaches influence attitudes and behaviours^31,32. For example, Maibach et al. (2023) highlight the importance of simple, clear messages, repeated often, by trusted messengers, and framing actions as easy, fun, and popular to encourage uptake³³. Similarly, Moser (2016) notes the importance of using climate imagery and appealing to the affective and emotional side of climate change (apart from just appealing to cognitions and facts)³¹. There are also calls for a “silver buckshot” approach, recognizing that no single communication method works for all audiences; instead, diverse strategies are required to resonate with varying perspectives and contexts³⁴. Despite these advances, research on the causal impact of climate communications on mass mobilisation for climate movements has been relatively neglected.

In this regard, social movement scholars offer valuable insights into how collective action can be mobilized. For instance, Benford and Snow (2000) identified three key types of frames typically used by social movements: diagnostic, prognostic, and motivational³⁵. While diagnostic frames identify the problem and attribute blame, and prognostic frames propose solutions, motivational frames are particularly critical for inspiring participation. These frames provide a rationale for engaging in collective action by emphasizing the urgency, severity, and efficacy of potential contributions, along with a sense of moral obligation or propriety. There is, however, very little causal evidence on about the impact of these strategies from ecologically valid field settings, including on social media, where climate and social movements are increasingly turning to mobilise the public.

How social media communications—the content of messages and posts—increase collective action has been studied in other contexts, including political mobilisation (e.g., 2010 US elections³⁶, Brexit and the 2016 USA elections³⁷), voter turnout³⁸, and civil society uprisings (e.g., 2020 Hong Kong protests³⁹). For example, Bond et al.³⁶ found that exposure to a social message (with pictures of friends who voted) increased the likelihood of choosing the “I Voted button” on Facebook compared to an informational message (without pictures of friends) in a randomised controlled trial (RCT)^36,38. While emerging studies also find social media messages and images can increase online engagement with environmental organisations^40,41, how they can increase engagement with information about local collective climate action opportunities has not been studied. Therefore, which type of messages and images increase interest in local collective climate action is unclear.

Indeed, effective social media communication strategies must balance multiple imperatives: highlighting both individual and collective agency whilst conveying urgency without fostering paralysis⁴². Vasi and Macy (2003) address this mobilizer’s dilemma, suggesting that effective communication strategies must combine messages that simultaneously address both the “free rider” problem (the temptation to benefit from collective action without contributing) and the “efficacy” problem (the perception that individual actions are too small to matter)⁴². By weaving a narrative that highlights both the severity of the threat and the tangible impact of collective effort, mobilizers can try to provide a compelling rationale for engagement.

Prior research suggests that how you ask affects whether people comply. Asking nicely through polite requests can increase compliance (especially when people are in a good mood)⁴³. Our scoping research with XR (using qualitative focus groups) also revealed that non-activists who were concerned about climate change preferred communications that were framed as polite requests rather than exhortations to act. However, other studies show that compliance depends on contextual aspects of the ask, including the behaviour in question (for example, whether declining itself threatens social image or negative face) and how contingent or serious it is, apart from the social distance and identities of who is asking and being asked^44,45. In contingent requests, asking nicely may even backfire by downgrading perceived contingency⁴⁶. Furthermore, the efficacy of requests in the context of climate change social media communications is uncertain, as previous studies have shown that people pay more attention to urgent and “angry” language⁴⁷, which can also mobilise people^48,49. In this study, we define exhortations as urgent, direct calls to action that emphasise the collective responsibility to act (e.g., “Don’t stand by idly”), whilst requests are framed as polite invitations that give the recipient more choice in responding (e.g., “Would you like to come along?”). It is unclear whether exhortations or asking nicely interact with the imagery used.

The choice of visual imagery in climate communications is equally important. Images depicting climate impacts (such as flooding), protest actions, or diverse participants may each resonate differently with audiences and interact with message framing in complex ways²⁹. Prior research suggests that vivid imagery of climate impacts can increase risk perception and concern⁴⁸ and increase personal behavioural change⁵⁰. Studies show images of protesters may influence concern and social identification with movements^22,27,28. Diversity in visual representation may affect perceptions of inclusivity and relevance to different demographic groups, but this aspect has not been examined²⁷. How these different types of imagery interact with message framing to influence engagement with collective climate action opportunities has not been systematically tested.

To test which types of social media messages can increase engagement with information about collective climate action opportunities, this study used a randomised controlled trial with Facebook users (n = 353,998) in the UK. We collaborated with Extinction Rebellion (XR)-UK for several reasons. First, XR is the UK’s largest grassroots climate movement, making it highly relevant for studying collective climate mobilisation. Second, XR actively uses social media to recruit participants for local events, providing an ecologically valid setting for testing communication strategies. Third, XR’s organizational structure relies heavily on local groups coordinating actions across the country, making local engagement crucial to their model. Existing research also suggests that XR’s apocalyptic framing and confrontational tactics can be alienating for some groups, particularly Black, Asian, and Minority Ethnic (BAME) communities and working-class individuals who may perceive the movement as predominantly white and middle-class^27,28. These organisational practices and characteristics influenced both the messages tested (as discussed below and in the methodology section). It may also influence who engages with XR’s communications on social media, and we return to these important issues in our discussion.

The trial was conducted amongst users in three cities—Birmingham (18th–24th January 2023), Cardiff (2nd–9th February 2023), and Oxford (24th February–1st March 2023). It aimed to increase awareness of, and engagement with, local community talks organised by XR, and to invite new members to engage with their local group. So, in this RCT, Facebook users were randomly assigned to an ad from XR (they were not randomly sampled from the population). The ad either exhorted or requested them to attend a local talk via a post in their ‘News Feed’ and provided a clickable link to do so (Fig. 1). Apart from varying how people were asked (i.e., exhortation versus request), images were also randomly varied to depict XR street protests (protest; the default image used in current XR social media communications), people from different backgrounds (diversity), or flooding (climate change impact; Fig. 2).

This experiment therefore, evaluated the impact of the type of message (exhortation or request; Fig. 1), image (protest, impact, or diversity; Fig. 2), and any interaction effects between the two. The design of this experiment allowed us to assess the impact of these treatments on a key information-seeking behavioural measure: clicking a button with a link to see more information about the talk. We note that link clicks represent engagement with information about local collective climate action opportunities, which can be seen as an important first step in the pathway from online exposure to real-world participation. We targeted users in and around 30 km from each location, aged over 18, and from all genders, to randomly receive one of the ads for one week before the talks were scheduled. While we were able to get disaggregated data by user age, gender and location, we could not obtain data by ethnicity, race, disability or socio-economic status.

The ad campaign reached 353,998 users. Of the total users reached, 39.5% were in Birmingham, followed by Cardiff (32.4%) and Oxford (28.1%). They were mostly aged above 55 (66.9%). In terms of gender, most users self-identified as male (52.7%) and female (46.2%). Around 1.1% of the users identified as either non-binary or preferred not to say (Table 1).

Around 6002 users—1.7% of total users—clicked the link. The greatest number of link clicks came from the ads that combined exhortation messaging with impact imagery (Fig. 3). When exhortation messages were combined with any other image (either protest or diversity), there seems to be little difference with request messages.

Since we are interested in understanding local collective action, we also disaggregated the results by location (Fig. 4). Birmingham users had the highest proportion of link clicks (2.01% of users exposed to any ad), followed by Cardiff (1.55%) and Oxford (1.43%). In Birmingham and Oxford, we find similar patterns to the pooled data in Fig. 1. In other words, users were likely to click the link when exposed to exhortation messages with impact imagery. However, a different pattern emerges in Cardiff, where more link clicks resulted from being exposed to exhortation messages with protest imagery, followed by exhortation messages with impact imagery. Some other differences also emerge; for instance, request messages are more effective when combined with diversity imagery in Birmingham and protest imagery in Oxford.

Direct effects of message and imagery

To examine whether these differences across treatment groups and locations are statistically significant, we report results from linear probability regression (LPM) models by location and for the pooled sample (with heteroskedasticity robust standard errors; without gender and age covariates in Model 1, and with covariates in Model 2). Figure 5 presents the regression coefficients estimating the direct effects of exhortation and request language, and imagery, in messages.

In all locations, and for all observations, requests were less likely to get link clicks compared to exhortation messages. In the pooled sample, for example, users were 0.54 times less likely to click on a link when exposed to a request compared to exhortation when controlling for the influence of different imagery and covariates like location, age, and gender (Fig. 5 All, Model 2 Coefficient (β) = −0.0054, Standard Error (SE) = 0.001, p < 0.01, 95% CI [−0.006 to −0.004]). Impact images were more likely to elicit link clicks than protest images in Birmingham (when controlling for request messages), Oxford and for the pooled sample. In the pooled sample, for example, users were 0.22 times more likely to click on a link when exposed to an impact versus protest image when controlling for message language and covariates (Fig. 5 All, Model 2 β = 0.0043, SE = 0.001, p < 0.01, 95% CI [0.0033 to 0.0054]). Wald tests show that impact imagery was more effective than diversity images (F(1,353985) = 82.40, p < 0.01). But there are differences by location. In Cardiff, the difference between protest and flood imagery is not statistically significant (Fig. 5, Cardiff Model 2). Taken together, this suggests that exhortations are more effective at engaging people on social media to consider participation in local collective climate actions than requests. Flood imagery is more effective—or at least as effective—as protest imagery. Older users (especially those falling in age groups 45–54, 55–64, and 65+) were more likely to click than those aged 18-25 years.

Interaction effects of message and imagery

Figure 6 presents the predicted link clicks from the interaction between the message and image treatment groups (estimated via LPM models with heteroskedasticity robust standard errors and controlling for covariates). The models revealed that requests lead to significantly lower link clicks with impact imagery than using exhortations in all locations and the pooled sample (Fig. 6, All β = −0.013, SE = 0.001, p < 0.01, 95% CI [−0.016 to −0.011]). The difference is largest in Birmingham (β = −0.018, p < 0.01), followed by Oxford (β = −0.016, p < 0.01) and the smallest in Cardiff (and marginally statistically significant; β = −0.003, p = 0.09). We find that the only instance where requests are marginally more effective than exhortation is when combined with diversity images in Birmingham (β = .003, p = 0.096).

We aimed to study how grassroots climate movements like XR can increase online engagement with information about local collective climate action opportunities through social media communications. Specifically, we conducted an RCT with 353,998 Facebook users to evaluate the impact of messages that varied in how people are asked to get involved (exhortation or request), and the imagery used (protest, impact, or diversity), on clicking a link to find out details of a local climate talk (organised by XR), in three UK cities.

Our results showed that exhortation messages are more effective at increasing the likelihood of link clicks compared to requests, especially when combined with flood imagery (in the pooled sample, Birmingham, and Oxford). This finding implies that the persuasive impact of ‘asking nicely’ can depend on contextual features of the ask itself, such as the visual imagery. It aligns with past research suggesting that when there is some contingency, politeness can downgrade the request, and thereby backfire⁴⁴. Conversely, congruence between the ask and nature of the contingency can promote compliance, i.e., exhortations to “not stand by idly” because there is a risk of severe negative impacts like floods.

There is a consistent positive effect of exhortation messages despite the relatively subtle differences in wording with the request message. This result echoes past literature about how negative and moral emotions are more prevalent—and may elicit greater reactions—in social media discussions about climate change²⁸, as well as the role of risk perception^10,23 and anger⁴⁸ in driving collective action. More broadly, it raises the possibility that stronger exhortation messages could have an even larger effect on public engagement. Effects may also be stronger if combined with images of multiple interconnected adverse impacts (e.g., flooding and wildfires), as the range of affected social groups addressed– and the mobilisation capacity of the message– can increase³⁵. Surprisingly, it contradicts the focus group results, where people said they preferred requests to participate in collective climate action. The results of this study, therefore, suggest that what people say they want may not always match their real-world behavioural responses. This supports evidence that people’s preferences, attitudes and revealed behaviours do not always match³³, especially in situations imposing a higher personal cost on individuals⁵¹. It underscores the importance of experimentally testing communication strategies in real-world contexts.

The practical implication of this finding is that climate movements could use exhortations with climate change impact visuals, like flooding, to increase British public user engagement in information about local collective climate action opportunities. This recommendation contrasts with current practices; for instance, XR largely uses images of protesters to increase engagement (as per our internal conversations with XR). Protest imagery may still be an effective strategy to engage people who are already interested in, or willing to engage in, collective climate action. Future research can explore how those already inclined to act²⁸ compared to unengaged groups. Notably, these results show what is persuasive to get people to take the first steps towards becoming a climate activist – obtaining information to attend a local talk. Once people are in the room or have joined their local group, different types of messaging or imagery may be necessary to motivate them to take more significant action, like attending a protest.

Although the magnitude and direction of the effects were similar in all locations, we found that the interaction effects between exhortation and flood imagery were more pronounced in Birmingham and Oxford compared to Cardiff. It is unclear why this pattern emerges. Cardiff is prone to greater flood risk and damage compared to Birmingham and Oxford⁴⁴. However, research finds that simply living in a flood risk region does not necessarily impact behaviour⁵². Instead, personal experience of climatic events⁶, including flooding⁵³, may be associated with higher risk perception and action. The effect of personal experiences, furthermore, may be moderated by economic conditions⁶. Future research could study whether appealing to personal experiences of these events is more effective at raising collective action. It could also examine which types of impacts (e.g., droughts, wildfires, and heatwaves) are more effective at raising engagement apart from flooding, all events which are projected to increase in severity and frequency in the UK, but with different local risk factors. Another potential reason there are regional differences in response to XR messaging could be that the local XR groups are more or less visible within the communities. Exposure to actual climate actions (and activists) may predispose residents to be more or less receptive to future messaging.

We also found that older users (aged 45 years and above) were more likely to click than younger users (in the 18-25 age group). This ‘age effect’ could have emerged for a few reasons. Market reports suggest that users from older age groups are more likely to be active on Facebook and likely to click on ads^54,55. We find that older users are over-represented in our study, although in our experimental settings, we told Facebook not to ‘optimise’ ads by preferentially showing them to those who were more likely to click. This propensity of older users to click on ads may be concerning if they are not disposed towards taking collective climate action. For instance, they may be unwilling due to the biographical consequences of activism⁵⁶ or because they are less worried and concerned⁵⁷. But it is also the case that, contrary to popular belief and media coverage, many who are willing to participate in climate activism do belong to older groups⁵⁸. For example, recent mass protests such as ‘climate grannines’ and ‘rocking chair rebellion’ were led by and for senior citizens. A recent survey found that the average age of activists involved at 2023 action at the White House Correspondent’s Dinner was 52 (25% reported being 69 or older)⁵⁸. In March 2024, two protesters who attempted to smash the Magna Carta in the British Library were aged over 80 years old. Indeed, climate movements are keen to mobilise this population, because people can be retired, feel responsible and thereby more willing to take risks⁵⁹. Since older users are heterogeneous in their climate concern and willingness to act, it is possible that multiple factors are driving the age effect that we observe. Future work can focus on more precise evaluations of how different age groups respond to ads based on prior concern to better understand such dynamics. It can also examine other social demographics and identities, such as ethnicity, race or class, in response.

The diversity imagery we tested was intended to signal inclusivity, yet its relatively weaker performance (compared to impact imagery) raises questions about whether visual representation alone can overcome deeper perceptions about movement identity and accessibility. Since we did not have access to Facebook user data on ethnicity, race, or class, our study cannot shed light on how users in diverse groups responded to different treatments. Additionally, we could not examine engagement patterns among people with disabilities, for whom online activism may be particularly important, as it can provide more accessible pathways to participation compared to physical protests. Understanding how different communication strategies resonate with disabled activists and supporters represents an important direction for future research on inclusive movement building.

In this study, we collaborated with a real-world grassroots movement to assess and improve ongoing strategies for collective climate engagement in a British sample using a large-scale natural field experiment on Facebook. The study was ecologically valid, as it was part of an actual campaign run by XR. The random assignment to different ads mitigates the possibility that people’s prior perceptions of XR drove the responses to ads. However, it is possible that the overall level of engagement with the ads is specific to XR and may not replicate exactly for other advocacy groups and campaigns. Collective climate action can be performed in diverse ways locally, and with a range of organisations, each of which may be attractive to certain types of audiences based on their public-facing values, narratives, tactics or other characteristics. Future work can vary the type of pf organisation, apart from the language and imagery, to understand how generalisable the results are.

We also measured causal impacts on a revealed online information-seeking behaviour, thereby extending past researc,h which primarily considers self-reported collective climate action and intentions^{10,22,23,24,25,27,28,48}. However, we were unable to estimate how different ads influenced actual attendance at the talk. More broadly, this study only focuses on digital platforms as a contemporary avenue for such mobilization. There is an important discussion within social movement literature regarding “digital activism.” While critiques, such as those arguing for “slacktivism” or the limitations of low-threshold online engagement⁶⁰, have questioned the efficacy of online activities, a more nuanced understanding has recently emerged. Online activities can be viewed not as replacements for traditional activism but as amplification tools^61,62 that can extend reach and build pressure. In this context, online platforms, particularly social media, facilitate the rapid dissemination of information, enable the formation of virtual communities, and allow for low-cost participation, thereby potentially broadening the base of individuals exposed to and engaging with a movement’s message.

Indeed, a significant body of work has focused on the role of pre-existing social ties in driving activism⁶³. It emphasizes that participation often precedes ideological commitment, with individuals being drawn into movements through existing social networks, rather than a direct, staged progression from an initial recruitment message to deep involvement. Our promotion of ad-clicking as a “first step” is not meant to imply a linear, inevitable progression, but rather a potential entry point into a broader ecosystem of engagement. While direct, pre-existing social ties are undoubtedly powerful mobilizers, digital platforms can serve as a conduit for weak ties or even introduce individuals to an issue, providing the initial spark that, in conjunction with other factors, could lead to deeper involvement. For individuals without immediate access to existing activist networks, therefore, an online advertisement might be their first exposure to a local climate group. Clicking on an ad or expressing interest online can lead to receiving further communications, being invited to virtual or in-person events, or connecting with other interested individuals. These subsequent interactions, which may involve forming new social ties, can then facilitate greater commitment and more substantial forms of activism. Our study specifically examines willingness to attend climate talks, which represents a step beyond mere ad-clicking and signals a greater intent to engage. Therefore, while not a guarantee of deep involvement, such online engagement may act as a crucial initial bridge, especially for those not already embedded in activist social networks.

That said, future work should study whether online social media engagement has real-world behavioural consequences, and whether it can indeed act as a pathway to collective climate action. It is possible that users may click the ad and never take any real-world action due to barriers to following up, or more worryingly, because clicking is a substitute for other forms of activism. It is also possible that users who are more likely to click on the ad are doing so to confirm their pre-existing actions and beliefs (so the ad is reinforcing rather than generating activism). Current empirical evidence about this issue is limited and mixed. Some studies, for example, Bond et al.³⁶, find that online campaigns (covering around 61 million users) can lead to real-world impacts on single actions (e.g., voting). However, other experiments in the climate and environmental domains, e.g., Shreedhar et al.⁴⁰ and Kubo et al.⁴¹, find that real-world charitable donations towards conservation organisations can be limited, despite impacts on social media engagement. In related research, Castiglione et al.⁶⁴ find that a 12-week video intervention (via Zoom, rather than via social media) did not increase attendance at activism events. Given the importance of pre-existing social ties in mobilising collective action, future work could also examine whether ads targeted at those already expressing an interest in climate activism and environmental issues, or having social relationships with those who do, are more broadly more receptive. Apart from differences in the types of behaviours themselves (voting versus money donations versus volunteering time), there are also significant differences in the overall context (i.e., performing a political duty versus counter-normative collective climate action) and the study design (i.e., number of users targeted given organisational budgets).

In this regard, more research is needed to understand the power of social media campaigns to affect real-world behaviour across different collective action contexts. This would be useful to set realistic expectations for movements and organisations. This may be especially important given climate movements increasingly face structural barriers to local community-level expansion, such as institutional and legal frameworks that delay, prohibit, or criminalise climate action⁶⁵, including peaceful public protest and civil disobedience. For example, the 2022 UK Police Crime and Sentencing Bill has led to increased arrests of climate activists and harsher sentencing⁶⁶; these trends may be especially detrimental to the participation of marginalised groups, who may already have low identification with movements. Social media communications can be one of many instruments to raise local collective climate engagement, apart from face-to-face campaigning and community initiatives. (2615 words).

Scoping research

To undertake this study, we undertook preliminary research using qualitative techniques. First, based on extensive internal discussions within XR, we invited people who were sympathetic to climate issues but currently not involved in any form of environmental activism to take part in focus group discussions. We ran four 1.5-hour focus groups with eight participants in each. They were broadly diverse in terms of age, gender, ethnicity, and class. Participants were asked to discuss their reactions to eight example Facebook ads, which asked people to attend a local talk run by XR. The inspiration for these ads was Britain Talks Climate’s attitudinal segmentation of the British public⁶⁷. The succinct descriptions of seven different attitudes towards the climate crisis led to the creation of seven different messages, plus one ad that was typical of XR⁶⁸. In addition, two key findings that emerged, which were relevant to this study was that participants preferred requests to exhortations and were not very moved by standard protest imagery. Following this, and after receiving university ethics approval (#115808), two pilot Facebook tests were conducted in November-December 202,2 reaching 10,078 and 8627 users in Tyneside and Cardiff, respectively. During the pilot, diversity and alternative protest imagery (featuring children and families) was trialled because insights from the focus group discussions revealed this type of portrayal was preferred to standard protest imagery. However, since this imagery type did not perform well during the pilots, climate impact imagery was chosen to be tested in the final experiment instead.

Final experiment

For the final experiment—whose results are reported and discussed in this article—A/B experimental testing ad sets were used on Facebook to run for a week prior to each local talk. We used a 2×3 between-subjects design, varying the type of messaging and imagery used in a series of Facebook ads. Three images were used for each imagery type (to mitigate potential concerns that any effects were driven by one specific image). All messages and imagery were developed in consultation with XR. The test set-up was based on past research^40,69 aiming to evaluate the effectiveness of Facebook ads (SI A contains details of the experimental settings). Facebook offers only aggregated test result data on link clicks, which is grouped at the test group level and estimated by proprietary algorithms. So we created a simulated individual-level dataset, which forms the basis for the analyses conducted in this article.

The ads were released in three different cities (Cardiff, Birmingham, and Oxford), making a total of 54 ads (Fig. 2). On Facebook’s ad test platform, we specified that users in and around 30 km from each city, aged over 18, and from all genders, would be targeted to randomly receive one of the ads for one week before the talks were scheduled. These cities were chosen because they had an upcoming climate talk scheduled. The cities had the additional advantage of being relatively big and lying in different UK regions, so we could explore differences by location. In each ad, there was a button that users could click for more information about the talks themselves. This button took users to an Eventbrite page specific to the ad, so that we could collect data on both ad clicks and registrations. SI A.5 contains registration data by treatment, which was very low; only 12 observations. This is likely because registration was not necessary to attend the talks since XR wanted to reduce barriers to entry. Furthermore, all the information needed to attend the talks was available without registration by clicking the link. Therefore, we focus only on link clicks in the main paper. We note that our collaboration with XR required rapid deployment around scheduled events, which made pre-registration challenging and is a methodological limitation. Supplementary Information (SI) A contains further details of the experimental settings, dates, users reached in each location, and by age and gender groups and other study design details.

Table 1 and SI Tables S3 to S6 show sample characteristics by treatment, age, gender, and location. Since these characteristics were not evenly distributed across locations, we controlled for them by including them as regressors in all our linear probability regression models. In Fig. 3, the regression results are also displayed without these socio-demographic covariates, and results are qualitatively similar (also see SI Tables S7–S14). In addition, results are replicated using probit regression models and again results are qualitatively similar (SI Tables S15 to S22). Furthermore, they match the aggregated test results data on the most cost-effective ads, estimated by Facebook (SI A.4).