Prediction markets are eating American Politics
We're all "monitoring the situation" now. Our new research shows that, in the clip economy, prediction market prices are replacing polls.
“All of the major candidates I know right now, running for office right now, are sending around their Kalshi numbers, not their poll numbers.”
–Sean Patrick Maloney, former chair of the Democratic Congressional Campaign
Our information environment is changing. Television, radio, newspapers, and polls used to be the bread and butter of American news during election season. Now we live in the so-called “clip economy”---where many people’s primary interface with American politics are the social-feed friendly clips and soundbites from podcasts and live commentary platforms.
“The Internet is real life” is how Erik Torenberg summed it up, while boosting the launch of a new a16z-backed live online tech show, Monitoring the Situation (MTS). “Politics are downstream of the internet now...”
Appearing on the show, Marc Andreessen expanded on how this new, fragmented, hyper-online ecosystem is changing elections, explaining how memes have become so dominant in politics and so fast to change that politics becomes chaotic: “by the time the election rolls around, whatever is the thing that we think is the thing that’s gonna tilt the election today is gonna be a hundred social media meme cycles old.”
Andreessen concluded: “Hence the need to monitor the situation.”
For decades, the polling number was the default quantitative input into political conversation—the main way to use data to “monitor the situation” with elections. Cable anchors led with it, op-ed writers cited it, campaigns lived and died by it.
But how does this play out today, in a much more fragmented media landscape for political debate? One where people don’t trust the news, don’t trust polls, and worry that AI is creating fake news?
Tarek Mansour, co-founder of Kalshi, has sought to publicly position prediction markets as the antidote to this distrust. In a recent press release, he declared: “More people are watching Kalshi’s forecasts than trading them, which says a lot: our data effectively complements news and polls... As misinformation grows more common, Kalshi offers accurate, unbiased data to help people better understand what’s going on in the world.”
At Free Systems, we’re working on building what we call political superintelligence. The first layer of political superintelligence is information—how we know what we need to know to make government work well. So when the information environment changes rapidly, we pay attention. As we move deeper into an AI-fueled clip economy centered around rapid meme cycles, are people shifting their media diet more towards prediction markets and less towards polls? And if they are, what will this mean for how we should structure our information environment going forward? That’s what we start to tackle in today’s post.
Prediction markets take over from polls on social media
To get a feel, we zeroed in on videos that specifically cite either polling data or prediction market data when discussing politics. This is obviously a narrow slice of the short-form video ecosystem, but it’s a particularly interesting one, because it’s the part that’s bringing hard data to the conversation. We wanted to know, are the data sources that creators are citing changing as prediction markets become more popular?
We began by collecting TikTok and YouTube videos matching over 50 search terms (e.g., platform names, generic phrases, poll-related keywords) which return tens of thousands of raw videos. We then used Whisper to generate the text for each video and ran LLM classifiers to filter down to only the videos that genuinely discussed U.S. politics. The final dataset contains roughly 8,000 videos across both platforms, with Youtube coverage stretching back to 2007 and TikTok from 2020, each with the transcripts, engagement metrics and content classifications. For our purposes in this post, we focus primarily on videos from 2023 to present, but we’ll be working with the broader data in our ongoing research.
Our main finding: prediction markets were once a small fraction of the relevant creator content, but the 2024 election was a sea change. Starting in the summer of 2024, prediction-market videos started to shoot up. By the time of the election in late 2024, prediction-market videos were far more numerous than polls-related videos.
And now, in 2026, prediction markets are now the dominant data source for creators making videos about politics who need to draw on probabilistic information.
When creators want to monitor the election situation, and want to bring data to the party, they’re increasingly turning to prediction markets—the always-on data source they can pull on, react to, and clip clip clip.
As we flagged, this is a relatively narrow slice of all videos online, so we shouldn’t over-interpret. The reach on these videos is not huge, in general, and during the 2024 election videos that referenced polls saw much more overall reach than the prediction-market videos, driven by one huge outlier (an MSNBC TikTok video celebrating the “Selzer Poll” bombshell…whoops).
In 2026, though, prediction-market videos are gaining more reach than poll videos in most months. And if I had to guess, I would predict that they will outperform polling videos in the lead-up to this November, given current trends.
The breadth and speed of prediction markets are a dream for monitoring the situation
Monitors of the situation need up to the minute, real-time information on a wide variety of political potentialities. But polls are expensive, so they have to focus on a few key things and measure them on a relatively slow cadence.
Prediction markets can list contracts on a much wider variety of things—not just how the American public feels about a particular election or person, but what will the Fed do, what will the military do, what will Congress do, and so forth. And they can provide up-to-the-second prices. For creators, clippers, and monitors of the situation, this breadth and speed comes in handy.
Of course, the presidential election is the 700 pound gorilla of American political events. Naturally, monitors of the situation are going to be particularly hungry for information on it, and our data confirms that the bulk of our videos cover the presidency.
But, we do see that the second-largest category is policy outcomes, including things like Fed rate decisions, regulatory outcomes, and other questions pollsters would be hard pressed to predict using polls. Down-ballot races round out the picture. Senate, House, party-control, and gubernatorial contracts together generated roughly 1,100 videos—a substantial share, and one likely to grow as the 2026 midterm cycle intensifies.
How creators are using prediction markets
Here are a few examples of recent content we found that cites prediction markets and helps to illustrate how it fits into the modern clip economy.
Monitoring the political drama
First, as we’ve argued, creators can use markets to speak to ongoing political dramas with hard probabilities in a way they couldn’t with polls. Here are three good examples.
(1) A group of analysts discusses the situation in the Strait of Hormuz, referencing prediction-market data.
(2) A newscast analyzes prospects for a US recession and stagflation using prediction-market odds.
(3) And here’s a fun one where a Thai creator tracks potential Fed rate moves using prediction-market data.
Tiktok failed to load.Enable 3rd party cookies or use another browser
Establishment media blending polls and prediction markets
In a different but related vein, major news outlets like CNN are now combining both polls and prediction markets when discussing upcoming elections. Here’s a great clip of Harry Enten unpacking what the prediction markets have to say about the cost of living and Trump’s popularity.
Degens trying to make money
And let’s not forget the degens. Sometimes when you’re monitoring the situation, you also like to make a little scratch monitoring the situation. In another common vein, here’s a prototypical degen talking about his Polymarket trading strategy in which he looks for signs of insider trading on Iran-related markets.
Tiktok failed to load.Enable 3rd party cookies or use another browser
The numbers that shape politics
For decades, the polling number was the default quantitative input into political conversation. Cable anchors led with it, op-ed writers cited it, campaigns lived and died by it. The creators driving a fast-growing share of political discourse on TikTok and YouTube are now reaching for something else. When they need a number to anchor a take, they increasingly pull a market price.
What does this all mean for the information layer and a potential better future? This shift could have real upsides. Prediction markets aggregate information that polls cannot—probabilistic views on specific policies, court rulings, and political events that never had a sampling frame to begin with. They update in real time. They carry financial skin in the game that partially disciplines them against wishful thinking. And as survey response rates collapse and pollsters struggle to detect AI-generated responses among real human respondents, markets offer a useful supplementary signal about political reality that polls alone increasingly cannot.
But markets also fail in different ways than polls do, and we should be attuned to these potential issues. A poll can be wrong because of sampling error, turnout modeling, or social desirability bias; a market can be wrong because of thin liquidity, coordinated manipulation, or self-reinforcing feedback loops when a price starts being treated as news, as I argued in a previous piece.
In a media environment where creators are pulling odds from Kalshi and Polymarket and narrating them to audiences that trust the market number more than any individual pollster, the political stakes of market integrity are much higher than they were when these platforms were a niche curiosity. The creator economy is now downstream of the market price.
That makes the governance questions we’ve written about elsewhere—liquidity standards, manipulation monitoring, disclosure rules for campaigns and senior political staff—more urgent, not less.
In our previous piece on Building the Truth Machine, we argued we should build towards a prediction market ecosystem in which the news reports on thick markets using manipulation-resistant prices; in which politically important markets are listed regularly with standardized and clear rules; and where platforms encourage liquidity in those markets through market-making incentives and agentic trading.
Building the truth machine
We are getting much better at predicting the future. AI forecasting systems are climbing leaderboards that were once the exclusive domain of elite human “superforecasters”—and they may soon surpass us at divining the trajectory of our messy, contingent world. Developments in AI dangle the tantalizing prospect that, some day, we might actually be able to…
We believe this is the path forward for a newly invigorated information environment that people can trust in a fractured world. As people monitor the situation one clip at a time, they should get the highest quality clips with the most nutrient-dense probabilities. We’re working on building the system for that, and we’ll be back with updates on it soon.
Disclosures: In addition to my appointments at Stanford GSB and the Hoover Institution, I receive consulting income as an advisor to a16z crypto and Forum AI. My writing is independent of this advising and I speak only on my own behalf.
 | A guest post by
|
But the issue with prediction markets in the US is that users skew heavily male, under 50, and non-white. They are not reflective of the US of the electorate, or at least those who voted in the 2024 general election.
Excited to hear more updates on this!
It would be great if more prediction markets on useful events start getting more liquidity. Maybe you can funnel trading fees from popular-but-useless markets to subsidize political/macro markets. Also many exchanges on blockchains have a variety of market making incentives that can be explored (especially perp DEXes that do a very good job of it). Alternatively, I am working on ways to do that by using counterfactual knowledge across markets - https://virajnadkarni.substack.com/p/two-uses-of-knowledge-in-society
Also would be interesting to consider how concepts like "futarchy" can be used to inform decision making for any organization/firm. That will turn insider trading that is considered toxic into something more positive sum. So it will be in the interest of the organization to subsidize any futarchy markets.