Directing Audience Responses: How Nonprofit Donation Appeals Shape User Comments on Social Media

Sentiment

Incentive motivation theory posits that emotionally charged messages function as external stimuli that energize and direct behavior by activating individuals’ tendencies to seek positive affective rewards or to alleviate aversive emotional states (Berridge, 2018). In the context of nonprofit communication, positive emotional appeals can foster favorable attitudes toward the organization and reinforce affective alignment with its mission, thereby encouraging expressions of appreciation and support (Erlandsson et al., 2018). Thus, positive sentiment is expected to increase gratitude-oriented comments because it frames the appeal around hope, appreciation, progress, or successful care rather than immediate distress. Such appeals may reinforce users’ favorable affect toward the organization, beneficiaries, or broader mission.

Conversely, negative emotional appeals heighten perceived need and moral concern, which in turn elicit sympathy and increase prosocial responses (Li et al., 2022). When translated into interactive social media environments, these affective cues not only shape engagement volume but also structure the semantic orientation of user comments. Negative sentiment is expected to increase action-oriented comments because it signals urgency, suffering, or unmet need, making the cost of inaction more emotionally salient. The same negative cues should also increase sympathy-oriented comments because they direct attention to the hardship of beneficiaries and invite empathic concern.

Accordingly, this study theorizes that emotional valence embedded in donation appeals systematically influences the types of comments they generate:

Narrative Mode

In addition to emotional framing, the narrative mode of donation appeals shapes how audiences interpret messages and respond behaviorally. Narrative mode refers to the perspective and focalization through which a message is conveyed, such as who is speaking, who is being addressed, and whose experience is foregrounded. Prior research on fundraising communication highlights the importance of narrative perspective, like first-person versus third-person storytelling, in influencing viewers’ intentions and engagement (Wang et al., 2022). In nonprofit contexts, first-person plural language (e.g., “we”) is commonly used to represent the organization and signal collective identity and mission alignment.

Within third-person narratives, however, important distinctions arise between personalized narratives that emphasize a single, identifiable beneficiary and group-centered narratives that portray outcomes affecting a broader collective. Building on the identifiable versus statistical victim framework, personalized narratives tend to elicit stronger engagement than abstract or aggregated portrayals (Mao & Nishide, 2025). By centering the appeal on a concrete individual with a name, image, or story, personalized narratives reduce psychological distance and allow audiences to more easily empathize with the beneficiary’s situation. This heightened empathic connection increases the likelihood that users express emotional reactions such as sympathy and gratitude toward the organization’s efforts.

At the same time, second-person, donor-centric narratives that directly address the audience using “you” position donors as active agents in problem solving, increasing perceived efficacy and motivation to act (NextAfter, 2023). Such narratives frame the audience as capable contributors rather than passive observers, lowering the cognitive and motivational barriers to action. By explicitly assigning responsibility and agency to the audience, this framing makes behavioral responses more salient, thereby increasing the likelihood of action-oriented comments.

Accordingly, this study theorizes that:

Obtaining Nonprofit List

The sampling of nonprofits is based on the CauseIQ Nonprofit Directory (www.causeiq.com), which contains information on around 425,000 nonprofits across the United States. It collects nonprofit data pulled from IRS tax forms, self-reported information, and other sources. The website allows users to view a list of nonprofits categorized by State and the National Taxonomy of Exempt Entities (NTEE) code. NTEE is used by the Internal Revenue Service (IRS) and the National Center for Charitable Statistics (NCCS) to classify NONPROFITS into 25 named categories plus 1 unknown category.

To provide contextual justification for the focal category, this study concentrates on NTEE code D (animal-related organizations). Prior research indicates that animal-related nonprofits are among the most active in fundraising engagement (Tuckman & Chang, 1998). Empirical evidence further suggests that these organizations attract more donors per organization than any other program service area, as observed in the Omaha Gives campaign (Bhati & McDonnell, 2020). This pattern is often attributed to donors’ stronger emotional attachment to animal welfare causes, which enhances responsiveness to fundraising appeals. Thus, animal-related nonprofits provide a solid context for examining donation appeals and user engagement on social media.

To capture a representative sample of the animal-related nonprofits across the United States, this study first employed a stratified sampling approach based on geographic distribution. The sample reflects the diversity of animal-related nonprofit regional representation. This research viewed the lists in descending order by revenue and recorded up to 60 results in the animal-related category in each of the 49 continental U.S. states (excluding Hawaii). The revenue ranking helps prioritize organizations with an operational scale and financial footprint to utilize social media for fundraising and stakeholder engagement. The initial list contains a total of 2,940 animal-related nonprofit names from CauseIQ.

Checking Facebook Presence

Through the search function in MCL API, nonprofit names captured from CauseIQ served as the search terms for finding the corresponding Facebook account. Specifically, the Facebook/pages/preview endpoint was used to conduct Facebook Page searches, and the first returned result was recorded as the presumed official account. The saved information includes the account name, Facebook ID, description, and other information. However, not every nonprofit has a Facebook account; there were also many irrelevant accounts in the results.

Comparing Name Similarity

To ensure the accounts actually belong to the nonprofits, this research tested the text similarity between nonprofit names in the CauseIQ list and the Facebook account name. The cosine similarity (scaled from 0 to 1) was first consulted to measure similarity based on words in the names. As this method only works accurately when the words in the name are separated by space, it returns zero (no similarity) when the nonprofit is named “XXX YYY ZZZ” and the account name is “XXXYYYZZZ.” Thus, a second method, the Jaccard score (scaled from 0 to 1), was used to measure similarity based on alphabet overlapping, which avoids space problems but has the possibility of making false judgments since it only compares the appearance of alphabets but not the spelling. In addition, some nonprofits use account names different from the organization’s name but have the full name in the account description. Thus, a Facebook account was considered correct if it met one of the two criteria: (a) at least one of the cosine similarity and Jaccard score is over 0.85 and (b) the nonprofit name is found in the account description. After screening, a total of 1,061 Facebook accounts were captured.

Kindly note that, given the variety of organization names, this study used 0.85 instead of 1.00 as the threshold because some names in the CauseIQ Nonprofit List differ slightly from their Facebook account names. For example, “Tenderloin Housing Clinic THC” has the capital of its full name, while its account name is “Tenderloin Housing Clinic.” Similarly, “Franklin Volunteer Fire Dept” has the abbreviation in its name while its account is “Franklin Volunteer Fire Department.” The threshold of 0.85 is a conservative estimate intended to minimize the risk of overlooking valid accounts but may occasionally result in the inclusion of some erroneous accounts.

Collecting Donation Appeals

Through MCL API, the Facebook/posts/preview endpoint was used to retrieve Facebook posts associated with each nonprofit’s Facebook ID, filtering posts that contained any of three donation-related keywords: “donate,” “donation,” or “donating.” The keywords were separated by “OR” so the search will return the posts containing any of the keywords. Given substantial variation in the volume of donation appeals across nonprofits, each query was capped at a maximum of 50 posts and ordered by view count in descending order, prioritizing the most widely viewed appeals. The Facebook accounts that never posted any donation-related content were then removed from the dataset. As a result, 45,647 donation appeals from 1,020 Facebook accounts were collected.

Collecting Comments

The Facebook/comments/job endpoint was used to retrieve user comments associated with each donation appeal based on the post IDs obtained in the previous step. Due to API quota limitations, and to ensure analytical relevance and sufficient user engagement, it was necessary to reduce the number of eligible posts for comment collection. Donation appeals were first sorted by the number of comments in descending order, and only the top 20 posts per nonprofit account were retained for comment collection. Posts were further filtered to include only those with at least 100 views, 10 comments, and one recorded reaction. From this refined set of highly engaged donation appeals, up to 5,000 comments per post were retrieved, forming the final dataset for comment-level analysis. Comments authored by the nonprofit organizations themselves were excluded to ensure that the dataset captured only user-generated responses.

The final dataset consists of 674,845 user comments associated with 14,319 donation appeals posted by 794 animal-related nonprofit Facebook accounts.

Coding: Donation Appeals

The sentiment of donation appeals was coded using VADER (Valence Aware Dictionary for sEntiment Reasoning) implemented in Python, a parsimonious, rule-based sentiment analysis model specifically designed for social media text (Hutto & Gilbert, 2014). VADER is based on a sentiment lexicon of over 7,500 words, each assigned a polarity indicating its positive or negative meaning. Using this lexicon-based method, VADER generated three sentiment scores, positive, neutral, and negative, for each donation appeal. These scores are expressed as proportions ranging from 0 to 1, representing the share of text classified into each sentiment category, and they sum to 1 for each appeal. Prior to analysis, domain-specific words including donate, donating, and donation were removed from the text. This preprocessing step was intended to avoid mechanically inflating positive sentiment scores due to the inherently positive connotation of donation-related terms.

The narrative modes of donation appeals were operationalized through a dictionary-based keyword coding scheme focused on pronoun usage, as illustrated in the coding rules. Second-person, donor-centric narratives were identified by the presence of second-person pronouns (e.g., you, your, yours, yourself, yourselves), which directly address the audience and position the donor as an active agent in the appeal.

Third-person narratives were divided into two subcategories. Personalized narratives were coded based on singular, gendered third-person pronouns (e.g., he, him, his, himself, she, her, hers, herself). In contrast, group-centered narratives were identified through plural third-person pronouns (e.g., they, them, their, theirs, themselves), emphasizing collective beneficiaries rather than a single person.

In addition, structural features of donation appeals were coded using metadata obtained from the Meta Content Library (MCL) API. Based on the multimedia information provided by the API, each appeal was classified by media type into text-only, photo-based, or video-based posts. These categories were operationalized as a categorical variable to capture differences in narrative medium that may influence audience interpretation and engagement (Mao & Nishide, 2025). Furthermore, a separate dummy variable was created to indicate whether a donation appeal contained an embedded URL link. This variable captures the presence of direct pathways to external information or donation pages, allowing the analysis to account for the potential role of call-to-action accessibility and informational cues in shaping audience responses.

Coding: Comments

User comments were also coded using a dictionary-based keyword approach. Three distinct dictionaries were constructed to capture substantively meaningful patterns in user responses to donation appeals: Action, Gratitude, and Sympathy. Each dictionary consisted of carefully curated keywords and short phrases commonly observed in nonprofit comment sections. For example, Action-oriented comments were identified by the presence of words or phrases indicating concrete engagement or behavioral commitment, such as making a donation, completing a payment, sharing the appeal, or explicitly confirming participation (e.g., “donated,” “sent,” “shared,” “done”). Gratitude-oriented comments were coded based on expressions of appreciation, positive evaluation, or blessings directed toward the organization or its mission, while Sympathy-oriented comments captured language conveying compassion, concern, or emotional support for beneficiaries or victims. A comment was assigned a value of 1 for a given category if it contained at least one corresponding keyword or phrase, and 0 otherwise. Because comments could contain multiple semantic signals, categories were not treated as mutually exclusive, allowing a single comment to be coded into more than one category when applicable. The full dictionary can be found in the Appendix.

Reliability Test

While VADER is designed to capture sentiment nuances in informal and colloquial language commonly found in social media (Arora et al., 2024), it remains a lexicon-based method whose performance depends on the coverage and appropriateness of its predefined vocabulary. To assess the reliability of sentiment analysis in this study, a validation exercise was conducted on a subset of 100 randomly selected posts. The positive and negative words identified by VADER for each appeal were first extracted for review. A human reviewer then examined these extracted words alongside the full post and corresponding sentiment scores, and indicated whether they agreed or disagreed with the results. Similarly, to assess the reliability of the dictionary-based keyword classification applied to user comments, a validation exercise was conducted using a random sample of 200 comments. Given that this approach relies on predefined keyword dictionaries to assign comments to categories, its performance depends on the adequacy and specificity of the selected terms. A human reviewer examined the classification results for each comment and evaluated whether the assigned category accurately reflected the content.

Overall, agreement was achieved in over 95% of the cases. Regarding post sentiment, discrepancies primarily arose from limitations in lexicon coverage and context sensitivity; for example, certain uncommon negative terms such as “malnourished” and “dehydrated” were not captured, while contextually neutral words like “original” were assigned positive valence and words such as “accidentally” were assigned negative valence. For comment classification, discrepancies were mainly driven by the variability and ambiguity of natural language expressions, including partial phrase matches (e.g., “I did” versus “I did not”). Additional discrepancies stemmed from contextual misclassification, such as phrases like “mail it to you” that indicate concrete action but were not captured by the keyword dictionary, and phrases like “send love” being interpreted as action-oriented when they do not necessarily reflect actual donation behavior. Despite these limitations in coverage, the overall dictionary performs well, and because the discrepancies were infrequent and did not materially alter the overall patterns, the results are considered acceptable for the purposes of this study.

Sentiment of Donation Appeals

Overall, the appeals exhibit a clear predominance of positive over negative language, consistent with the inherently prosocial and affirmational nature of charitable solicitation. Table 1 illustrates the top 10 most frequent positive and negative words captured in the 14,319 donation appeals, revealing substantial variation in both the volume and function of emotional expressions. Notably, the frequency of highly salient positive terms, such as help, exceeds the total number of appeals, indicating that these words are often repeated multiple times within individual posts rather than appearing only once per appeal. For example, one appeal states: “Our Humane Officer brought in 10 cats today that a woman in Somerset is/was trying to help . We are asking for help with donations of kitten food, wet and dry, and pine litter. Monetary donations for their medical care would be appreciated and you can donate through our website. . . Thank you.” This excerpt illustrates how positive emotional language is repeatedly deployed within a single appeal to reinforce compassion and collective responsibility.

OPEN IN VIEWER

Table 1.

Most Frequent Sentiment Word in Donation Appeals.

Top	Positive word	Count	Negative word	Count
1	Help	16,433	No	4,166
2	Care	10,301	Emergency	1,574
3	Please	10,160	Pain	1,139
4	Thank	7,528	Hard	1,112
5	Like	7,172	Poor	741
6	Support	5,214	Broken	713
7	Rescue	4,754	Drop	710
8	Love	3,864	Unfortunately	709
9	Well	3,066	Alone	676
10	Good	2,405	Low	663

By contrast, the overall frequency of negative words is considerably lower. The most frequent negative term is no, which typically appears in constructions such as “no creature deserves that,” “no resources,” or “no time,” reflecting constraints, deprivation, or moral negation rather than direct expressions of hostility. In summary, positive words primarily emphasize themes of help, care, and gratitude, underscoring nonprofits’ efforts to frame giving as compassionate, supportive, and socially meaningful. In contrast, negative sentiment appears less frequently and is concentrated on terms related to emergency and pain, which are used strategically to signal urgency and highlight beneficiaries’ need.

Narrative Mode of Donation Appeals

Table 2 presents the frequency and percentage of personalized narratives, group-centered narratives, and second-person, donor-centric narratives across all donation appeals. Overall, group-centered narratives appear more frequently than personalized narratives (58% versus 49%), suggesting that nonprofits more often frame need at the collective level rather than focusing exclusively on a single identifiable beneficiary. It is important to note that, as two distinct forms of third-person narration, personalized and group-centered narratives are not mutually exclusive and can co-occur within the same appeal. For instance, an appeal may reference a specific individual while simultaneously situating that individual within a broader collective or species, as illustrated by the statement: “Little Miss Walrus is giving you her best raspberry to say, “It’s the last day for a sticker!” . . . donation now helps our program give our patients the best chance at life, whether returning to where they were found, or thriving in their new forever homes as ambassadors for their species.” In this example, the narrative foregrounds a particular animal while also emphasizing the well-being of the larger group to which it belongs.

OPEN IN VIEWER

Table 2.

Distribution of Narrative Mode.

Narrative	Count	Percent
Personalized	7,006	49
Group-centered	8,303	58
Second-person	12,579	88

Meanwhile, second-person, donor-centric narratives are used most extensively, appearing in 88% of all donation appeals. This narrative mode facilitates direct communication with potential donors by explicitly addressing them and inviting action through phrases such as “if you can donate please do,” “help in any way you can,” and “here is how you can help,” thereby positioning donors as active participants in the charitable process.

Other Features of Donation Appeals

Donation appeals also vary in the forms of media they employ, including text, photos, videos, and external links. Table 3 demonstrates the frequency distribution of these media types across all donation appeals. It is important to note that Facebook donation appeals always contain some textual content; therefore, text in this context refers specifically to appeals that rely on pure text without accompanying other media. The majority of donation appeals incorporate photos (77%), highlighting the central role of visual imagery in fundraising communication. In comparison, a smaller proportion of appeals use videos (10%) or rely solely on pure text (9%), while only a limited share of appeals include external links to other media, such as news coverage or off-platform videos (3%). Meanwhile, 46% of donation appeals contain embedded URLs directing users to the nonprofit’s external donation webpages (Table 4), indicating that nearly half of appeals explicitly provide direct access to donation portals.

OPEN IN VIEWER

Table 3.

Distribution of Narrative Media.

Media type	Count	Percent
Photo	11,064	77
Video	1,508	11
Text	1,337	9
Links	410	3
Total	14,319	100

OPEN IN VIEWER

Table 4.

Distribution of Appeals with URL Link.

Has URL	Count	Percent
No	7,693	54
Yes	6,626	46
Total	14,319	100

In addition, the length of the textual content in donation appeals varies substantially, ranging from as few as three words to more than 2,000 words, with an average length of 178 words. The shortest appeals are often highly concise and action-oriented, such as “Welcome, Sasha! tcwr.org/donate,” which serves to redirect audiences to an external news story about a rescued cougar named Sasha. Other brief appeals similarly rely on minimal text, phrases like “Please donate and share” or “Thank you for your donation,” typically accompanied by photos that convey context visually. In contrast, longer donation appeals adopt a narrative, diary-like style, providing detailed stories about individual people or animals, elaborating on their circumstances, personal feelings, and ongoing needs.

Action-Oriented Comments

Action-oriented comments constitute approximately 13% of all 674,825 user comments responding to nonprofit donation appeals (Table 5). These comments primarily reflect self-reported behavioral engagement, in which users explicitly state actions they have already taken in response to the appeal. Typical expressions include concise confirmations such as “Donation made,” “Sent two bags and some kitten food,” or brief acknowledgments like “Done !.” Rather than conveying reactions or moral support alone, action-oriented comments are signals of compliance and participation, indicating that the appeal successfully translated exposure into concrete action.

OPEN IN VIEWER

Table 5.

Distribution of Comments Type.

Comment	Count	Percent
Action	88,791	13
Gratitude	275,819	41
Sympathy	100,874	15
Other	292,883	43

Gratitude-Oriented Comment

Gratitude-oriented comments represent the largest group of all three types of comments, accounting for approximately 41% of all comments posted under nonprofit donation appeals. These comments primarily express appreciation directed toward nonprofit organizations and affiliated individuals, acknowledging their efforts, impact, and commitment. Typical examples include statements such as “I will always be thankful for the work that you do. If not for your organization, we would not have our beautiful Noah!!!” and “A big thank you to everyone who adopted pets and to DCHS for all you do.” In addition to organization-focused appreciation, a subset of gratitude-oriented comments conveys positive affect toward the beneficiaries, often through expressions of happiness, hope, or prayer (e.g., “So happy for them all” or “Sending prayers for Bunny”). Collectively, the gratitude-oriented comments primarily recognize nonprofit efforts while simultaneously expressing goodwill toward those the appeals aim to support.

Sympathy-Oriented Comments

Sympathy-oriented comments account for approximately 15% of all user comments responding to nonprofit donation appeals. These comments primarily convey empathetic and compassionate reactions toward the beneficiaries featured in the appeals, often emphasizing their vulnerability, suffering, or need for care. Typical expressions include emotionally charged statements such as “Oh my heart you poor sweet babies,” “Such sad eyes!!!!,” and “So heartbreaking to do to an animal. Hopefully he’s in good hands and will find a loving home.” Rather than reporting concrete actions or offering direct gratitude, sympathy-oriented comments reflect users’ affective engagement with the plight of beneficiaries, signaling emotional resonance and moral concern.

Other Comments

It is important to note that the three comment types are not mutually exclusive, as users may simultaneously express multiple sentiments or behaviors within a single comment. For example, comments such as “DONATED! Poor sweet baby, hope she gets all the help she needs and finds a wonderful home” contain elements of action, sympathy, and gratitude concurrently. Beyond these overlapping cases, approximately 43% of all comments were not classified into any of the three focal categories. A substantial portion of these uncategorized comments consisted of brief affective reactions directed toward the animals themselves, including remarks such as “how cute,” “adorable,” “beautiful girl,” or “WOW!” Given that the dataset primarily comprises donation appeals from animal-related nonprofit organizations, the prevalence of such appearance-focused and exclamatory comments is unsurprising. While these comments contribute to overall engagement and visibility, they do not directly reflect the behavioral, appreciative, or empathetic responses that are the central focus of this study.

Action-Oriented Comments

Consistent with Hypotheses H1a and H2a, action-oriented comments are more likely to appear in response to donation appeals that contain stronger negative sentiment and employ second-person, donor-centric narratives, confirming the mobilizing effect of urgency- and agency-focused messaging. In contrast, and somewhat unexpectedly, both third-person narrative modes, personalized narratives emphasizing an individual beneficiary and group-centered narratives highlighting a collective, significantly decrease the likelihood of action-oriented comments. This suggests that shifting attention away from the donor’s role may dampen users’ propensity to publicly signal action.

In addition, donation appeals presented in video format are significantly less likely to elicit action-oriented comments compared to the more commonly used photo-based posts, possibly because watching videos requires greater time and cognitive investment, reducing immediate behavioral responses. Finally, appeals that include external links are more likely to generate action-oriented comments, indicating that providing access to supplementary information or engagement opportunities beyond the post itself may facilitate users’ expressions of action, whereas the inclusion of a direct donation URL within the appeal text does not exhibit a statistically significant effect.

Gratitude-Oriented Comments

Consistent with Hypotheses H1b and H2b, gratitude-oriented comments are more likely to appear in response to donation appeals that contain stronger positive sentiment and employ personalized narratives, suggesting that emotionally affirming messages centered on identifiable beneficiaries encourage expressions of appreciation.

In contrast to the pattern observed for action-oriented comments, the use of second-person narratives significantly decreases the likelihood of gratitude-oriented comments, indicating that direct calls to donor action may crowd out expressions of thanks. Similarly, the inclusion of a direct donation URL is associated with a lower probability of gratitude-oriented comments, potentially because transactional cues shift user attention away from affective responses.

Sympathy-Oriented Comments

Consistent with H1a, sympathy-oriented comments are more likely to appear in response to donation appeals that contain stronger negative sentiment, underscoring the role of distress-focused emotional framing in eliciting empathetic reactions. Interestingly, both forms of third-person narrative significantly increase the likelihood of sympathy-oriented comments, with the coefficient for personalized narratives larger than that for group-centered narratives, providing partial support for H2b.

Meanwhile, the use of second-person narratives significantly decreases sympathy-oriented comments, as direct appeals to the donor may shift attention away from the beneficiaries’ suffering. Similarly, donation appeals presented in video format are less likely to generate sympathy-oriented comments, possibly because videos require greater cognitive engagement which reduce users’ propensity to articulate empathetic responses.