Luís Inácio Lula da Silva’s Oct. 25 election win for an unprecedented fourth, non-consecutive term as Brazil’s president was a surprising come-from-behind victory, as he defeated Flávio Bolsonaro by 4.4 million votes. Contrast that to his loss in the first round by 2.3 million votes, the first presidential candidate since Brazil’s return to democracy to register such a turnaround. And he scored that reversal against the son of former president Jair Bolsonaro, who, while serving a 27-year sentence for plotting to overturn the 2022 election, nevertheless remains popular across Brazil.
Beyond its historic attributes, Brazil’s runoff has something to teach other democracies, not only about its efficient electronic vote counting, though that alone is striking: with 128.2 million votes cast, official results came in only a couple of hours after polls closed. The bigger lesson is about online political mobilization. In the three weeks between rounds, the left and a broader coalition of pro-democracy forces mobilized at large scale in digital spaces that the far right has dominated for the past decade. At the center was “Lula’s Army,” a WhatsApp channel that connected 10.8 million supporters and mobilized them to use the algorithmic levers the right had long mastered: hashtags, search trends, coordinated sharing.
Online Mobilization
A campaign that had opened as a celebration of his legacy, and that Lula himself lamented as his “least organized” since 1989, changed course after the first-round loss and turned to the digital world and the streets. The candidate changed, too. Lula, who turned 81 two days after the runoff, got his first smartphone after the first round. He began recording selfie videos and going live, speaking directly to supporters in formats familiar to many of them but new to him. His first selfie video reportedly garnered more than 23.5 million views in just 24 hours.
The mobilization reached beyond the usual activists, influencers, artists, and politicians. It drew in fan communities and people with no prior political activism who came out against Flávio Bolsonaro, if not necessarily for Lula. It also drew in older voters, including those who, like the candidate himself, were learning to find a voice in this new online political arena.
Often what united them was less enthusiasm for Lula than memory of the first Bolsonaro government — its handling of the Covid-19 pandemic that killed more than 700,000 Brazilians, its attacks on democratic institutions that culminated in the coup plot, and allegations of corruption and involvement with organized-crime.
Much of that mobilization ran through a single WhatsApp channel, Exército do Lula (“Lula’s Army”), created and run by Congressman André Janones, himself a social-media influencer. By Oct. 23, two days before the runoff, the channel had passed 10 million followers. Followers received daily “missions,” extending the military metaphor, to share videos, make memes, push hashtags, and run Google searches on campaign-related terms to lift them in search trends. The channel also recruited content producers to contest narratives, answer disinformation, and spread the campaign’s message in plain, popular language. It trained followers to work the platforms’ algorithms: how to create viral content, spread it, and extend its reach. Just as important, it taught them not to engage with opponents’ content, since even hostile replies and sharing of posts signal engagement and feed the algorithm that amplifies it.
The channel’s reach quickly drew the attention of Flávio’s campaign, prompting it to petition the Superior Electoral Court (TSE), alleging Lula’s channel coordinated the spread of disinformation and offensive content against him. On Oct. 11, the court’s president, Justice Kassio Nunes Marques, ordered the channel to be shut down. Another judge from the same court overturned the decision the same day. The independent fact-checking news outlet Agência Lupa reviewed the 721 messages the channel published between Oct. 5 and 12 and found no disinformation among them. By contrast, Lupa found that the rival channel, “Tropa do Flávio” (“Flávio’s Troops”), created in response to “Lula’s Army,” had produced at least eight misleading posts, including fabricated images and statements by Lula taken out of context.
Fighting Big Tech
Beyond the courts, the mobilization had to contend with the platforms themselves, Meta in particular. The dispute with Meta involved two distinct dimensions: the algorithmic distribution of political content and the generation of campaign-related images through Meta AI. Lula’s campaign accused the company of restricting the reach of its posts and paid advertisements while disproportionately amplifying content associated with Bolsonaro. Lula’s campaign told the TSE that its posts and paid ads were being throttled (deliberately slowed) on Facebook and Instagram while content tied to Flávio was amplified. According to an analysis by Lula’s Workers Party (PT), Bolsonaro’s social media account generated, on average, 2.96 times more engagement than Lula’s, despite having 17.2 percent fewer followers. Separately, tests by Projeto Comprova, CNN Brasil, BBC News Brasil, and other news organizations found that Meta AI refused requests to generate campaign-related images featuring Lula while fulfilling comparable requests referring to Flávio Bolsonaro, even when only his first name was used. On Oct. 15, the electoral court ordered Meta to ensure equal treatment of the candidates in content distribution and AI-generated outputs, subject to a daily fine of R$500,000 for noncompliance. Meta attributed the reported inconsistencies to technical problems and denied deliberate interference or political bias.
Regardless of intent, the effect was real. The lesson is to remember that algorithms are not neutral. They reflect their designers’ choices, the data they are trained on and process, and the commercial incentives of the companies that deploy them. They shape what millions of voters see, read, and learn about, and, just as important, what they never encounter. Generative AI now “decides” what images a campaign can make through its tools. In an election decided by a few points, such asymmetries may make a difference.
Democracy depends not only on the right to vote, but also on citizens’ ability to form political judgments freely and participate in public debate on fair terms. When AI systems shape political visibility and expression through opaque decisions, they can affect the conditions under which democratic choices are made.
Brazil’s electoral court caught this one before the vote; most democracies still lack effective mechanisms to timely detect and address algorithmic asymmetries. Too often, voters and campaigns may discover such problems only after ballots have been cast, when the damage to democratic debate may already be irreversible.
Election Day
The mobilization was reflected at the polls. Turnout in Brazil usually falls between the first round and the runoff. This time, however, about 3 million more people voted in the runoff. But turnout alone does not explain Lula’s win. He turned a 2.2 million-vote deficit into a 4.4 million-vote margin, a 6.5 million-vote swing. Even if every additional voter had chosen Lula, that would account for less than half. The rest came from voters who chose other candidates in the first round, cast blank or spoiled ballots, and possibly from some who voted for Flávio.
An otherwise uneventful election day turned chaotic in its final hours. In several cities with large numbers of pro-Lula voters, public transportation ran well below required levels, leaving voters stranded or facing long waits. Voting is mandatory in Brazil (though not everyone complies), and electoral rules require free public transit on election day at weekday frequency. In Salvador in the country’s northeast, where the bus fleet was reportedly cut by more than half, the regional electoral court went as far as ordering the arrest of the city’s transportation secretary, who was in Paris. Both campaigns asked for voting time to be extended. Lula cited the transit shortfalls, while Flávio cited rain and flooding in the south and southeast, where he ran strongest in the first round. With the 5 p.m. close approaching, a Supreme Court Justice extended voting nationwide to 7 p.m., the first such extension in Brazilian history. But the decision became public only minutes after the scheduled close, and by the time word reached the country’s roughly 100,000 polling places, many had already shut down. A follow-up clarification limited the extension to polling places still open, still leaving officials, volunteers, and voters confused. In the end, only about 34,000 people voted during the extended hours, less than 1 percent of Lula’s 4.4 million-vote margin.
Reflections
AI researcher Joy Buolamwini uses the term “excoded” for people harmed by algorithmic systems or left at their margins. Extending this concept to the political sphere, Lula’s campaign was “excoded” when Meta’s AI would not generate its images, and, by the campaign’s account, its content reached fewer people. The lesson is how Lula’s Army found a path and worked around the code: its volunteers shared campaign content directly through WhatsApp and personal networks, recruited others to amplify it, and coordinated the sharing of videos and hashtags to increase visibility across social media. These efforts helped the campaign pierce information bubbles and reach audiences beyond those served by algorithmically curated feeds. That may help explain both the higher runoff turnout and the voters who changed their minds.
In refusing to remain excoded, the campaign’s supporters recovered a way of communicating that risks being eroded by the algorithm-driven public sphere, in the absence of adequate governance. As American philosopher John Dewey wrote in Democracy and Education, “a democracy is more than a form of government; it is primarily a mode of associated living, of conjoint communicated experience.” Brazilian educator and philosopher Paulo Freire offered a complementary insight, understanding dialogue not merely as an exchange of words, but as a form of collective learning, reflection, and action through which people can transform their world. Seen in this light, the achievement of Lula’s Army went beyond winning votes. It brought people together to communicate and reclaim human agency in that digital public sphere increasingly shaped by algorithms. Online mobilization also translated into offline activism, bringing people together for public demonstrations. In doing so, it helped revive the concept of a democratic society in which citizens are not passive consumers of algorithmically curated content, but participants in shaping their collective future. The second round of the 2026 Brazilian Elections was not only a vote to be won, but an association and shared experience of communicating in common.
These lessons also carry a warning. The tactics that may have helped Lula win this runoff are not fundamentally different from those the far right has used for years: centralized channels, coordinated action, and strategies to influence algorithmic visibility. The difference this time was who used them and why. But it’s also important that their use remains accountable to democratic principles. Lula’s Army answers to the one congressman, André Janones, who originated the idea, and it is likely to become a lasting instrument of political mobilization, as Lula has called on supporters to remain active after the election. Similar WhatsApp mobilization channels, inspired by Janones’ strategy, have emerged in France, Argentina, and Spain.
Lula has already signaled that digital mobilization will be central to his next term. The test is whether the left can keep a 10-million-strong army committed to accurate information now that the campaign is over, and whether it can resist the incentives of a polarized, algorithm-driven system that rewards outrage over truth, loyalty over scrutiny, and power over democratic norms.
