Judge suspends poll showing Bolsonaro's son losing ground after controversial audio

The audio came after, on a separate page, after all voting questions were locked in.
AtlasIntel's defense against claims that survey respondents were shown damaging recordings before answering about Bolsonaro.

No Brasil, onde a confiança nas instituições eleitorais é um bem frágil e disputado, um juiz do Tribunal Superior Eleitoral suspendeu a divulgação de uma pesquisa que mostrava Flávio Bolsonaro perdendo terreno para Lula às vésperas de 2026. A decisão, concedida a pedido da defesa do pré-candidato, levanta uma questão que transcende o caso concreto: até onde pode chegar o poder judicial de silenciar dados que o público já conhece? O episódio revela como a batalha eleitoral brasileira se trava cada vez mais nos tribunais, onde metodologia e timing se tornam armas tão poderosas quanto votos.

  • Uma pesquisa da AtlasIntel mostrando queda no apoio a Flávio Bolsonaro foi suspensa por ordem judicial antes mesmo de ser amplamente distribuída — mas depois de já ter circulado.
  • A defesa de Bolsonaro alega que gravações comprometedoras foram exibidas aos entrevistados antes das perguntas de intenção de voto, contaminando os resultados; a AtlasIntel nega categoricamente a sequência.
  • Ministros do TSE, em conversas reservadas, criticam a decisão por considerá-la tecnicamente frágil e potencialmente lesiva à credibilidade das pesquisas eleitorais no país.
  • A AtlasIntel tem dois dias para apresentar documentação técnica, após o que o plenário do TSE decidirá em definitivo — mantendo a suspensão como uma medida provisória e contestada.
  • A posição do ministro Dias Toffoli, que se declarou impedido em casos ligados ao banco Master citado nas gravações, permanece incerta e pode ser decisiva no julgamento final.

Um juiz do Tribunal Superior Eleitoral suspendeu a divulgação de uma pesquisa da AtlasIntel que registrava queda no apoio a Flávio Bolsonaro e crescimento da vantagem do presidente Lula. A decisão do ministro Nunes Marques atendeu a um pedido da defesa do pré-candidato, que questionou a metodologia do levantamento realizado em maio — justamente após o portal Intercept Brasil publicar gravações de Bolsonaro pressionando o banqueiro Daniel Vorcaro, dono do banco Master, a financiar um filme biográfico sobre seu pai, intitulado Dark Horse.

O argumento central da defesa era técnico: os advogados queriam saber se o áudio comprometedor havia sido reproduzido aos entrevistados antes das perguntas sobre intenção de voto, o que teria artificialmente deprimido os números do candidato. A AtlasIntel respondeu que a gravação só foi apresentada após o encerramento completo do questionário principal, por meio de uma ferramenta separada chamada Atlas VRC, sem possibilidade de revisão das respostas anteriores. A empresa também apontou que outras pesquisas chegaram a conclusões semelhantes, sugerindo que a queda refletia uma mudança real na opinião pública.

A ordem de suspensão, no entanto, gerou tensão nos bastidores do tribunal. Ministros que preferiram não se identificar criticaram a medida, argumentando que bloquear a circulação de dados já públicos carece de sentido prático e que qualquer restrição a pesquisas deveria ser respaldada por análise técnica independente. A preocupação mais ampla é com o precedente: num país onde a desconfiança em relação às pesquisas já é elevada, decisões judiciais desse tipo podem aprofundar o ceticismo sobre a integridade de qualquer levantamento eleitoral.

O caso ainda não está encerrado. A AtlasIntel tem prazo para apresentar documentação técnica, o Ministério Público Eleitoral terá espaço para se manifestar, e o plenário do TSE dará a palavra final. Paira também a incógnita sobre o voto do ministro Dias Toffoli, que se declarou impedido em processos ligados ao banco Master e ainda não sinalizou sua posição neste caso específico. O desfecho dirá não apenas o destino desta pesquisa, mas o quanto o Judiciário está disposto a intervir no fluxo de informação eleitoral no Brasil.

A federal judge in Brazil's electoral court has frozen a political poll that showed Flávio Bolsonaro, the former president's son and a leading candidate for the 2026 presidency, losing ground to incumbent Luiz Inácio Lula da Silva. The decision came after Bolsonaro's legal team challenged the survey's methodology, arguing that voters had been manipulated into giving negative responses about their candidate.

The poll in question was conducted by AtlasIntel in May, shortly after the news outlet Intercept Brasil published audio recordings of Bolsonaro pressuring a former banker named Daniel Vorcaro to finance a biographical film about his father. The recordings were damaging—they showed Bolsonaro demanding money from Vorcaro, who owns the Master bank, for a movie project called Dark Horse. The timing mattered. AtlasIntel's subsequent survey found Bolsonaro's support slipping while Lula's advantage widened, a result that Bolsonaro's camp immediately disputed.

On Monday, June 8th, Judge Nunes Marques issued a temporary order suspending the poll's release, distribution, and promotion across all of AtlasIntel's official channels. In his written decision, Nunes Marques concluded that the survey showed signs of methodological contamination—that the questionnaire had been designed or administered in ways that pushed respondents toward negative views of Bolsonaro. He specifically cited the public disclosure of the audio recordings as evidence that the polling environment had been poisoned, making it impossible to measure genuine voter sentiment.

Bolsonaro's lawyers had raised a specific technical objection: they wanted to know whether the controversial audio had been played to survey respondents before they were asked about their voting intentions. If it had been, they argued, it would have artificially depressed support for their candidate. AtlasIntel responded that the audio was only played after the main questionnaire had been completely finished, with no opportunity for respondents to go back and change their answers. The firm said participants were redirected to a separate page to record their reactions while listening to the recording through a tool called Atlas VRC. AtlasIntel also noted that other polling firms had found similar results, suggesting the decline in Bolsonaro's numbers reflected genuine shifts in voter preference, not methodological tricks.

The judge gave AtlasIntel two days to submit additional technical documentation defending its methodology. After that, Brazil's Electoral Public Ministry will have one day to respond. The full merits of the case will then go before the TSE's full panel for a final ruling, meaning Nunes Marques's suspension order is not the last word.

Behind the scenes, however, the decision is generating significant friction among the electoral court's other ministers. Several of them, speaking on condition of anonymity, have criticized Nunes Marques's move. They argue that granting an emergency suspension after the poll had already circulated publicly makes little sense, and that any order restricting the spread of polling data should be accompanied by independent technical analysis, not just the judge's say-so. Some ministers have also expressed worry that suspending polls could damage public confidence in electoral surveys more broadly—a concern that carries weight in a country where skepticism about polling integrity is already high. One minister noted that such a decision could feed existing doubts about whether any survey can be trusted, potentially poisoning the well for all future polling.

There is also uncertainty about how Dias Toffoli, a recently appointed justice who took over a seat vacated by Cármen Lúcia, will vote. Toffoli was mentioned in the Master bank scandal himself and has already recused himself from related cases at Brazil's Supreme Court, saying he needs time to study the matter. His position on this poll suspension remains unclear.

AtlasIntel has maintained that its work was conducted with full technical rigor and scientific precision, guided by impartiality and accuracy. The firm's defense rests on the timeline: the audio came after the voting questions, so it could not have influenced those answers. Whether the TSE's full panel agrees will determine whether this poll ever sees the light of day, and whether judicial intervention in polling becomes a new front in Brazil's increasingly contentious electoral politics.

The poll was conducted with full technical rigor and scientific precision, guided by impartiality and accuracy.
— AtlasIntel, in a statement
The audio was only played after the main questionnaire had been completely finished, with no opportunity for respondents to go back and change their answers.
— AtlasIntel's defense
The Hearth Conversation Another angle on the story
Inventor

Why would a judge suspend a poll that's already been published? Doesn't that seem like closing the barn door after the horse has escaped?

Model

That's exactly what some of the other ministers are saying. But Nunes Marques seems to be treating it as a matter of principle—he wants to prevent further circulation and amplification of what he sees as a contaminated result. Once something's out there, though, you can't really un-ring that bell.

Inventor

What's the actual complaint? That the poll is wrong, or that it was conducted unfairly?

Model

The complaint is about fairness. Bolsonaro's team says voters were shown the damaging audio before answering questions about him, which would obviously poison their responses. AtlasIntel says no—the audio came after, on a separate page, after all the voting questions were locked in.

Inventor

And which version is true?

Model

That's what the full court will have to decide. But here's the tension: if you believe the audio contaminated the environment just by existing and being public, then maybe the timing of when it was shown in the survey doesn't matter. The damage was already done to the information space.

Inventor

So this is really about whether judges should be able to suppress polling data at all.

Model

Exactly. Some ministers worry that if courts start blocking polls they don't like, it erodes trust in all polling. Others think that's a risk worth taking if the alternative is letting manipulated data shape the election. It's a genuine collision between two legitimate concerns.

Inventor

What happens next?

Model

AtlasIntel has two days to defend itself technically. Then the Electoral Public Ministry responds. Then the full panel votes on whether the suspension stands. The judge's order is temporary, but it sets a precedent either way.

Contact Us FAQ