Nobody buys the answer.

Simple assumes nothing

The output of a prediction market is a probability, and it is usually more accurate than polls, pundits, or a company's internal forecast. Nobody pays for it.

The fees come from people who want action, not from people who want the number. So the market lives on gambling demand and investor money, and the accuracy is a side effect nobody has ever bought.

Compare it to a credit rating. Someone pays for that, which is the entire reason the rating industry exists. Prediction markets have the better product and no buyer. Finding the buyer is a research problem rather than a sales problem, because the market probably has to be built differently to produce something anyone would pay for.

Moderate assumes you know what a market is

The demand side is empty because the product is generic. A public probability on a public event is a public good. Whoever funds it cannot capture it, and everyone else reads the price for free. That is the structural reason there is no revenue, and better marketing does not touch it.

Two escapes exist. Make the signal excludable, which means private or delayed markets where the price itself is the product you sell. Financial data has worked this way for decades. Or make the signal decision-relevant to a specific principal who will pay to have their own uncertainty reduced. An insurer pricing a novel risk. A government evaluating a programme. A company choosing between two launches.

The second escape is the interesting one, because the buyer has a decision, the decision has a cost of being wrong, and a market can be pointed at exactly that. It has also never been built and closed with a paying customer. Corporate prediction markets were tried at HP, Google, and Ford. They worked, and they were shut down anyway, which is a fact worth understanding before rebuilding them.

Technical state of the art and the gap

This is an information-goods failure. Prices are non-excludable and non-rival, so the private return to funding a market sits far below the social return, and the equilibrium provision is roughly what we observe, which is close to zero paid provision.

Hanson's decision markets are the canonical answer. Condition the contract on a decision and the principal pays, because the conditional price is an input they cannot get anywhere else. That runs straight into decision-selection bias, which means the payment channel and the mechanism defect are the same problem seen from two sides. Solving one without the other does not produce a customer.

The specific open quantity: willingness to pay for a conditional probability, as a function of the decision's stakes and the principal's prior uncertainty. A value-of-information calculation gives the upper bound in an afternoon. Nobody has measured the actual number for any real buyer, so "who pays for the signal" stays rhetorical rather than empirical.

One adjacent buyer is underexplored and already consuming market data. The CGV thesis frames prediction market prices as a grounding layer for AI systems, calibrated priors piped into model context, and RSS3's MCP server already does the piping. That buyer has budget and a decision loop.