Arceneaux, P., Anderson, J., Lukito, J., Shah, M., & Kiousis, S. (2026). Social bots as agenda-builders: Evaluating the impact of algorithmic amplification on organizational messaging. Journal of Public Relations Research, 1–34. https://doi.org/10.1080/1062726x.2025.2606676
Summary
This study reconceptualizes social bots as agenda-builders — non-human political communicators that compete with organizations, press, and the public for influence over public discourse. Drawing on political public relations (PPR) and agenda-building theory, the authors analyze over 935,000 tweets from the 2022 Ohio midterm elections to test whether algorithmically amplified bot accounts can transfer object, attribute, and network salience to human actors. They find that bots were particularly effective at influencing campaign messaging — especially at the second level, driving negative sentiment — while the press remained largely insulated. The paper argues that classical agenda-building theory, built on human-centric assumptions, must be revised to accommodate machine actors operating within human-computer information ecosystems.
Key Contributions
- First empirical public relations assessment of how social bots interfere with organizational news and strategic issues management.
- Extends agenda-building theory by adding non-human, algorithmic actors as a fourth communicator category alongside organizations, press, and public.
- Frames social bots as a medium-specific information subsidy unique to online environments.
- Offers a mechanism for why agenda-building fails in digital ecosystems, tied to bot-driven information disorder.
- Applies structuration theory to online ecosystems, treating algorithms and platform affordances as governance structures for machine actors.
- Draws out practical implications for campaigns weighing ROI on paid and earned media amid algorithmic competitors.
Methods
- Automated content analysis of 935,021 tweets collected via X’s API (May 4–November 8, 2022).
- Sampled 32 candidate campaign accounts (Gubernatorial, Senate, House), 47 Ohio newspaper accounts, and public users identified through 24 election-related keywords/hashtags.
- Bot classification via Tweetbotornot (0.5 threshold), yielding 2,064 bot accounts (9,141 tweets).
- 17 manually built keyword dictionaries (333 indicators) covering issues, stakeholders, campaign rhetoric, and Ohio cities, validated by a subject-matter expert.
- Sentiment coded with LIWC-22 (positive/negative tone).
- Granger causality models tested directional influence across time series; QAP assessed third-level network salience, disaggregated by race type and party.
Findings
- Bots transferred object salience to campaigns on four issues but had limited first-level influence on press (one object) and public (one object).
- Bots were the strongest second-level agenda-builders, driving both positive and negative sentiment — effects roughly doubled for negative tone.
- The public exerted the strongest first-level influence on the bot agenda (leading across eight objects), suggesting bottom-up agenda-building.
- Senate and Democratic campaigns influenced bot discourse more than Gubernatorial, House, or Republican campaigns.
- Republican campaigns were marginally more susceptible to bot agenda-building (six objects) than Democratic campaigns (three objects).
- Issue networks across all sources were highly correlated at every monthly time point, indicating shared third-level network agendas.
- The press neither influenced nor was influenced by bots (aside from one campaigning-related exchange).
Connections
This work sits within the broader literature on automated political influence and bot-driven amplification, connecting to studies of bot behavior and impact such as DeVerna2025-dl, Minici2024-tf, and Yang2025-iv. Its focus on how coordinated, algorithmically amplified accounts shape discourse relates to agenda-setting and cross-actor influence work like Rohrbach2026-rc and Kulichkina2026-zk, while its framing of bot activity as a vector of information disorder links it to the disorder-oriented scholarship represented by Starbird2025-jj and Marwick2025-ov.
Podcast
A research-radio episode discusses this paper: 🎧 MP3 · Spotify · Apple Podcasts