Aggressiveness among candidates and gender-based political violence mark Brazil's 2024 municipal elections

By: FGV Comunicação Rio

By: FGV Comunicação Rio

The 2024 Brazilian municipal elections took center stage within the Mídia e Democracia project’s initiatives during September and October, reflecting the election’s significance as a barometer of the national political landscape and the growing role of digital media in shaping public debate on issues such as hate speech, misinformation, artificial intelligence, and distrust in democratic institutions. Over this period, 10 studies and 12 misinformation fact-checking initiatives were conducted, alongside meetings of the Media and Democracy Council.

The studies addressed various aspects of the elections, ranging from candidacies and campaigns in strategic capitals such as Rio de Janeiro, São Paulo, and Fortaleza to cross-cutting themes with direct influence on the electoral process. Topics investigated included toxicity in televised debates among candidates, the impact of the climate agenda on the electoral sphere, gender-based political violence targeting female candidates, and the strategies employed by candidates to increase engagement and expand their reach on social media.

This policy paper begins with the section “Analysis Lines: Period Overview,” presenting findings from three studies that focused, respectively, on the level of toxicity in the speeches of candidates in São Paulo, the influence of climate change on the electoral agenda of candidates in Porto Alegre following the May 2024 floods, and a mapping of gender-based political violence targeting female candidates in different cities.

These findings, alongside the in-depth analysis conducted for these studies, enabled the identification of potential action lines, outlined in the section “Action Lines: Thematic Directions.” Finally, the “Retrospective” section highlights the main activities undertaken by the project between September and October 2024.

 

  1. Analysis Lines: The Period Overview

The 2024 municipal elections were examined from various perspectives within the scope of the Mídia e Democracia project. Key topics included: gender-based political violence specifically targeting female candidates; the digital reach of candidates and their primary strategies for gaining visibility on social media; the toxicity in candidates' speeches during televised debates among competitors for the São Paulo mayoral race; the influence of the climate agenda on campaigns for the Porto Alegre mayoral race following severe floods that inundated southern Brazil in May; and the promotion of electoral-themed posts by candidates in Meta’s Ads Library, among other focal points.

It is worth noting that the U.S. presidential elections, which resulted in the re-election of Republican candidate Donald Trump, were also included in the project’s scope—both in studies and fact-checking initiatives. Monitoring reports revealed that Brazilian YouTube users exhibit strong polarization in their opinions about Kamala Harris and Donald Trump. The high toxicity of comments, particularly those targeting Kamala Harris's candidacy, highlighted the radicalization, intolerance, and aggression present in discussions surrounding the elections. 

1.1 An Analysis of Toxicity Among Candidates for the São Paulo Mayoral Race

Methodology

The study investigated the level of toxicity in the statements made by participants in the electoral debates for the São Paulo mayoral race. To do so, all electoral debates broadcast on free-to-air television during the first round of the election were selected, except for those on TV Globo. The debates included were from Band, TV Cultura, RedeTV, SBT, and Record. To analyze the selected debates, an artificial intelligence tool from OpenAI was used to transcribe the audio from videos collected on the YouTube channels of the networks responsible for broadcasting the debates. This tool also identified the speakers in one of the debates, attributing them to candidates or moderators based on segmentation conducted by researchers.

For each statement, corresponding to a candidate's complete speech (question, answer, rebuttal, counter-rebuttal, opening statement, or closing statement), an instruction rule was created—also using OpenAI—that enabled the identification of speech segments across four levels of toxicity, as well as the emotional appeal employed by the candidates.

From this initial classification, a sample of 50 speech segments from the candidates and the moderator in the TV Cultura debate was reclassified by researchers and incorporated into the model to improve results. Based on this information, the model classified all five selected debates, identifying statements as toxic or non-toxic, determining the level of toxicity, and assessing the emotional appeal involved in the statements. To verify the reliability of the classification, all speech segments from the Band debate were reclassified by researchers in a blind test, achieving an accuracy rate of 89.4% in toxicity classification compared to the model’s results. Using this data, both quantitative and qualitative analyses were conducted, considering the model's classifications, the content of the candidates' statements, and the context of these statements, which was examined through video analysis of the debates and the broader political context of the election.

Analysis

To observe the variation in toxicity during the debates and among the candidates, only the statements of the six candidates participating in the selected electoral debates were analyzed, excluding statements made by moderators, journalists, announcers, and external guests. The statements were evaluated based on the level of toxicity classified by the model, which can be detailed as follows: 1) Non-Toxic; 2) Mild Toxicity; 3) Moderate Toxicity; and 4) High Toxicity. Level 1 includes statements that maintain a consistent level of civility, without aggressiveness, insults, or a markedly offensive tone. Level 2 comprises statements that, while potentially critical, sarcastic, or ironic, do not directly offend opponents or exhibit a strongly aggressive dimension. Level 3 encompasses statements containing accusations and veiled or explicit insults that may trigger negative emotional reactions, without reaching an extreme level. Finally, Level 4 includes statements with direct insults and aggressive, disrespectful language, intentionally aimed at provoking, humiliating, or attacking the opponent.

Evolution of toxicity throughout the debates

Total number of statements classified: 363 statements | Total percentage of toxic statements: 51.2%

 

 

grafico

Source: YouTube | Prepared by: FGV Communications Rio

Analyzing the electoral debates individually, it is possible to observe how the percentage of toxic statements among the candidates fluctuated during the first round of the elections. The debates on RedeTV, TV Cultura, and Band had the highest levels of toxic statements, with 62.9%, 57.8%, and 57.3% of statements falling within the toxicity spectrum, respectively. These were also the three debates that recorded statements with high toxicity (level 4). Notably, the two debates with the highest toxicity—TV Cultura and RedeTV—were marked by significant physical and verbal altercations among participants, including Datena's chair strike on Pablo Marçal and the heated argument between Ricardo Nunes and Marçal. Interestingly, in the RedeTV debate, the candidates' chairs were fixed to the floor beforehand to prevent them from being used as weapons. Additionally, physical altercations between members of Marçal and Nunes' teams were reported during the Flow Group debate.

Following these two instances of aggression, the next debate recorded the lowest percentage of toxic statements, at 31.7%, indicating a softening of the candidates' rhetoric after these incidents. This more moderate tone carried into the Record debate: although it still recorded 48.5% toxic statements, it had a percentage of moderately toxic statements (17%) and no highly toxic statements, similar to SBT and lower than other debates. This analysis highlights differences among the debates throughout the electoral race and points to a reduction in political toxicity following an initial period of heightened tensions and acts of aggression among participants.

1.2 Debate on climate crisis among candidates for Mayor of Porto Alegre

Methodology

The report examined the online political discourse of candidates for the mayor of Porto Alegre, the capital of the state of Rio Grande do Sul, Brazil, following the severe floods that struck the region in May 2024. We analyzed the topics addressed by the candidates since the beginning of the current year, specifically observing the debate surrounding climate-related issues, considering the recent flooding that affected the city. Using both quantitative and qualitative analysis methods, we analyzed Instagram posts from four mayoral candidates from January 1 to August 31, 2024. The researchers used a social listening tool, FanPage Karma, to collect data from Instagram. A total of 2,149 posts were collected from the four leading candidates: the current mayor, Sebastião Melo (MDB), Felipe Camozzato (Novo), Juliana Brizola (PDT), and Maria do Rosário (PT).

Analysis

For the quantitative analysis, two main methods were used to interpret the posts: (i) topic modeling, a technique that allows us to combine words found in Instagram comments with their likelihood of association with broader topics, and (ii) volume analysis, comparing the number of posts related to climate with that of general posts. For each quantitative analysis, qualitative checks and data interpretation were carried out.

Main topics discussed by candidates for Mayor of Porto Alegre

 

grafico

Data source: Instagram Prepared by: Democracy Reporting International

As shown in a previous analysis by DRI, municipal elections represent a crucial opportunity for candidates to consolidate or regain influence and support in their regions. National politics, the economy, and issues such as public health and education significantly impact local elections in Brazil.

The graph above presents the main topics and subtopics discussed by the mayoral candidates in Porto Alegre. The analysis shows that posts about the floods were predominant, totaling 883 posts (41%), followed by various other topics, such as debates on national politics (423 posts, 19.7%) and municipal policies on specific issues (352 posts, 16.3%). Other posts covered various subjects (491 posts, 23%).

When debating "floods and climate-related issues," the candidates most frequently posted content with: (i) messages of solidarity, writing about hope for a better future and a prosperous, rebuilt city; (ii) flood alerts, advising people in affected areas to leave their homes; (iii) comments on the federal government's actions to support the municipal government; (iv) messages directly attributing the disaster to climate change; and (v) posts related to public safety and shelters available for relocated individuals.

The candidates also focused on national and municipal politics in their Instagram posts. While posts generally supported or criticized the current federal government on national issues, local discussions were more specific, covering topics like public education, women's rights, urban mobility, and public health.

The "various" category includes posts about specific holidays (Mother's Day, Carnival, Labor Day, etc.), personal posts about the candidates' private lives, and posts highlighting their decision to run for mayor in the municipal elections.

When observing the distribution of these topics in recent months, we saw that posts related to climate reached a peak after the first floods hit Porto Alegre, as expected. The data also indicates that candidates prioritized the debate on environmental issues and flooding over other topics in the months immediately following the floods.

1.3 Political gender violence in municipal elections

Methodology

In this research, the focus was on comments directed at mayoral candidates in YouTube videos to understand the nature of user interactions during the electoral race. Our main hypothesis, following previous studies such as those conducted by MonitorA – an observatory of political violence on social media developed by the organizations InternetLab, Revista Azmina, and Volt DataLab – was that female candidacies would receive more comments and offensive attacks than male candidacies, which would tend to receive comments merely disagreeing with their opinions.

Additionally, the level of aggression in the comments could be higher depending on other social markers of the candidates, such as belonging to diverse gender identity groups, political spectrum, and region. The research analyzed comments on YouTube videos of debates, interviews, and Q&A sessions with candidates who represent gender and racial diversity in the race for mayor in Brazilian municipalities, between the pre-campaign and campaign periods.

Analysis

To understand how gender-based political violence occurs on social media, the analysis was conducted by classifying the comments into 7 categories: i) support; ii) disagreement; iii) dehumanization; iv) infantilization; v) sexualization/fetishization; vi) gender-based misinformation; and vii) misinformation.

Category mobilized by gender

grafico

Source: YouTube | Prepared by: FGV Direito Rio 

As detailed in the presentation of this report, one of the hypotheses proposed, based on previous studies, was that female candidacies receive more aggressive comments, while male candidates receive comments of mere disagreement. The graph above confirms this hypothesis by showing the difference between the volume of comments with content of mere disagreement for men (with an average of 64 comments) and the volume of comments directed at female candidates classified into more harmful categories, such as infantilization and dehumanization (a total of 1,569 comments with this type of content for women) – for which no incidence of such discourse was identified for men. Male candidates stood out in comments with infantilization due to the high volume of ironic and mocking comments. This volume may be derived from typical behavior among social media users, using jokes and memes to ridicule and infantilize candidates, creating derogatory nicknames through puns. In this sense, Eduardo Paes received a large volume of comments of this type (37.2%).

  1. Categories by political spectrum

Regarding the differences between political fields, the graph below showed that right-wing female candidates consistently received more support and fewer harmful comments compared to left-wing female candidates.

Category mobilized by political spectrum

 

grafico

Source: YouTube | Prepared by: FGV Direito Rio

 

Only 14.9% of support messages were directed at left-wing female candidates, while in the right-wing camp, 90.8% of comments directed at female candidates were supportive.

Left-wing female candidates were the target of 96.9% of the attacks. Duda Salabert (PDT-MG), the first trans woman running for a municipal election in a state capital, and Tabata Amaral stood out as the primary targets of violent content in the sample, representing 67.4% and 19% of the attacks, respectively. Together, they were the target of 86% of the recorded aggression.

Of the total comments directed at left-wing female candidates, 53.4% were dehumanizing, while for right-wing female candidates, comments aiming to dehumanize accounted for 1%.

Among right-wing candidates, the one who suffered the most attacks was Clarissa Tércio (PP-PE), with about 1.3% of the identified harmful content directed at her. 

2. Action Lines: Thematic Guidelines

The analyses conducted this semester focused on key themes for strengthening Brazilian democracy and highlight the urgency of improving regulation in the country's political and electoral landscape. The studies show that recurring challenges in the national political spectrum are also reflected in the electoral context, pointing to the need to overcome obstacles related to both candidates' behavior and voters' participation. Moreover, there is a highlighted urgency to reach a consensus on the role of digital platforms in amplifying these issues, reinforcing the need for regulation that acknowledges the position these corporations occupy in mediating dialogue between voters and candidates.

The evidence produced in the analyses, along with the reflections from the Council's periodic meetings, suggest pathways and inspirations for the formulation of public policies and advocacy actions aimed at building an electoral environment based on values such as transparency and auditability. These guidelines should guide democratic electoral competition, enabling a return to dialogue between different ideological perspectives and overcoming the hostile, toxic, and defamatory nature that political conversation has assumed in the country.

Facing a challenge of this magnitude requires collaboration among various actors committed to consolidating democracy, promoting network actions that encourage institutional innovations implemented permanently, with elections serving as moments for evaluation and the peak of these initiatives. In this context, coordination between civil society, Big Techs, and governments, especially through the TSE and TREs, is essential to strengthening a robust democratic structure that is adapted to current challenges.

Among the necessary actions, the improvement of public policies stands out, with the creation of institutional mechanisms that guide and influence the behavior of candidates, voters, and digital platforms on crucial issues such as climate change and political violence. Strengthening technical devices and network architecture is also essential, focused on the integrated development of human and algorithmic moderation, aiming to remove toxic and violent content and identify illegal practices by voters and candidates. Additionally, advocacy and educational actions promoted by civil society should be expanded to pressure candidates and governments and conduct campaigns that encourage engagement in climate issues and strengthen informational skills, enabling the population to identify, report, and combat the spread of harmful content.

Finally, transparency and access to data circulating on digital platforms—by researchers, journalists, and electoral observers—remain one of the most urgent demands to be implemented by companies that hold the monopoly on message circulation. The provision of APIs is a basic transparency criterion that must be immediately incorporated by Big Techs.

Thus, the following lines of action are proposed:

2.1 Public policies

2.1.1  Toxic discourses: Develop conceptual and objective definitions of toxicity that are agreed upon between candidacies and the TSE and regional TREs, as well as establish penalties for the use of such content by candidates. Taking the events of 2024 as an example, promote campaigns and a public pact against political violence in elections. 

2.1.2 Gender-based Violence in the Electoral and Digital Context: Implement a robust action—similar to what was done regarding the use of AI in elections—that informs about penalties and specific reporting protocols for gender-based violence. Increase public awareness about the direct reporting line for gender-based violence, the formation of specialized electoral justice teams that can actively search for such events, expand support actions for female candidates who are victims of violence, and reduce costs associated with criminal/legal processes that may need to be triggered.

2.1.3 2. Public Funding Linked to Climate Commitments: The Superior Electoral Court (TSE) could establish that part of the resources from the Special Campaign Finance Fund (FEFC) be allocated primarily to campaigns and parties that include explicit environmental policies aligned with the sustainability demands of the municipalities/states in which they are running for election. This could help ensure that environmental issues are discussed and prioritized in elections, while also discouraging superficial/opportunistic promises to address climate problems.

2.2 Digital platform companies

2.2.1 Provide access to APIs and user-friendly interfaces, so that ordinary users, research institutions, journalists, and electoral observers can access and audit the messages circulating on platforms, enabling both data analysis and the ability to report, minimizing the circulation of such content on social media as much as possible. 

2.2.2 Create an "electoral observatory" funded by the platforms and composed of various representatives from civil society, government, and electoral campaigns, so that content monitoring is done in real-time, with the right to recommend modifications to the platform's architecture during the ongoing electoral process. 

2.2.3 In partnership with civil society and the TSE, develop, before the electoral period, protocols for the classification of toxic content and gender-based violence contextualized with the Brazilian electoral and cultural reality. These protocols should enable the detection and training of algorithms to ensure the removal of such content, preventing the widespread dissemination of harmful messages and promoting a safer and more democratic electoral environment.

2.3 Civil society organizations

2.3.1 Promote campaigns encouraging candidates to publicly commit to climate issues in their government plans, such as signing letters and manifestos. Additionally, advocate for campaigns to incorporate, in a manner contextualized to local realities, specific climate-related content in the political and electoral communication throughout the electoral process. 

2.3.2 Create a toxicity index where organizations monitor and provide real-time information on the toxicity level of candidates, pressuring them to modify their discourse. The initiative could involve an online code of conduct—identified by a seal or a filter on profile pictures—where users express their intention to engage in respectful and constructive political conversation with people of different ideological views.

2.3.3 Fund existing organizations and networks dedicated to combating gender-based violence to carry out collaborative monitoring focused on identifying and reporting cases of political violence during the electoral period. The funding could include a platform for collecting reports, with examples that define various types of violence. Additionally, campaigns could be conducted at local levels, such as in schools and neighborhoods, to raise voter awareness about what constitutes gender-based violence and its impacts on women's political participation in the electoral process. 

3. Retrospective

Monitoring

Between September and October, 10 weekly reports were produced with a focus on electoral processes. In addition to issues related to the Brazilian municipal elections, the U.S. presidential elections were also analyzed within the scope of the project, revealing important intersections between both elections, such as gender-based political violence aimed at female candidates. All the monitoring and activities of the project can be followed on the website.

Fact-Checking

Based on social media monitoring, Lupa conducted 12 fact-checks on disinformative content circulating on platforms. In the context of the elections, attacks and lies against candidates were prominent, including false accusations of attacks between political opponents.

Reflecting the well-known gender-based political violence in the electoral context, notable fact-checks included the circulation of deep nudes of São Paulo mayoral candidates Tabata Amaral and Marina Helena. In the context of the U.S. presidential elections, false and transphobic claims were spread, stating that Vice President Kamala Harris and former First Lady Michelle Obama, two cisgender women, were hiding their alleged trans identities. There was also a fact-check regarding dozens of transphobic posts by elected councilor Lucas Pavanato (PL-SP), who received the highest number of votes in São Paulo city.

 

Media and Democracy Council

The Council remains a space for fostering debates between civil society organizations, experts, and political authorities. During the period in question, specifically on October 18, the Council hosted the Director of the Federal Police, Andrei Rodrigues, who discussed relevant issues related to Public Security in the electoral context, highlighting the efforts of the corporation to maintain democratic integrity in the municipal elections.