Social bots as agenda-builders: Evaluating the impact of algorithmic amplification on organizational messaging

Summary

This paper asks whether social bots, through algorithmic amplification, can serve as agenda-builders that compete with organizations, press, and the public in shaping political discourse. Using the 2022 Ohio midterm elections as a case, the authors analyze over 935,000 tweets across campaigns, newspapers, the public, and bot-classified accounts to test first- (object), second- (attribute), and third-level (network) agenda-building transfers. They find that bots most powerfully shape attribute salience—particularly negative sentiment—in campaign messaging, while having little measurable effect on the press. The core theoretical move is to argue that classical agenda-building theory, premised on human actors, must be revised to incorporate machine actors as a fourth category of political communicator in human-computer information ecosystems.

Key Contributions

• Extends agenda-building theory by positing social bots as a fourth category of communicator alongside organizations, press, and public.
• Provides the first public relations–focused empirical assessment of how bots interfere with news management and strategic issues management.
• Reframes social bots as a medium-specific information subsidy unique to online environments.
• Offers an explanatory mechanism for failure of organizational agenda-building, tying it to bot-driven information disorder.
• Connects structuration theory to algorithmic platforms, treating platform affordances as governance structures for machine actors.
• Draws out practical implications about ROI on paid/earned media when algorithmic competitors are active in the ecosystem.

Methods

Automated content analysis of 935,021 tweets collected via X’s API (May–November 2022) from 32 campaign accounts (Gubernatorial, Senate, House), 47 Ohio newspaper accounts, public users surfaced via 24 election keywords/hashtags, and 2,064 bot-classified accounts identified via Tweetbotornot at a 0.5 threshold. Issue coding used 17 expert-validated keyword dictionaries (333 indicators); sentiment was scored with LIWC-22. Granger causality models tested directional time-series influence among the four actor types, while QAP assessed third-level network salience and issue co-occurrence, disaggregated by race type and party.

Findings

• Bots were the strongest second-level agenda-builders, driving both positive and negative sentiment in campaign messaging, with negative-tone effects roughly twice as strong.
• First-level (object) influence of bots was modest: four issues onto campaigns, only one each onto press and public.
• The public exerted the strongest first-level influence on bots (eight objects), suggesting a bottom-up dynamic in which bots amplify pre-existing public discourse.
• Senate and Democratic campaigns more strongly shaped bot discourse; Republican campaigns appeared marginally more susceptible to bot influence (six vs. three objects).
• Third-level network agendas were highly correlated across all sources at every monthly time point.
• The press was essentially insulated: no meaningful bidirectional object or attribute influence with bots, apart from one campaigning-related exchange.

Connections

This study sits at the intersection of bot/automation research and agenda-setting/agenda-building theory, complementing computational audits of bot influence such as DeVerna2025-dl and Bollenbacher2026-vz, and resonating with work on the limited persuasive reach of inauthentic accounts like Mosleh2024-op. Its concern with how algorithmic amplification reshapes information ecosystems connects to broader treatments of platform-mediated disorder in Starbird2025-jj and Budak2024-ef, while its focus on attribute and especially negative-tone transfer aligns with analyses of affective and partisan dynamics in Gonzalez-Bailon2024-rq.

Podcast

A research-radio episode discusses this paper: Listen