Electoral polls carry real power to shape political contests
No coração de Brasília, o Tribunal Superior Eleitoral pausou um julgamento que vai muito além de uma pesquisa controversa: está em jogo a fronteira entre medir a opinião pública e moldá-la. A disputa em torno de uma sondagem da AtlasIntel sobre Flávio Bolsonaro levanta uma questão que toda democracia enfrenta — quando uma pergunta deixa de ser instrumento de escuta e passa a ser instrumento de persuasão? A resposta que o TSE construir definirá as regras do jogo eleitoral para todos os candidatos, não apenas para este.
- Uma pesquisa que mostrou queda na popularidade de Flávio Bolsonaro foi suspensa do ar após denúncias de que suas perguntas induziam os respondentes a associá-lo a escândalos financeiros antes de medir a intenção de voto.
- O relator Kássio Nunes Marques votou pela anulação, citando até uma entrevista em que o próprio CEO da AtlasIntel admitiu 'viés político' nas questões — um argumento que transformou a metodologia da empresa em evidência contra ela.
- A defesa rebateu que pesquisas reflexivas online são métodos legítimos e já validados, e alertou para o risco de dois pesos e duas medidas: a mesma metodologia foi usada para avaliar Lula após o carnaval, sem contestação.
- A ministra Estella Aranha pediu mais tempo para estudar o caso, mantendo a pesquisa suspensa e adiando uma decisão que, segundo o ministro Toffoli, precisa estabelecer padrões uniformes para todos os candidatos.
- O TSE se vê diante de um dilema institucional: agir com rapidez arrisca parecer parcial; agir com cautela prolonga a incerteza sobre o que separa pesquisa legítima de propaganda disfarçada de dado.
O Tribunal Superior Eleitoral adiou na terça-feira um julgamento que expôs uma tensão fundamental na regulação de pesquisas eleitorais no Brasil. A ministra Estella Aranha pediu mais tempo para analisar o caso envolvendo uma sondagem da AtlasIntel que mostrava queda no apoio a Flávio Bolsonaro — senador e pré-candidato à presidência — e que permanece suspensa de circulação pública.
A pesquisa foi divulgada em 19 de maio, dias após vir à tona um áudio em que Bolsonaro pedia dinheiro ao fundador do Banco Master para financiar um documentário sobre seu pai. A metodologia da AtlasIntel é singular: recruta participantes por anúncios online e os submete a questionários longos que estimulam reflexão sobre o cenário político antes de perguntar sobre intenção de voto. O relator Kássio Nunes Marques votou pela anulação, argumentando que essa estrutura cruzou a linha entre medir e influenciar — e reforçou seu argumento citando uma entrevista em que o CEO da empresa admitiu 'viés político' nas perguntas que ligavam Bolsonaro aos escândalos do Banco Master.
A defesa veio de dois lados. A advogada do Partido Liberal sustentou que a metodologia era projetada para conduzir os respondentes, pedindo regras mais claras para os institutos. Já o advogado da AtlasIntel argumentou que pesquisas reflexivas online são métodos já validados — e lembrou que a mesma abordagem foi usada para avaliar Lula após o carnaval, sem qualquer contestação, sugerindo que o tribunal deveria aplicar seus critérios de forma consistente.
O nó técnico está na ordem das perguntas: pedir que o respondente avalie o cenário político antes de declarar sua intenção de voto pode contaminar a resposta. Com o pedido de vista de Aranha, o julgamento foi suspenso — mas o adiamento também sinaliza que o TSE sabe que está construindo precedente. O ministro Dias Toffoli, empossado no mesmo dia, defendeu padrões uniformes para todos os candidatos. A decisão que virá determinará não apenas o destino desta pesquisa, mas onde o Brasil traça a fronteira entre ciência eleitoral e manipulação.
Brazil's Electoral Court hit pause on a consequential decision Tuesday when Minister Estella Aranha requested more time to study a case that has exposed a fundamental tension in how the country regulates political polling. At stake is whether a survey by AtlasIntel—showing a sharp drop in support for Flávio Bolsonaro, the senator and presidential hopeful—should be scrubbed from public view for allegedly manipulating respondents rather than measuring their genuine views.
The poll in question was released on May 19th, days after an audio recording surfaced in which Bolsonaro asked Daniel Vorcaro, founder of Banco Master, for money to finance a documentary about his father, the former president. AtlasIntel's survey showed Bolsonaro's popularity declining in the aftermath. The company's methodology is distinctive: it recruits respondents online through ads on websites and blogs across Brazil, then presents them with lengthy questionnaires designed to prompt reflection on the political landscape before asking about voting intention. The Court's rapporteur, Minister Kássio Nunes Marques, voted to invalidate the poll, arguing that its structure—which linked electoral preference to recent events involving Bolsonaro—crossed the line from measurement into influence.
Nunes Marques was blunt about the stakes. Electoral polls, he said, carry real power to shape political contests, which is precisely why they must follow strict rules. He pointed to an interview the AtlasIntel CEO gave to CNN, in which the executive acknowledged a "political bias" in questions that tied Bolsonaro to the financial scandals at Banco Master. The rapporteur saw this as evidence that the researchers had strayed beyond the legitimate bounds of opinion measurement into something closer to propaganda dressed up as data.
The defense of the poll came from two quarters. Maria Cláudia Bucchianeri, a lawyer for the Liberal Party and former Electoral Court minister, argued during oral arguments that the survey's methodology was unusual and designed to lead respondents—that in a tight electoral race, polls themselves become instruments of persuasion, and institutes need clearer guardrails. AtlasIntel's attorney countered that internet-based surveys with reflective questions are already validated as legitimate research. He pointed out that the same methodology had been used to measure public perception of President Luiz Inácio Lula da Silva after he was honored by a samba school's parade theme—suggesting the Court should apply its standards consistently across all candidates.
The technical dispute hinges on question order. There is nothing inherently wrong with asking respondents to evaluate the political landscape before or after asking about voting intention. But when the sequence is reversed—when evaluation comes first—researchers acknowledge the risk that earlier questions will contaminate answers about actual voting preference. AtlasIntel's approach invited respondents to reflect on political scenarios in a way that, critics argue, primed them to view Bolsonaro unfavorably before they were asked directly about supporting him.
With Aranha's request for additional time, the poll remains suspended from public websites. But the delay also signals that the Court recognizes it is writing rules that will apply far beyond this single case. Minister Dias Toffoli, who took office the same day as the hearing, emphasized that the Court needs to establish uniform standards that bind all candidates equally—a recognition that ad hoc decisions invite accusations of bias. Minister André Mendonça called Aranha's deliberation a sign of the Court's seriousness. The decision ahead will determine not just whether this poll stands or falls, but how Brazil's Electoral Court polices the boundary between legitimate research and electoral manipulation.
Notable Quotes
Electoral polls carry real power to shape political contests, which is why they must follow strict rules to avoid manipulating respondents.— Minister Kássio Nunes Marques
The Court needs to establish uniform standards that bind all candidates equally.— Minister Dias Toffoli
The Hearth Conversation Another angle on the story
Why does the order of questions in a poll matter so much here? Isn't it just a survey?
Because the order shapes what people think before they answer. If you ask someone to reflect on recent scandals involving a candidate, then ask if they'll vote for that candidate, you've already moved their thinking. The Court sees that as unfair influence.
But didn't AtlasIntel's own CEO admit there was political bias in the questions?
He acknowledged it in a CNN interview. That's what made Nunes Marques vote to invalidate the whole thing. Once the researchers admit the bias exists, it's harder to argue the poll was just neutral measurement.
The defense said the same methodology was used for Lula. Doesn't that cut both ways?
It does, which is why Toffoli said the Court needs uniform rules. If you invalidate this poll but allowed the same method for Lula, you look partisan. That's the real problem the Court is trying to avoid.
So what happens if Aranha votes to keep the poll suspended?
Then the Court has essentially said this methodology crosses a line. But they'll have to decide what counts as legitimate reflection versus manipulation—and apply that to every poll going forward.
Why did Nunes Marques bring back oral arguments after they were suspended for years?
It signals he wants more transparency and input. Lawyers for both sides got to make their case directly. It's a shift in how the Court operates, not just a procedural detail.