Do communication content functions drive engagement among interest group audiences? An analysis of organizational communication on Twitter

Interest group organizations (IGOs) use social media to influence beliefs, mobilize collective action, and shape policy debates. Engagement on social media reflects users’ behavioral, cognitive, and affective responses (e.g., likes, comments, shares/retweets). Despite widespread adoption, many organizations struggle to achieve effective engagement, and prior research offers limited insight into which specific communication content functions drive engagement and how multimedia affects this relationship. This study asks: (a) What communication content functions do IGOs create on social media? (b) How do these functions affect user engagement? (c) How does multimedia inclusion interact with communication content functions to affect engagement? Addressing these questions is important for understanding how IGOs can design communications that maximize audience interaction and dissemination on Twitter.

The review situates IGOs’ social media use within resource-based and strategic communication perspectives, noting competing views of social media as tools for resource-poor NGOs versus established organizations. Social media are used to shape news/policy debates and manage organizational image, reflecting shifts toward professionalization in advocacy. Engagement behaviors online are influenced by account characteristics, contextual factors, and content-specific features, but fewer studies examine the function of posts per se. Prior work shows mixed findings on the role of multimedia; effectiveness may depend on alignment with content characteristics. The study adopts the information–community–action framework (Guo & Saxton, 2014) and extends it by extracting sub-functions via topic modeling, focusing on dissemination (retweets) as a key engagement outcome for IGOs.

Study context and sampling: Using the EU Transparency Register, 500 NGOs were randomly selected from 2,897; verified English-language official Twitter accounts were identified, yielding 121 IGO accounts for analysis. Data collection: Tweets were collected using Export Tweets Professional for the period September 2017–August 2019. A total of 12,766 tweets were gathered; 1,600 were used for coding reliability in accordance with Riffe et al. (2014). Text mining and topic modeling: To address short-text sparsity, a Bi-term Topic Model (BTM) with Variational Expectation Maximization was applied. Models with K=2–21 were evaluated; the 9-topic solution showed the highest coherence and perplexity fit (perplexity score −8.24346), revealing nine latent communication content functions. Coding scheme: Based on BTM outputs and an inductive review, nine sub-functions were coded: (1) Event (activities, updates, community work); (2) Report (research, group updates); (3) Trade (business/trade/news); (4) Period (date, time, season, occasions); (5) Cancer (health education/tips); (6) People (appreciation/recognition); (7) Unite (call for action); (8) Sign (online activism/petition); (9) Glean (garner support). Coding was performed by two authors and two trained coders. Inter-coder reliability (Cohen’s kappa) ranged from 0.72 to 0.98 across categories (e.g., Event 0.98; Report 0.92; People 0.84; Sign 0.89; Trade 0.87; Unite 0.96; Period 0.92; Cancer 0.77; Glean 0.72). Engagement measure and controls: Retweets per post served as the dependent variable. Multimedia inclusion (images/photos/videos) was coded dichotomously (1=present; 0=absent). Number of followers per organization controlled for group size. Statistical analysis: A multilevel generalized linear regression with a log link modeled the association between content functions and retweets, with and without multimedia inclusion. Model specification: Log(Y_ij) = β0j + β1j T_ij + β2j I_ij + β3j P_ij, where Y_ij is retweets, T_ij function type, I_ij/P_ij multimedia indicators. Model fit: Omnibus χ2=566.7, p<0.001; −2LL=11129.838; Cox & Snell R2=0.219; Nagelkerke R2=0.259; Hosmer–Lemeshow χ2=6.744, p=0.601. Descriptive text statistics reported over 350k tokens and >40k unique terms; top 25 terms accounted for ~15% of tokens.

- Nine latent communication content functions were identified via BTM: Event (48.93%), People (14.40%), Period (10.22%), Report (3.96%), Trade (3.09%), Cancer/Health (2.80%), Sign (6.74%), Unite (1.59%), Glean (8.27%). Communications emphasized information provision (especially event/activity updates) with community-building content also present. - Without multimedia inclusion (baseline text-only): • Negative associations with retweets: Report (B=−1.753, p<0.001, Exp(B)=0.173); Event (B=−1.899, p<0.001, Exp(B)=0.150); Period (B=−0.320, p<0.001, Exp(B)=0.726). • Positive associations with retweets: People (B=0.579, p=0.005, Exp(B)=1.785); Unite (B=1.104, p<0.001, Exp(B)=3.016); Sign (B=1.158, p<0.001, Exp(B)=3.183); Glean (B=0.733, p<0.001, Exp(B)=2.082). • Non-significant: Trade (B=0.132, p=0.637), Cancer (B=0.051, p=0.406). - With multimedia inclusion: • Positive associations with retweets: Report (B=1.805, p<0.001, Exp(B)=6.078); Event (B=1.648, p<0.001, Exp(B)=5.195); Period (B=0.327, p<0.001, Exp(B)=1.387); People (B=0.565, p=0.006, Exp(B)=1.568). Main effect of multimedia inclusion was positive (B=0.291, p<0.001, Exp(B)=1.338). • Negative associations with retweets: Unite (B=−1.105, p<0.001, Exp(B)=0.331); Sign (B=−1.035, p<0.001, Exp(B)=0.355); Glean (B=−0.721, p<0.001, Exp(B)=0.486). • Non-significant: Trade (B=−0.180, p=0.524), Cancer (B=−0.059, p=0.340). - Interpretation: Information-oriented functions (event, report, period, people) drive higher engagement when paired with multimedia, whereas action-oriented appeals (unite, sign, glean) garner more retransmission when shared without multimedia.

The findings address the research questions by showing that IGOs predominantly create information-focused posts (especially event/activity updates) and that the function of a post interacts with multimedia inclusion to shape engagement outcomes. Information functions—reporting outcomes, event updates, time/period information, and recognizing people—benefit from multimedia elements, suggesting that visuals aid comprehension and sharing of informational content. Conversely, action-oriented calls (unite, sign, glean) perform better without multimedia, indicating that straightforward, text-based calls to action may reduce cognitive load and increase clarity and urgency for retransmission. These results suggest social media, including Twitter, can serve not only community-building and awareness (outside strategies) but also inside strategies of sharing policy-relevant information in “bite-sized” formats to influence debates and lobbying efforts. The study refines understanding of how specific content functions, rather than broad categories alone, drive behavioral engagement and highlights the importance of aligning form (multimedia) with function to maximize reach.

This study identifies nine communication content functions used by IGOs on Twitter and demonstrates that their effects on engagement depend on multimedia inclusion. Event, report, period, and people functions increase engagement when accompanied by multimedia, whereas unite, sign, and glean perform better without multimedia. Practically, communicators should match content function to appropriate media format (e.g., infographics for reports/events; text-forward appeals for petitions/support). Theoretically, the work extends non-profit and service literature by moving beyond broad message categories to sub-functions and by evidencing the moderating role of multimedia on engagement. Future research should examine multiple platforms and incorporate broader engagement metrics (likes, comments, sentiments) to generalize and deepen insights.

- Platform scope: Only Twitter was analyzed; IGOs operate across multiple platforms with different interaction mechanics. - Engagement metric: Retweets were the sole outcome; other behaviors (likes, comments, sentiment/emoticons) were not included. - Generalizability: Sample limited to English-language posts from EU-registered NGOs; findings may not generalize to other regions or languages. - Content nuance: While topic modeling captured latent functions, finer-grained linguistic and visual analyses were not conducted. - Time frame: Data from 2017–2019 may not reflect subsequent platform or behavioral changes.