Quaest trains AI agents to simulate Brazilian voter profiles

Quaest is training AI agents on Brazilian polling and social-network data to simulate voter profiles and test reactions to campaigns, products and speeches.

Wire notes

  • Quaest completes 10 years in 2026 and is developing digital agents trained on survey data and social-network monitoring.
  • The agents can simulate Lula-aligned, Bolsonaro-aligned and independent political profiles.
  • The tool can test hypotheses and estimate reactions to campaigns, products and speeches.
  • A survey of 10,000 people identified nine groups; two changed behavior in the latest election.
  • Social-liberal voters moved from Lula to Flávio and back to the ptista candidate, while individual entrepreneurs switched from Lula to bolonarism.
Voices 2026: Felipe Nunes revela que Quaest desenvolve ferramenta de IA para monitorar redes e simular perfis políticos
PhotoVoices 2026: Felipe Nunes revela que Quaest desenvolve ferramenta de IA para monitorar redes e simular perfis políticos — Firmbee / pixabay

By James Calloway

Quaest, which completes 10 years in 2026, is training AI agents on survey data and social-network monitoring to simulate Lula-aligned, Bolsonaro-aligned and independent political profiles, CEO Felipe Nunes revealed at Voices 2026.

Nunes, a political scientist, said the digital agents can test hypotheses and estimate possible reactions to campaigns, products and speeches. The tool remains in development.

Quaest is training the AI to behave as if it represented Brazil’s population. Nunes said the independent profile interests him most because it is the least predictable and often produces answers that escape other groups’ certainties.

What can Quaest’s AI agents test?

According to Nunes, the agents combine Quaest’s polling data with monitoring of online networks. Their simulations cover political profiles from lulistas and bolonaristas through independent voters.

“Our view is that AI enhances researchers’ abilities,” Nunes said. “What we have done is focus on simulating human behavior. This allows us to understand how groups may potentially behave.”

Nunes’s approach adds simulation to the research rather than replacing research judgment. He said the independent case is the least predictable part of the simulation.

Takeaway: For campaign, product and messaging teams, the tool offers a way to test likely reactions among distinct voter groups.

Nunes presented the project during the “O Brasil além das bolhas” (“Brazil Beyond the Bubbles”) panel at Voices 2026. André Miranda, GLOBO’s executive editor, moderated the session.

Voices 2026 took place at the Centro de Convenções Prodigy by Wish in Rio de Janeiro. The event brought specialists and businesspeople together to discuss technology, innovation and recent behavioral changes affecting daily life.

Voices 2026 was produced by GLOBO with Valor Econômico, Pequenas Empresas & Grandes Negócios, TechTudo, CBN and Rádio Globo. Rio’s city government and Municipal Education Department sponsored the event, the Câmara Municipal do Rio supported it, and Play9 partnered.

What does Brazil’s voting split reveal?

Nunes used the recent October elections as context. He said Brazil should not be treated as a country divided in two, even though close contests between candidates from opposite camps can suggest otherwise. His reading is of a country with many nuances and contradictions that cannot be reduced to the old left-right division.

Brazil’s electoral system was designed to produce two rounds, Nunes said. In theory, voters can support anyone in the first round, but a runoff forces them to choose the candidate they consider least bad. Because every election has reached a runoff since 2002, Nunes said, many people behave as though only the second round exists and the first never happened.

Nunes said rivalry between two major parties is healthy and part of democratic competition. The warning sign comes when disagreement shifts from ideas to social coexistence. He calls the latter affective polarization: treating an opponent as an enemy rather than an adversary and making coexistence, tolerance and acceptance difficult.

Nunes said affective polarization tends to produce discrimination, prejudice, intolerance and violence.

Takeaway: The danger is not partisan disagreement itself, but the point at which opponents become social enemies.

How do nine groups differ?

Nunes, the author of O Brasil no espelho, identified nine large groups of Brazilians from a survey of 10,000 respondents. He differentiated them by values, attitudes and preferences rather than placing everyone inside a single left-right frame.

Two groups changed their behavior in the last election:

  • Social liberals: Associated with the former PSDB electorate, they voted for Lula in 2022, moved to Flávio in the 2026 first round and returned to the ptista, or Workers’ Party, candidate in the runoff.
  • Individual entrepreneurs: These low-income workers see themselves as part of the middle class. They moved in the opposite direction, switching from Lula to bolonarism and keeping that choice through the end of the contest.

Nunes said the shifts show that Brazilian identities are more multifaceted than a two-camp model suggests. He called any claim to know Brazil arrogant because the country is continental and undergoing transformation. Brazil, he said, is explained not by the average but by variance, standard deviation and plurality.

Takeaway: The electorate is moving in layered ways that two opposing camps cannot fully describe.

How do algorithms intensify the split?

Nunes said virtual communities reinforce real-life bubbles. Mechanisms identify users’ preferences and offer content increasingly aligned with their convictions. The result, in his description, is a mirror that returns each user a flattering image.

Nunes put it plainly: “The principle of the algorithm is to make us think we are better than we are. It works on our narcissism.”

Nunes said research institutes cannot understand Brazil by asking questions and counting answers alone. What people declare can differ from how they behave. He cited racism: almost no one admits practicing discriminatory behavior, yet practically everyone says they know someone who commits racial prejudice.

The gap between public statements and daily conduct can reveal where society stands, Nunes said.

Takeaway: The hardest signal for researchers and marketers is the distance between what people say and what they do.

Marketers will judge the tool by one measure: whether its simulated audiences can anticipate real responses without smoothing away Brazil’s independents.

via oglobo.com (Source)

Filed under

Share this article:

More from James Calloway

James Calloway

Show full bio

Market editor covering technology and product news in marketing, advertising, marketing technology, media, SEO and search at Marketing Herald.

384 articles

More on the wire

« Previous article