Feeding negativity on social media: how compensating customers creates vicious cycles of complaint Open Access

Imagine an airline facing negative comments posted on social media after a delayed flight. Passengers are complaining publicly on Facebook and drawing attention to the negative experience that they had. Should managers step in and publicly respond to these complaints? Could well-intended responses spiral into a storm of negativity?

These and other interactive-marketing questions have sparked considerable interest among managers eager to engage customers through social media platforms such as Facebook, Instagram and Twitter (Aral et al., 2013; Dong et al., 2024; Goh et al., 2013; Yang et al., 2019a, b; Wang et al., 2021). In this environment, users connect with one another, engage directly with brands and participate actively online – placing customer interaction at the very core of modern interactive marketing (Wang, 2023, 2024; Peltier et al., 2024). While consumer contributions online can be positive, negative complaints are also prevalent, posing challenges for brand managers. To address these, platforms have established response features that allow brands to publicly respond to consumer complaints and protect their image. Given that negative feedback can severely damage a brand’s reputation, sales and even stock prices (Chen et al., 2011; Chintagunta et al., 2010; Ho-Dac et al., 2013; Liu, 2006; Luo, 2009; Moe and Schweidel, 2012; Ward and Ostrom, 2006), it’s crucial for managers to understand the implications of responding to negative comments on social media.

Offline, brands often respond directly to consumer complaints to mitigate the impact of product or service failures (Fornell and Wernerfelt, 1988; Homburg and Fürst, 2005). Theoretically, such responses can address complaints and reduce future issues, potentially increasing customer satisfaction, loyalty and future purchases (Smith and Bolton, 1998; Tax et al., 1998). These actions can also improve future brand evaluations (Knox and van Oest, 2014; Van Laer and de Ruyter, 2010).

However, in online brand communities, the impact of managerial responses is more complex. Public responses to complaints might encourage others to imitate negative sentiments, especially if compensation is involved, potentially leading to a vicious cycle where more users voice complaints to seek similar rewards (Smith and Bolton, 1998; Luo, 2007; Chevalier et al., 2018). Conversely, resolving a complaint publicly might reassure other users of the brand’s sincerity, creating a virtuous cycle that reduces overall complaints (Tax et al., 1998; Bapna et al., 2019; Yang et al., 2019a). While there has been evidence for both positive and negative impacts of responding, it is unclear what are the types of responses that firms should provide.

This study examines the impact of firms’ responses to negative comments on subsequent consumer behavior within social brand communities. We leverage a unique dataset from Facebook brand pages across the airline, automotive and hotel industries, we analyze public activities, including posts, comments and responses. However, there are formidable challenges in identifying the causal impact of firms’ managerial responses to negative comments on subsequent consumer complaint behavior. First, a brand’s decision to respond to consumer complaints is not random but rather endogenous. Second, each post on Facebook is ranked based on Facebook’s algorithms. As researchers do not have access to the ranking of each post for each user, Facebook’s ranking algorithm could boost posts with managerial responses. To address these issues, we utilize an exogenous Facebook policy change in a natural quasi-experiment design, supported by instrumental variables and difference-in-differences (DID) models. By comparing firms that responded to complaints before and after the policy change with non-responding firms, and using different response types (e.g. compensation vs. apology), we control for potential endogeneity related to Facebook’s algorithms.

Our results provide evidence that responding to negative comments increases the proportion and number of subsequent negative posts and intensifies the negative tone. The effects are primarily driven by response text that contains compensation-related content (e.g. a refund, discount or freebie). Thus, paradoxically, a firm’s public response with a financial incentive to address one complaint fuels other users following the brand to voice their own complaints to get reparation and compensation from the responding brand, in a vicious cycle. Compensation responses not only increase the proportion of negative comments but also decrease the sentiment of comments, suggesting an economic incentive mechanism for the vicious cycle of responding to negative comments on social media brand pages. By contrast, we find that apology-based responses (e.g. sorry or apologize) have no significant impact on the number and proportion of negative comments. Thus, our findings reveal that the heterogeneity in managerial response types is crucially important.

Our work contributes to this literature in the following ways. First, we seek to provide empirical evidence of the impact of both the occurrence and message type of managerial responses in the context of brand communities, complementing existing work that documents the sole effects of the occurrence of managerial responses (Chevalier et al., 2018, Proserpio and Zervas, 2017; Kumar et al., 2018; Wang and Chaudhry, 2018; Chen et al., 2019; Yang et al., 2019b). Second, we provide insights into why some firm response types could drive a vicious cycle of complaints. Responding with a form of financial compensation encourages more negative comments, while apologizing or using other non-compensation-based responses discourages negative comments. These findings shed light on the nuances in textual response types that are essential for firms to succeed in the current bilateral communication environment of interactive marketing (Wang, 2024).

Our work is related to two streams of literature: (1) online word of mouth and brand communities and (2) responses to consumer complaints.

The literature extensively explores the drivers and impact of user-generated content such as online reviews and brand community posts. Early studies, including those by Chevalier and Mayzlin (2006), show that online review valence influences sales rank for online retailers. Other researchers have established similar relationships for other products (e.g. Chen et al., 2011; Chintagunta et al., 2010; Dellarocas et al., 2007; Duan et al., 2008; Liu, 2006; Luo, 2009; Moe and Trusov, 2011). Research also examines consumers’ motivations for posting, highlighting self-selection effects (Li and Hitt, 2008), dual decisions on whether and what to post (Moe and Schweidel, 2012) and the influence of rewards, both intrinsic and extrinsic (Khern-am-nuai et al., 2018). Research has also found that firms can stimulate reviews through financial incentives (Burtch et al., 2018).

A growing body of work addresses customer engagement in brand communities, key components of firms’ marketing strategies (Goh et al., 2013, Yang et al., 2019a, b, Ma et al., 2015). Research shows that firm posts stimulating engagement can promote community growth (Bapna et al., 2019) and identify how post valence and content characteristics affect user engagement (Yang et al., 2019a, b). Given that community engagement impacts consumer purchase behavior (Goh et al., 2013; Zhang et al., 2017), understanding how managerial responses influence engagement is critical.

The consumer complaint literature finds that negative opinions can potentially harm the firm with respect to sales, brand equity and company stock prices (East et al., 2008; Ho-Dac et al., 2013; Luo, 2009). Researchers have found that acknowledging the problem or offering some form of compensation can be an effective way to recover from customer complaints (Smith and Bolton, 1998). Other researchers have found that refunds, credits or apologies are often considered fair and sufficient (Tax et al., 1998).

More recently, in the online environment, researchers have found a mix of positive and negative effects of responding to consumer complaints. Managerial response to online ratings can have a positive impact on subsequent ratings (Wang and Chaudhry, 2018) but may lead to longer negative reviews (Proserpio and Zervas, 2017). Managerial responses can also improve the opinions of the responded customer (López-López et al., 2021; Zhao et al., 2020; Gunarathne et al., 2017). This can also increase credibility, which has been shown to affect purchase intentions (Muda and Hamzah, 2021). In contrast, Chevalier et al. (2018) found that responding to online ratings hotels results in more negative reviews. Firm responses can also impact the volume of subsequent reviews that they receive (Chen et al., 2019).

While the literature finds both positive and negative impacts of managerial response to online reviews, little attention has been paid to the unintended consequences in online brand communities – particularly the potential for a paradox where compensation responses actually intensify negative feedback rather than mitigating it. Day and Landon (1976) showed that offline customers often complain in order to receive services from firms. As a result, many firms often provide complaining customers with some type of compensation. For example, refunds, store credits or forms of compensation are often considered equitable and sufficient for customers who have a valid complaint (Smith and Bolton, 1998; Tax et al., 1998). However, in the online environment, managerial responses are observable to the entire community and could incentive others to complain as well, leading to a negative complaint cycle (Ma et al., 2015).

Theoretically, responding to complaints should alleviate the customer’s concerns. Timely complaint resolutions in private setting are often shown to increase customer satisfaction, loyalty and future purchases (e.g. Smith and Bolton, 1998; Tax et al., 1998; Knox and van Oest, 2014). Offline, firms routinely address complaints to mitigate service failures (Fornell and Wernerfelt, 1988; Homburg and Fürst, 2005). In the online setting, a managerial response could not only discourage indefensible short negative reviews (Proserpio and Zervas, 2017) but also encourage customers to check into the store (Kumar et al., 2018). This suggests that marketers might consider addressing negative word-of-mouth by replying to consumer’s online comments. The implication, for the focal user, is that a response might prevent additional negative comments.

However, in an online social context such as Facebook, responses are posted publicly and observable (Javornik et al., 2020). While intuitively it seems beneficial for a firm to respond publicly to negative comments, we posit an intriguing paradox – responses meant to resolve complaints, especially those including financial compensation, incite a greater volume of subsequent negative interactions, creating a potentially vicious feedback loop. We hypothesize that a response to negative comments. Two mechanisms may explain this: (1) consumers feel recognized by the firm, enhancing their self-image (Hennig-Thurau et al., 2004), and (2) responses offering refunds or discounts provide direct economic incentives for others to post complaints (Hennig-Thurau et al., 2004). We argue the latter is the dominant mechanism, leading to our first hypothesis:

We note that this mechanism should occur only if other users observe the initial response. On social media, there are certain characteristics of firms or posts that increase the likelihood that other users observe the response. These include the number of followers for the firm and the number of “likes” for the post (Toubia and Stephen, 2013; Zhang et al., 2016). Based on this, we propose the following hypotheses:

To provide further theoretical foundations for the economic utility mechanism, we investigate the types of responses that a firm can post. We often observe several types of replies such as a message for the user to contact the firm, an apology or even some form of compensation (Smith and Bolton, 1998; Tax et al., 1998). Certain response messages that include a discount or a refund can be interpreted as an economic incentive for users to post content online (Hennig-Thurau et al., 2004; Cheung and Lee, 2012). As a result, we expect the effects to be stronger if firms respond with monetary incentives, which can encourage others to post their own negative comments. This leads to our third hypothesis:

We collected data from Facebook brand pages of the largest companies in the airline, automotive and hotel industries (see Web Appendix A for company names). Marketers post various types of content, including text, images and videos, on these pages. Users can comment on these posts, and while brands prefer positive feedback, they cannot control or prevent negative comments unless the commenting functionality is disabled. All the brand pages in our dataset allowed user comments.

The Facebook Graph API allows us to download all visible activities on a brand page, including posts, responses to comments, user comments, likes and shares. For each brand page, we collected all user–brand interactions since its inception. Our analysis focuses on the airline industry data from one year before and after the March 25, 2013 policy change (with additional data from the hotel and automotive industries). We selected 133 firms based on three criteria: (1) brand pages where at least 90% of comments are in English, as our sentiment analysis is tailored to English text; (2) pages with a minimum of 100,000 interactions since inception; (3) representativeness of major U.S. airlines, excluding small regional airlines that operate under major brand names.

On March 25, 2013, Facebook introduced a comment response interface, enhancing interaction between firms and consumers (see Figure 1). This policy change was implemented without prior notice, preventing firms from strategically timing their responses. Before the change, comments were posted chronologically, making it difficult for firms to respond directly to specific comments. Consequently, firm responses were rare before the policy change, with only 2 out of 133 firms using the “@name” convention to reply. After the policy change, brand page administrators gained the ability to directly respond to individual comments via a new reply button. This feature allowed firms to engage more effectively with customers, as responses now appear directly beneath the original comment. Following the policy change, 78% of firms responded to negative comments. Next, we describe our data and model.

Table 1 presents our data descriptives. Additional details of the data cleansing and sentiment identification are available in Web Appendix B. For the model and results section, we focus on the airline industry, which includes 67 firms. We include the results for the hotel and auto industry in the web appendix. Approximately 78% of airline firms responded to negative comments. The dataset contains about 2 million comments, with an average of 3.6 negative comments per post for the airlines, and 94,000 followers.

Figure 2 presents model-free evidence for the spike in negative comments after the Facebook policy change. It reports the results separately for firms that respond (the solid line indicates the response group) and firms that do not (the dashed line indicates the baseline group). The y-axis reports the average number of negative comments per post. The vertical dash-dot line indicates the date (i.e. March 25, 2013) when Facebook announced and implemented the new response feature (i.e. the policy date). There appears to be no difference before March 25 between the response group (treatment) and the baseline group (control) that did not respond to comments. After the policy implementation date, we observe a large visible spike in negative comments. In contrast, there appears to be little change in the number of negative comments for the baseline firms. This provides preliminary evidence that responding to negative comments could potentially increase the number of future negative comments.

Our models are based on the difference-in-differences (DID) framework and leverage the exogenous event study of Facebook policy change. It is unlikely that Facebook timed this policy change to benefit any particular firm or industry in our data set, making it an unexpected exogenous shock. However, firms may still self-select into groups that respond based on factors that might also affect negative commenting activity. Thus, we account for this self-selection and possible endogeneity bias. Our model specification is DID with self-selection correction. More specifically, we estimate the following endogenous DID model with IV self-selection correction:

where the dependent variable $NegativeComments$ is gauged by three measures: the proportion of negative comments, the number of negative comments per post and the average sentiment of the negative comments posted on firm j’s brand page during month t. Although we focus on negative comments, we also separately analyze total comments as a dependent variable because responses could boost all types of commenting activity. $Treatment$ is an indicator variable that is equal to one if the treated firm j follows responds to negative comments immediately after the policy change, and zero for baseline firms. $PostPolicy$ is an indicator variable that is equal to one if the time period occurs after the Facebook policy change, and zero for the time periods before it. We also include a vector of covariates X (e.g. the lagged number of followers who are exposed to the firm’s posts each month and the number of likes). Our main analyses use the two-month period before/after the Facebook policy change. This is because two months are a narrow window of time to attribute the differences to the Facebook policy change (rather than any other events), that is neither too long to be confounded by other events nor too short to capture the delayed effects of the Facebook policy change. Nevertheless, we also check results robustness by using 1 month before/after the policy, as well as 12 months before/after. We also assume that the error terms have a bivariate normal distribution with mean of 0 and covariance matrix $[σ2ρσρσ1]$⁠. The main estimate of interest is the DID parameter of $β1$⁠, which can gauge how firms that respond to negative comments affect different levels of negative comments relative to firms that do not respond after the Facebook’s policy change relative to before it.

With our differencing approach, we assume that parallel trends hold. That is, in the absence of managerial response, the differences between the treatment group and the control group are constant over time. Visual inspection of Figure 2 provides initial evidence that the two groups have similar trends pre-policy change. Moreover, the trends for both groups are similar in levels pre-policy change. Of course, we recognize that any violation of this assumption can result in biased estimates. As a result, we gather additional monthly complaint data from the US Department of Transportation’s Air Travel Consumer Report (USDOT) to be used as a counterfactual of the trend that the response group would have followed in a no response scenario. The USDOT tracks user reported complaints for all airlines and provides a monthly count, which gives us a baseline level of complaints on a platform where firms do not directly respond. We collect the monthly number of complaints reported by the USDOT for the 12-month pre and post policy change period for each firm.

We have three main endogeneity concerns: time-invariant unobservables, time-variant unobservables and the impact of Facebook’s algorithmic ranking. The first concern is unobservable factors, like a firm’s social media expertise or general quality. Our model accounts for these time-invariant firm-level factors by comparing the levels of negative comments before and after the policy change across treated and baseline firms. As long as the impact of social media expertise remains constant over time, the pre-policy controls will account for such unobservables. We also address time-invariant unobservables related to general social media conditions by including baseline firms that do not respond as a control group. Additionally, the shorter time window used in our analysis mitigates concerns related to firm quality, aligning with the short-window identification strategy used by Chevalier et al. (2018).

Second, time-variant factors, such as temporary quality issues around the policy change, could also influence both response decisions and negative comments. To address this, we include prior period sentiment controls to account for recent changes in negative commenting activity, such as temporary quality shifts, which might influence a firm’s decision to respond.

Third, we also consider Facebook’s algorithm, which might boost the ranking of posts with firm responses, attracting more comments. To mitigate this, we include the number of likes as a proxy for post visibility in our model, assuming that posts with more likes are ranked higher. This approach should yield consistent estimates as long as the ranking factor is uncorrelated with the decision to respond. Further, we use monthly complaint data from the USDOT’s Air Travel Consumer Report as a control. This data captures baseline complaint levels over time for airlines, where firms cannot respond to complaints, helping to control for time-variant factors like weather or flight delays.

Finally, we include instrumental variables, such as the number of employees and the length of time a brand has been on Facebook, to address unobservables and reverse causality. These variables may influence a firm’s response decision but are assumed to be orthogonal to time-varying unobservables around the policy change. Since the policy change was unexpected, firms could not adjust these variables in response, ensuring their suitability as instruments.

Table 2 reports the results. We use the etregress package in STATA to estimate our model. Columns (1) and (2) examine the impact of responding on the proportion and absolute number of negative comments. This provides our primary analysis of the impact of managerial response on brand pages. We find a significant positive coefficient on the $Treatment×PostPolicy$ interaction (⁠$β1=0.243$ and $β1=3.46$⁠, respectively), indicating that responding to negative comments increases the levels of negative comments. Model 1 controls for the overall commenting activity level by using a proportional measure for the dependent variable. Model 2 controls for the levels of commenting activity using the number of positive comments as a control. We note that in column (3), responding does not affect the total amount of commenting activity. Finally, column (4) assesses the impact on the sentiment of comments. The estimate suggests that responding decreases comment sentiment, or in other words exacerbates the intensity of complaints ^[1]. Thus, responding to negative comments has significant harmful implications for subsequent negative commenting on social brand communities. With respect to the other covariates, we find that the number of followers and the number of positive comments is positively associated with the number of negative and total commenting activity. In the short two-month window, our results also suggest that the previous period’s sentiment has no impact on commenting activity.

A potential explanation for why publicly replying to negative comments generates more negative comments is that it incentivizes consumers to also post negative comments. A response is generally written to satisfy the customer but observable to others, which paradoxically, might incentive others to post in order to also gain utility from the firm responding to their own post. We investigate this explanation by conducting two additional analyses. First, we explore the number of followers for each firm and the number of likes for each negative comment prior to the policy date. This allows us to test if greater exposure of the response to other users will moderate the effects. We expand Equation (1) to include the interaction between the treatment and the number of followers from each firm (and separately the interaction between the treatment and the number of likes from each firm). We note that for the Likes model, we calculate $Likesij,t−1$ for each individual i, as the total number of “likes” indicated by all other users across all previous comments posted by all other users (i.e. excluding user 1). Thus, this measure excludes comments posted by i and “likes” indicated by i. The identification assumption is that the comments posted by i does not influence others to “like” other user’s comments. This is likely to be reasonable given that there are on average 150,000 followers for the brand pages in our dataset (less likely for one comment to impact the “likes” for other comments).

Table 3 reports the results. We observe a significant coefficient on $Treatment×PostPolicy×$ $Followers$ (⁠$β=4.698$⁠. Thus, consistent with H2a, these findings confirm that to the extent firms that have more followers receive more attention for the messages they post, the treatment effect is indeed stronger in the case of more followers in the post policy period ^[2]. We also observe a significant coefficient on $Treatment×PostPeriod×Likes$ (⁠$β=0.00039$ ^[3]. Thus, in support of H2b, the impact of replying on the number of negative comments is stronger if a firm’s negative comments are highlighted to the public through greater number of “likes”. That is, publicly replying to negative comments generates more negative comments because it brings to attention to negative issues to more users (through the number of follower sand likes) and incentivizes more consumers to imitate and voice negative comments.

On social media, firms could craft several types of responses to their customers. In our data, many of the response types fall into two categories: compensation and apology (see Web Appendix C for examples). Thus, in addition to assessing whether to respond to negative commenting activity (e.g. Ma et al., 2015; Chevalier et al., 2018), we also investigate how to respond. In this section, we assess the impact of the firm’s response type on subsequent negative commenting activity.

In our analysis, we gauge response types with two distinct measures. First, we create categorical dummy variables indicating whether a firm’s response falls under one of the three categories: compensation, apology and others, based on the response text. That is, if a firm uses a financial incentive related keywords (refund, upgrade or freebie) in its response text, then the compensation dummy variable is equal to one for that firm, and zero otherwise. Similarly, if a firm uses the apology-related keywords (sorry, apologize) in its response text, then the apology dummy variable is equal to one and zero otherwise. The baseline response type is a generic response with a general acknowledgement and no financial incentive or apology in the text. Second, we generate continuous variables that measure the latent probability that a firm’s response belongs to one of the three categories, based on the Latent Dirichlet Allocation (Tirunillai and Tellis, 2014; Toubia et al., 2019). Specifically, we leverage the guided Latent Dirichlet Allocation approach (Toubia et al., 2019) that models our latent topics according to the themes of apology and compensation, as guided by previous consumer complaint literature (e.g. Luo, 2007; Smith and Bolton, 1998; Tax et al., 1998) ^[4]. This enables us to create continuous measures for the variables of apology and compensation, which provides flexibility to categorize a message under multiple topics based on a continuous probability measure. See more details of guided LDA results in Web Appendix C.

With these response type variables, we specify the following DID model with interactions between compensation and apology response type with the post-policy period:

Note that in this model, apologizing or providing compensation is conditional on responding. Thus, the $Treatment×PostPolicy$ coefficient estimates the lift for responding group relative to the baseline group, the $Compensation×PostPolicy$ coefficient estimates the impact of a compensation response after the Facebook policy change and the $Apologize×PostPolicy$ coefficient estimates the impact of an apologizing response after the Facebook policy change.

Tables 4 and 5 report the results with categorical and continuous variables of apology and compensation, respectively. Consistent with an economic-based mechanism, responding with compensation (with both categorical and continuous measures) indeed increases both the proportion and absolute levels of negative comments and decreases the sentiment of comments (Columns 1, 2 and 4). In contrast, we find that apology-based responses have no significant effects. Thus, our findings reveal a striking paradox: managerial responses offering compensation actually increase negativity, confirming our suspicion of a “vicious cycle.” In contrast, simply apologizing deflates further negativity.

Thus, these findings suggest that the vicious cycle of negativity is driven by financial compensation responses, rather than apology responses. In other words, the vicious cycle of responding to negative comments on social media brand pages manifests, such that the adverse impact becomes more severe when firms offer financial compensation in their response content on Facebook brand communities. This is consistent with H3, which predicts that messages with economic utility can lead to greater levels of subsequent negative commenting activity. Moreover, similar to our main findings, there is no statistically significant impact of compensation-based responses on the number of total comments or positive comments. With respect to apology responses, we find that it has no significant effect on these types of commenting activity. Furthermore, the larger magnitude estimates on treatment x post policy indicates that our results are primarily driven by firms that use compensation response types.

Online social media has given consumers countless opportunities to share their opinions and access those of their peers. However, marketers face the challenge of deciding how to respond to negative feedback on platforms like Facebook. In this paper, we examine how firms’ responses to negative comments influence other consumers’ willingness to post feedback. We exploit a change in Facebook’s policy that allowed firms and users to interact directly on brand pages and introduce a machine learning technique to analyze unstructured social media data, including emoticons and Internet slang. Our research resolves the managerial paradox by clearly delineating response strategies: compensatory gestures cannot be handled publicly to avoid incentivizing a negative spiral, while public apologies offer a safer, reputation-preserving path. For marketers, this means navigating the fine line between generosity and restraint to manage the social community effectively. They must also be judicious with compensatory responses, ensuring these gestures don’t inadvertently spark additional negativity.

Our study makes several important contributions to the existing literature. First, we provide empirical insights into the dual impact of managerial response in brand communities, examining not only the occurrence of such responses but also the nuances of message type. This expands upon prior work that primarily focused on the presence of managerial responses (e.g. Chevalier et al., 2018; Proserpio and Zervas, 2017; Kumar et al., 2018; Wang and Chaudhry, 2018; Chen et al., 2019). Second, we explore the heterogeneity of response effects, uncovering that financial compensation in responses tends to elicit more negative comments, whereas non-compensatory responses, such as apologies, have a mitigating effect. By highlighting these distinctions, our findings add depth to the understanding of how textual variations in managerial responses can drive divergent consumer behaviors, addressing an underexplored dimension in the existing body of research. This allows researchers and managers to have a better understanding of how to engage with their audience online, which is a critical characteristic of interactive marketing (Wang, 2025, 2023).

Our research is, of course, not without its limitations. First, we are unable to definitively assert whether responding to negative comments affects company sales. On the one hand, responding leads to more negative comments on Facebook brand communities, which might have a downward relationship with sales. On the other hand, responding also increases the number of followers, which could be positively related to sales. In addition, since responding reveals complaints that would not have been posted otherwise, firms can now address the new complaints in order to increase customer lifetime value. Managers should assess the potential costs and benefits with respect to other important metrics before making their decision on whether and how to respond. Second, our research does not assess how consumers interpret negative comments once the firm responds. Many customers who post negative comments get their grievances heard and addressed. This could be a potential benefit of responding to negative comments. Also, we are unable to assess whether consumers’ issues were properly addressed with our current data. Thus, future studies might investigate how responding affects satisfaction and other customer loyalty metrics in the context of managerial responses on Facebook brand pages. Moreover, our data consists of mostly large companies. Consumers may react differently to communication from smaller community businesses. For example, a local small business might receive a kinder or gentler reaction from a consumer when compared to a larger multi-national corporation. Future research on smaller companies and brands could find interesting moderation effects. In addition, at firms could respond to negative posts privately. This strategy would eliminate the observability of the response, thus reducing concerns of the vicious cycle. Our current data does not allow us to track private conversations and thus is outside the scope of this research. Finally, our results are based on only one social media platform of Facebook in a specific time window. Thus, the findings may not be generalizable to other social media platforms for other companies in other time periods. Future research may examine the dynamic effects of different algorithm types and boundary conditions of managerial responses on Twitter, Snapchat, Pinterest, Instagram, TikTok and other social media platforms.

In conclusion, this research highlights the potential drawbacks of firms responding to negative comments on Facebook brand pages, and we encourage further studies in social media marketing.

The supplementary material for this article can be found online.