Predictive Markets: Publishers’ Risky Bet
Prediction markets and silicon sampling offer publishers a rich new source of stories, but as Axios found out, they come with a serious risk – violating reader trust.
Earlier this spring, Axios ran a maternal health story citing ‘findings’ that a majority of people trusted their own doctors and nurses. Whether this statement is true or not is beside the point. Why? Because these ‘findings’ did not come from a poll of real people, they were made up.
Fabricated.
When this was duly pointed out to Axios – a title I otherwise highly respect – the publisher sheepishly added an Editor’s note: “This story has been updated to note that Aaru is an AI simulation research firm.” This correction identified the problem while sidestepping the real issue.
Axios used findings from AI start-up Aaru, which calls its method “silicon sampling”: AI is used to simulate how people might answer survey questions. This removes the need to survey real people but deceptively retains the authority of polling language.
In a guest essay entitled This Is What Will Ruin Public Opinion Polling for Good, University of California, Berkeley, computer sciences professor Benjamin Recht and Digital Theory Lab director Leif Weatherby weighed in on the Axios debacle:
“Phone polling has become exponentially harder. Web polling is too uncertain. Silicon sampling removes the messy, costly part of asking people what they think. But this undermines the very idea of the opinion poll.” — Leif Weatherby & Benjamin Recht
Polling is imperfect, but it still begins with people being asked questions. Put simply, Axios published machine-made opinion where readers would expect evidence from real people.
Why this is dangerous
The issue goes directly to the heart of audience trust. Publishers have to know the difference between evidence and imitation, and so do their readers.
Weatherby and Recht add that “pure fictions are on the brink of being treated as scientific and political knowledge.” That should stop editors cold, because fabricated opinion becomes far more powerful once a trusted news brand carries it, especially one with the weight and reputation of Axios.
A company, campaign, or pressure group can now commission synthetic opinion, frame it as research, and use a publisher’s reputation to give it credence. The reading public is then not being informed about what people think, but nudged by a manufactured version of consensus — the sort of tactic any government “Nudge Unit” (Copyright HM Government) would recognise immediately as useful.
This is a brutal trust risk. If publishers carry simulated opinion as though it reflects real people, they do not just weaken one article; they make all evidence look suspect.
The danger is compounded by the fact that AI models are not neutral. They are trained on selected data, shaped by design choices and constrained by safety rules, which means their answers inevitably carry someone else’s view of the world — often Silicon Valley’s. Dr Seth Dobrin, former IBM Global Chief AI Officer and President of the Responsible AI Institute, bluntly calls it technological colonialism.
Prediction markets want the same thing
It is not just silicon sampling that poses a threat. Kalshi, Polymarket and others are moving into partnerships with major media companies, including CNBC, CNN, Fox News, AP, Substack and Dow Jones. Basically, prediction markets want the trust, reach and respectability of established media brands.
A prediction market price is a price created by people betting on an outcome. Once odds are presented as a guide to what is likely to happen, rather than as the output of a betting system, the publisher has done more than report on the market, it has helped sell it. This is an ethical overreach.
Speaking to Nieman Lab, Kate Knibbs, a senior writer at Wired covering the prediction market, says she’s on the lookout for “the first big journalist insider trading scandal.”
Bo Sacks, one of publishing’s best-known industry watchers, warns:
“Prediction markets are only the headline. The real story is that outside industries keep arriving at your door with money and ambition, hoping to rent the credibility you spent decades building.” — Bo Sacks
Bottom Line
The editorial standard here is not particularly complicated: Polling begins with people and if no real people were surveyed, the output is not public opinion, however the vendor chooses to frame it.
Similarly, betting markets reflect the aggregate of wagers placed, not an assessment of what is true or likely. If a company stands to benefit from being cited by a credible news brand, the nature of that relationship and the limitations of the method need to be visible to the reader.
Axios showed how quickly this standard can slip with its subsequent editorial correction looking more like one of embarrassment than a willingness to front up.
Going forward, publishers will be offered (much) more of this. After all, AI research firms, prediction platforms and data vendors understand that a trend or number become significantly more valuable once a respected media brand has carried it.
Publishers need to tread warily, and with their eyes wide open.