The Menu, The Movies, and The Midterms: How Cultural Taste Data Called the Iowa Senate Upset

NEW YORK, NY – November 09, 2026 — For decades, political forecasting has relied on a straightforward, if increasingly flawed, premise: if you want to know how people will vote, you ask them. But as response rates plummet and social desirability bias skews survey results, the traditional barometer of public sentiment is breaking. In the wake of the November 3 midterms, a new model has emerged, suggesting that the most accurate predictor of a community's political leaning might not be found in a telephone survey, but in its Netflix queues, favorite local restaurants, and preferred retail brands.

This paradigm shift was thrust into the spotlight today when an artificial intelligence forecasting tool named Undertow was crowned the winner of a Global Agentic Hackathon. The competition, which drew nearly 3,000 developers worldwide, was hosted by Qloo, a firm specializing in mapping consumer taste and cultural data.

While the hackathon was designed to push the boundaries of autonomous digital agents, the victorious application managed something far more tangible: it correctly predicted the year's biggest political upset. By blending public polling with localized cultural preference data, the winning platform foresaw Democrat Josh Turek's unexpected victory over Republican Ashley Hinson in the race for Iowa's open U.S. Senate seat—a contest that traditional models had confidently handed to the conservative candidate.

Beyond the Polls: When Lifestyle Predicts the Ballot

The Iowa Senate race, an open seat following the retirement of Republican Joni Ernst, was widely viewed as a bellwether for the 2026 midterms. Conventional polling aggregators narrowly favored Hinson, the former U.S. Representative for Iowa's 2nd congressional district. Turek, a state representative and former Paralympian, was considered a long shot in a state that has leaned heavily right in recent election cycles.

Undertow's initial, polling-only model agreed with the consensus, projecting a Hinson victory. But the tool's architect, engineer Abdal Hassani, introduced a novel variable. Leveraging an API that accesses over 600 million entities across entertainment, dining, travel, and fashion, Hassani built county-level cultural profiles for the state. His software analyzed how these specific regional tastes aligned with historical voting patterns from the 2020 elections, combining those localized estimates into a state-level forecast.

When the lifestyle metrics were injected into the algorithm, the prediction flipped. The cultural data indicated a subtle but decisive shift in the electorate's underlying sentiment, pointing toward a Turek win.

"I built Undertow with Qloo's taste data to test a simple question. Could what people watch, where they eat and the brands they like add something useful to an election forecast?" said Abdal Hassani, CEO at Undertow. "Iowa was where that extra context changed our pick."

The Iowa upset was not an isolated success. In a preliminary evaluation of 33 Senate races called by midday on November 4, the hybrid forecast correctly identified 32 winners. A control model relying exclusively on polling data managed 31. Both scorecards were verifiably published before Election Day, cementing the application's methodology as predictive rather than retrospective.

The $25,000 Bet on Agentic Culture

While political analysts are dissecting the implications for future campaigns, the technology sector is eyeing a different prize. The ability to accurately predict human behavior based on aggregated cultural consumption has massive commercial applications, a reality that became immediately apparent during the hackathon's final rounds.

On November 5, during a live broadcast of the finals on the This Week in Startups podcast, angel investor Jason Calacanis pledged a $25,000 direct investment in Hassani's project. The financial backing was not driven by a desire to upend the polling industry, but by the tool's clear potential to revolutionize corporate strategy.

During the judging panel, which featured Calacanis alongside entertainer Cedric the Entertainer, the conversation quickly pivoted from ballot boxes to boardrooms. If an algorithm can determine a zip code's political vulnerability based on its affinity for specific indie films or fast-casual dining chains, that same architecture can autonomously direct millions of dollars in corporate spending.

"Abdal got my attention with the Iowa call. What made me want to invest was how quickly I could see the idea working beyond elections," said Jason Calacanis. "If you're deciding where to spend an ad budget or which cities to put on a tour, understanding what people in those markets care about can change the decision."

This evolution—from passive data analysis to active, autonomous decision-making—is the frontier of "agentic" AI. As digital assistants become more integrated into business operations, their efficacy relies entirely on their understanding of human nuance.

"AI is learning to reason, plan, and act. The less we want to instruct our agents, the more they need to understand us," said Alex Elias, founder and CEO of the company hosting the hackathon. "Undertow shows how a developer can take that foundation into an entirely different field and make it measurable."

Psychographics 2.0: Intelligence Without Identity

The intersection of behavioral data and political targeting inevitably evokes the specter of Cambridge Analytica, the defunct consulting firm that weaponized millions of psychological profiles harvested without consent. In the post-GDPR era, the concept of "psychographic profiling" remains a regulatory minefield and a source of deep public anxiety.

However, the architecture underpinning this new wave of predictive modeling attempts to thread a delicate ethical needle. The underlying database powering Hassani's tool operates on a strict non-PII (Personally Identifiable Information) mandate. The system does not know who an individual voter is, nor does it ingest personal identifiers. Instead, it relies on pre-anonymized, stateless inference—analyzing the relationships between cultural entities rather than tracking human subjects.

It is a framework designed to deliver "intelligence without identity." By mapping the cultural DNA of a county rather than the browsing history of a specific resident, the technology circumvents direct privacy violations while still capturing the zeitgeist of a community.

Yet, even aggregated behavioral data raises complex questions about the future of civic engagement. Independent data privacy experts note that while anonymized regional profiling complies with current consumer protection laws, it still facilitates a form of hyper-targeted influence. If campaigns and corporations can map our collective subconscious through our consumption habits, they can craft messaging that exploits regional anxieties or desires with unprecedented precision. The line between understanding a community and manipulating it remains uncomfortably thin.

The New Barometer of Public Sentiment

The success of this hybrid forecasting model exposes a fundamental truth about modern society: our stated preferences are often aspirational, defensive, or guarded, but our revealed preferences—where we spend our time and money—rarely lie.

Traditional polling asks citizens to articulate their worldview to a stranger over the phone. Cultural taste analysis simply observes the world they have already built around themselves. As response rates to traditional surveys continue to dwindle, the reliance on alternative data sources will only accelerate.

The 2026 midterms may be remembered as the moment when the political and commercial spheres realized that the most accurate map of human intention is drawn not with survey questions, but with the mundane, everyday choices that define our culture. As artificial intelligence continues to untangle the complex web of human behavior, the industries that govern our lives are learning that they no longer need to ask us what we want; they simply need to watch what we consume.