Battle over Meta’s alleged Bolsonaro bias intensifies days before Brazil presidential runoff

Five days after Brazil’s first presidential-election round, Lula’s campaign was unable to boost Instagram posts for almost 24 hours. His lawyers submitted the episode to the electoral court as possible evidence that Meta was interfering in the election. They pointed to the apparent contrast with Flávio Bolsonaro’s campaign, which could continue boosting posts.

A separate incident occurred on 23 September, when Meta disabled Lula’s campaign advertising account for more than an hour. Meta said an anti-fraud review mistakenly treated an advertisement as an attempt to impersonate a government social programme. The company called the incident accidental, not political.

The allegations gained force because many Lula supporters also reported removed content, reduced reach, reposting restrictions, and account blocks. A watchdog found mostly leftwing profiles among its cases. Meta denied favouritism, and a judge said there was no evidence of deliberate action, but ordered the company to show the problems were fixed.

Meta’s content-moderation system applies platform rules to posts, accounts, and advertisements. Its anti-fraud tools look for suspicious activity, including suspected bots, impersonation, or unusual advertising behaviour. These systems matter during elections because they can determine whether political messages remain visible and whether campaigns can pay to reach voters.

The article gives several examples of automated restrictions. Meta said it wrongly flagged the US No Kings account as bot traffic and disabled Lula’s advertising account after misclassifying an advertisement as government-programme impersonation. Meta also said a gap in its safeguards caused an AI assistant to treat requests involving Lula differently from requests involving Flávio.

Restrictions can remove posts, reduce their reach, prevent reposting, or block accounts and advertisements. Meta later said it introduced human review for Lula’s campaign. Its oversight board nevertheless cited recurring enforcement failures and urged more transparent, consistent election practices that protect political speech.

The scale of reported problems was substantial. CTRL+Z, a watchdog founded by a former Meta WhatsApp executive, catalogued 1,121 reports involving lost or suppressed reach, reposting restrictions, and account bans or blocks. The reports affected mostly leftwing profiles, though Meta dismissed the catalogue as anecdotes and users’ perceptions.

Meta’s financial stake was also large. Political advertising on its services earned the company £43m during this election. That was 85% more than it made during the last presidential election. The figure shows why access to Meta’s paid campaigning tools mattered to both candidates and why restrictions could become politically sensitive.

Google and TikTok were not allowing political advertising in this election. As a result, Meta’s services were the only large-scale channels for paid digital campaigning. The article does not establish that the reported cases changed the result, but their scale prompted complaints, court action, and calls for better scrutiny.

Brazil’s electoral court was involved because the alleged restrictions affected an election campaign. The court could examine complaints, require information from Meta, and impose instructions about equal treatment. The article does not describe every power the court possesses, but it records direct orders against the company.

The court ordered Meta not to give Lula and Bolsonaro “unequal treatment”. Justice Marco Aurélio Bellizze then gave Meta 24 hours to prove that the problems affecting Lula had been fixed and would not happen again. After he said Meta had failed to do so, he granted another 24 hours.

Meta responded by denying favouritism and saying it had introduced human review to prevent automated restrictions. Bellizze said he had found no evidence of deliberate action, which would require further investigation. The episode shows the court could demand corrective steps and explanations, but not immediately establish intentional interference.

Technical restrictions can create an electoral advantage by reducing one candidate’s ability to communicate while another campaign continues reaching voters. A blocked advertisement or lower distribution means fewer people may see a message, especially when campaigns rely on paid digital promotion during a short runoff period.

The article describes Lula’s inability to boost Instagram posts for almost 24 hours, while Flávio Bolsonaro could do so. It also reports that Lula’s advertising account was disabled for more than an hour. Meta called these incidents mistakes, but even temporary restrictions could matter when campaign messages compete for attention immediately before voting.

The imbalance was sharper because Google and TikTok were not allowing political advertising. Meta’s services were therefore the only large-scale channels for paid digital campaigning. The article does not prove that restrictions changed votes, but unequal access created a credible reason for complaints, court intervention, and demands for transparent enforcement.

Distinguishing deliberate bias from mistakes requires evidence about what happened inside Meta’s systems. Investigators would need decision logs, account histories, ad reviews, error reports, and comparisons showing how similar material from both campaigns was handled. They would also need to establish whether restrictions followed political affiliation rather than the same technical triggers.

The article highlights the central problem: only Meta has access to its internal data. External observers therefore cannot easily test whether Lula’s cases were isolated errors or part of a wider pattern. The oversight board found examples of over-enforcement, while Meta attributed them to mistaken anti-fraud classifications and a safeguards gap.

A larger, independently reviewable dataset could show whether leftwing accounts were restricted disproportionately and whether automated errors were corrected consistently. The article says the existing evidence is a collection of cases that might eventually support a conclusion, but that no one can be certain now because information is missing.

Private platforms have electoral influence because campaigns use them to distribute messages and buy access to voters. In this election, Meta’s services were the only large-scale channels for paid digital campaigning. That made decisions about visibility, advertising, and account access politically important, especially during the final days before the runoff.

The same systems are designed to prevent fraud, bots, impersonation, and other abuse. Yet Meta said those safeguards wrongly flagged the No Kings account as bot traffic and misclassified Lula’s advertisement. The company also acknowledged a safeguards gap involving AI artwork requests. Such errors can restrict political speech even without deliberate bias.

This creates a difficult balance. Platforms must enforce rules consistently while protecting lawful political expression, and outsiders need enough information to check their decisions. João Victor Archegas said external scrutiny was not real because Meta controlled the data. The oversight board called for enforcement that is transparent, consistent, and protective of political speech.