Echoes of Automation: How Bots Shaped Political Discourse in Brazil

In an era where social media platforms are central to political communication, the activity of bots raises pressing concerns about amplification, manipulation, and misinformation. Drawing on more than 315 million tweets posted from August 2018 to June 2022, we examine behavioural patterns, sentiment dynamics, and the thematic focus of bot- versus human-generated content spanning the 2018 Brazilian presidential election and the lead-up to the 2022 contest. Our analysis shows that bots relied disproportionately on retweets and replies, with reply activity spiking after the 2018 election, suggesting tactics of conversational infiltration and amplification. Sentiment analysis indicates that bots maintained a narrower emotional tone, in contrast to humans, whose sentiment fluctuated more strongly with political events. Topic modelling further reveals bots’ repetitive, Bolsonaro-centric messaging, while human users engaged with a broader range of candidates, civic concerns, and personal reflections. These findings underscore bots’ role as amplifiers of narrow agendas and their potential to distort online political discourse.

💡 Research Summary

This paper presents a comprehensive analysis of the influence of social media bots on political discourse in Brazil, spanning from the 2018 presidential election to the prelude of the 2022 contest. Utilizing a massive dataset of over 315 million tweets, the study systematically compares behavioral patterns, sentiment dynamics, and thematic content between bot-generated and human-generated tweets.

The methodological approach is robust and multi-faceted. Bot accounts were identified using BotometerLite scores, with analysis conducted at three confidence thresholds (0.5, 0.7, 0.9) to assess the sensitivity of results. Text preprocessing filtered for Portuguese content and cleaned tweets for analysis. The study addressed three primary research questions through temporal analysis of tweet types (original, retweet, reply, quote), lexicon-based sentiment analysis using the Sentilex-PT02 dictionary, and topic modeling via Latent Dirichlet Allocation (LDA) applied to the intense campaign period of October 2018.

Key findings reveal stark contrasts between bot and human behavior. Bots disproportionately relied on retweets and replies, with a dramatic spike in reply activity immediately following the 2018 election—particularly at the strictest (0.9) detection threshold. This suggests a tactical shift towards infiltrating and potentially disrupting conversational threads to amplify messages. In contrast, human activity was consistently dominated by retweeting.

Sentiment analysis showed that human discourse exhibited greater emotional volatility, with sentiment scores fluctuating strongly in response to major political events. Bot sentiment, however, remained notably more stable and neutral across most thresholds, indicating a narrower, potentially scripted emotional range. Interestingly, at the highest (0.9) threshold, bots displayed more extreme sentiment spikes, hinting at a subset of highly specialized or provocative automated accounts.

The most pronounced difference emerged from topic modeling. Bot-generated discourse was overwhelmingly narrow and repetitive, heavily focused on amplifying pro-Bolsonaro messaging and keywords. Human discourse, conversely, engaged with a much broader and decentralized spectrum of topics, including discussions about multiple candidates (Haddad, Lula, Ciro Gomes, etc.), civic concerns like voting logistics and corruption, socio-economic issues, and personal reflections on democracy.

The study concludes that bots in the Brazilian political context functioned primarily as amplifiers of a narrow agenda rather than originators of novel content. Their strategic use of replies for conversational infiltration and their repetitive, candidate-centric messaging highlight their potential to distort online public discourse by reducing thematic diversity and creating artificial amplification. These findings underscore significant challenges for the integrity of digital public spheres and carry important implications for platform governance and policies aimed at mitigating automated manipulation.