The prediction markets are just measures of people's willingness to bet on something. There's no underlying data.

For some bets, sure. Same for betting on the coin toss on the Super Bowl. Some of those prop bets are there purely for folks to take a flyer on.

But the same general principle that makes most sports betting super accurate, and makes up the efficient market hypothesis, also underlies a lot of the more substantive bets. If you give the public the opportunity to make money by having a better understanding of the likelihood of the future events by betting on/investing in them, then the price of those bets/investments will incorporate the information the public has in a meaningful way. It will also incentivize lots of people to learn relevant information about the events.

That's why betting lines for major sporting events are such amazing predictors of game outcomes, often far better than efforts to statistically model them. NFL games land pretty much at a near-exact 50/50 against the closing spreads, for example - showing that those betting lines are doing a very good job of accurately reflecting real world information.

I think you're criticizing a claim that even the prediction markets aren't making. If there's a prop on a truly random event (like the coin toss), no one's suggesting that the prediction market on that event is based on underlying data. But for the types of markets where there is underlying data, like "conventional" sports bets or elections predictions or what have you, then the market price can convey meaningful information that is aggregated from bettors about the real world.