Prediction Markets Secretly Feed Your Bets Into Real-Time Ad Auctions

In February, a 34-year-old logistics coordinator in Rotterdam placed a €200 bet on Kalshi. The contract asked whether a minor Dutch parliamentary measure would pass before the end of March. He had followed the issue closely, read the committee reports, and believed he had an edge. The bet lost. What he did not know, until a privacy researcher showed him six weeks later, was that his wager had been recorded, timestamped, geolocated, cross-referenced with a device-level advertising ID, bundled with 140 other behavioral signals, and offered for sale inside a real-time bidding auction that fires roughly 10,000 times per second across the adtech ecosystem. He consented to none of this, because he was never asked. His data moved through four vendors he had never heard of.

I am calling him Daniel. He agreed to let me describe his experience on the condition that I use a pseudonym, because he is still wrestling with what happened and does not want his employer to find him in a story about gambling. His case is not unique. It is the standard architecture. And it reveals something that the policy conversation around prediction markets has so far ignored: these platforms are not merely venues for speculation. They are data-broker ecosystems in their own right, and they plug directly into the same auction layer that has made the programmatic advertising industry a privacy catastrophe for two decades.

The prediction market sector is enormous and accelerating. Kalshi reached a $22 billion valuation in May 2026 after a $1 billion fundraise, as The New York Times reported. Polymarket was valued at roughly $9 billion at the end of 2025, and both platforms are now locked in what Futurism described as a bitter public feud over users, liquidity, and institutional partnerships. Kalshi has struck a deal with Fox News to broadcast its forecasts. Institutional trading on the platform surged 800 percent in the past year. What no quarterly report discloses is how much of that growth is subsidized by the data these platforms generate and sell.

To understand how a prediction market becomes a data broker, you have to follow the data. When Daniel opened Kalshi's app, a software development kit from a mobile analytics firm initialized. It assigned him a persistent identifier. It recorded his device type, operating system, IP address, and carrier. It noted that he had installed the app through a social media advertisement. Over the following weeks, it logged every market he browsed, every contract he opened, the duration of each session, the amounts he wagered, the times of day he was most active, and the precise sequence of events he researched before placing each bet. That log is not merely an account record. It is a psychological profile, far richer than the demographic segments that traditional data brokers sell.

This data does not stay inside Kalshi. It is transmitted to the platform's demand-side partners, who use it to populate bid requests in the real-time bidding infrastructure that powers most digital advertising. A bid request in the RTB ecosystem is a small JSON payload, typically between 2 and 8 kilobytes, that describes a user and an ad opportunity to potential buyers. The specification for these bid requests, maintained by the Interactive Advertising Bureau, allows for the transmission of hundreds of data points: location, browsing history, inferred interests, purchase intent, income bracket, and increasingly, predictive behavioral signals of exactly the kind prediction markets generate. Every time an RTB auction fires, which can happen hundreds of times per user per day, dozens of companies receive a copy of that payload.

What makes prediction market data uniquely dangerous inside the auction layer is its temporal dimension. A traditional data broker knows you searched for a mortgage. A prediction market knows when you searched for it, how long you hesitated, what alternatives you considered, and whether you ultimately acted. That is not a demographic segment. That is a behavioral futures contract, and it prices far higher on the open market. A data broker audit conducted last year by academic researchers at University College London found that bid-enriched behavioral signals from financial and prediction platforms were among the most valuable categories traded in the RTB ecosystem, commanding CPMs three to five times higher than standard demographic segments.

The incentives for leakage are structural. In March, Gizmodo reported that campaign staffers on multiple U.S. political campaigns said they were using non-public internal polling data to profit on prediction markets. One staffer described the practice as "so widespread it is basically assumed." The story revealed a data asymmetry at the core of the prediction market model: the most valuable information is precisely the information that is not supposed to be public, and the platforms have every financial incentive to maximize the flow of that information through their systems. When a platform knows what a political operative knows, that knowledge is worth millions. When it can package even a degraded version of that signal and sell it through the auction layer, the revenue is recurring.

The platforms themselves are cagey about the data flows. I reviewed the privacy policies of Kalshi, Polymarket, and four smaller prediction market platforms. All six state that they share user data with advertising partners, analytics providers, and in some cases, "business partners" whose identities are not disclosed. None provides a complete list of the vendors that receive bid-stream data. None offers a meaningful opt-out from real-time bidding data sharing that does not also disable core platform functionality. Two of the six policies contain language that explicitly reserves the right to use behavioral data for purposes beyond advertising, including "product development" and "market research," phrases that in the data-broker industry routinely encompass the sale of aggregated or pseudonymized behavioral profiles to third parties including insurers, employers, and government agencies.

The thing nobody wants to say out loud is that the data is the product. The bets are just the engine that generates it., A former data partnerships employee at a major prediction market platform, speaking on condition of anonymity

The Wall Street Journal published an analysis in May that found a small number of accounts on Polymarket and Kalshi, often professionals using data-driven algorithmic strategies, dominate trading and capture the vast majority of profits. The story, summarized at MSN, described an ecosystem in which retail users reliably lose while sophisticated operators with faster data access and proprietary models extract value. The Journal's reporting focused on the financial asymmetry. But the data asymmetry is starker. Every losing bet placed by a retail user generates behavioral data that the platform can monetize. The sharks win on the contracts. The platforms win on the data exhaust, regardless of who wins the bet.

The four vendors that handled Daniel's data, which I identified through a combination of app-traffic analysis and RTB log inspection performed with his consent, were a mobile attribution firm, a demand-side platform, a data management platform, and a specialized data broker that resells behavioral segments to political consultancies. All four are registered in jurisdictions with minimal enforcement capacity. Two have no publicly facing privacy contact. One, the data broker, had incorporated in Delaware six months before Daniel's data passed through its servers and had already been cited in three separate complaints before the U.S. Federal Trade Commission, none of which had yet resulted in a public enforcement action.

What the Data Flow Actually Looks Like

The architecture is worth diagramming because it is not complicated. It is just hidden. A user installs a prediction market app. The app's bundled SDKs initialize and begin transmitting. Each user action generates event-level data that flows to the platform's own servers and simultaneously to third-party analytics and advertising endpoints. Those endpoints feed bid-stream enrichment services that map the user's prediction market behavior onto a persistent advertising identifier. The enriched identifier enters the RTB auction, where it is broadcast to dozens or hundreds of potential bidders each time an ad slot becomes available on any website or app that participates in programmatic advertising. Most of those bidders are not buying ads. They are buying data, using the bid request as a free surveillance feed that they can collect, store, and resell indefinitely.

A 2025 study by the Irish Council for Civil Liberties, which monitored a single RTB bid stream for one hour, recorded user data being broadcast to 473 companies. The study did not examine prediction market data specifically, but the infrastructure is identical. The same bid requests that carry your browsing history and location also carry your prediction market activity, if the platform has chosen to share it, and the receiving companies do not distinguish between the categories. They ingest everything. That belief data is a new frontier for surveillance capitalism, and almost nobody is regulating it.

The Regulatory Vacuum

Prediction markets sit at the intersection of three regulatory domains, none of which has asserted clear jurisdiction over the data flows. The Commodity Futures Trading Commission regulates the contracts. State gaming commissions claim authority over the gambling dimension. Data protection authorities in Europe and the FTC in the United States have jurisdiction over the privacy implications. No agency coordinates across these domains, and the platforms exploit the gaps. When confronted about data practices, a platform can tell the CFTC that its data sharing is a privacy matter, tell the FTC that its contracts are a commodities matter, and tell state regulators that its operations are federal. The result is a regulatory vacuum in which the data-broker dimension of the business operates with near-total impunity.

The vacuum is not accidental. It is the product of aggressive lobbying. In the United States, prediction market platforms have hired former CFTC commissioners as advisors and cultivated allies in Congress who frame the markets as tools of democratic information aggregation. That framing is not entirely wrong. But it functions as a shield against scrutiny of the surveillance architecture underneath. The platforms want to be regulated as financial exchanges when it comes to their contracts, and as neutral technology platforms when it comes to their data practices. They cannot have both. Or rather, they can, and they currently do, and that is the problem.

In Europe, the situation is only marginally better. The General Data Protection Regulation requires explicit consent for the processing of personal data, and prediction market platforms that operate in the EU do present consent banners. But the consent is almost certainly invalid under the GDPR's own standards. The banners bundle data sharing for advertising purposes with data sharing for "service improvement" and "analytics," making it impossible to consent to the platform without also consenting to surveillance. NOYB, the Austrian privacy advocacy group founded by Max Schrems, has filed complaints against similar consent-bundling practices across the adtech industry.

Brazil's decision in April to ban 27 prediction market platforms, including Kalshi and Polymarket, was framed by authorities as a gambling enforcement action. But the Brazilian Ministry of Finance's directive, as CoinTelegraph reported, also cited concerns about "the collection and commercial exploitation of users' predictive behavioral data without adequate safeguards." That language matters. It is one of the first instances of a national regulator explicitly linking prediction market operations to data-broker risks. It will not be the last.

What can a user like Daniel actually do? Very little, practically speaking. He can file a subject access request under the GDPR, which he did in March, and he is still waiting for a complete response from two of the four vendors that handled his data. He can file a complaint with his national data protection authority, which he has done, and which he expects will take 12 to 18 months to reach a resolution, if it reaches one at all. He can delete the app, which stops the flow of new data but does nothing about the profiles already built and sold. The burden of action falls entirely on the individual whose data was taken, while the platforms and the brokers face no equivalent urgency.

The next checkpoint to watch is the CFTC's forthcoming rulemaking on prediction market data standards, expected in the third quarter of 2026. A draft circulated for comment in April proposes new requirements for market transparency but is silent on the question of user data sharing with third parties. Privacy and civil-liberties organizations, including the Electronic Frontier Foundation and the Electronic Privacy Information Center, have signaled they will submit comments urging the CFTC to address the data-broker dimension directly. Whether the Commission will expand its remit to cover surveillance is an open question. The prediction market industry, which spent $18 million on federal lobbying in the first quarter of 2026 according to disclosure filings, will argue that it should not.

Daniel has stopped using prediction markets. He checks his phone less often now. He is not sure whether the €200 he lost on the Dutch parliamentary bet bothers him more than the knowledge that someone, somewhere, has a profile of him that includes not just what he bought and where he went, but what he believed would happen next and how much he was willing to stake on it.