A 21-year-old built an anonymous polling operation with AI tools and no real polling track record. The episode shows how quickly fabricated election data can move through campaigns, media and political information systems.
A 21-year-old recent college graduate built a polling website by himself, using AI tools, under no real firm and no disclosed methodology. According to reporting from The Guardian, The Hill, and the Los Angeles Times, that alone was enough to move real political coverage, including a mayoral campaign’s own messaging.
The trail of who bit
Rahil Prakash has identified himself as the creator of Median Strategies, an anonymous site that published fabricated polls in races across California, Wisconsin, and Nevada. The Los Angeles Times reports the fake Median poll showed Mayor Karen Bass leading challenger Nithya Raman by nearly 12 percentage points in the L.A. mayoral race, a number Bass’s campaign promoted before deleting it once the operation was exposed. A local outlet, the California Post, also cited the same survey, according to multiple reports. Separately, a fabricated poll in the Nevada governor’s race was picked up by a polling aggregator on X. And in Wisconsin, Median’s numbers had Francesca Hong leading the Democratic gubernatorial primary by 20 points; she went on to lose that race by half a percentage point.
None of these were fringe outlets or anonymous accounts amplifying the numbers. They were a mayoral campaign, a political news outlet, and an aggregator whose entire function is supposed to be tracking legitimate polling.
The common thread: the number was one they wanted
Every instance above shares something; the poll being cited showed the side citing it, or the outcome expected, ahead. A campaign sharing a poll that shows it winning by double digits isn’t neutral data-sharing; it’s what campaigns do, amplify good news. The uncomfortable part isn’t that a campaign wanted to believe a favorable number. It’s that wanting to believe it seems to have stood in for actually checking where the number came from.
Prakash told the Guardian he built the site to test exactly that gap of whether fabricated polling data could enter and spread through the political information ecosystem without independent verification. By his own account to The Hill, he was “a bit shocked” at how fast it spread once it did.
Why “fooled” isn’t quite the same as “deceived” and why some experts are skeptical of his story
It’s worth separating a few distinct failures. A campaign sharing a number that flattered it is an unforced error nobody had to trick them into wanting good news. An outlet citing an anonymous, unverified pollster as if it were established is a different and arguably more serious failure: a basic reporting standard went unapplied. And an aggregator repeating a fake poll shows the same failure at the curation layer, where vetting is the entire value proposition.
It’s also worth noting that Prakash’s framing of this purely as a public-interest “social experiment” hasn’t gone unchallenged. The Hill reported that experts in the polling industry were skeptical of his stated motives, pointing out the range of political and financial incentives, including the rise of prediction markets, where bettors wager on election outcomes, that can motivate fake polling, even though Prakash denied placing any such bets himself. Analysts also noted that past fake pollsters generally haven’t looked as credible as Median did, whether because they skipped a website altogether or built one that looked visibly suspicious which may be part of why this one traveled as far as it did.
Why this points to a regulatory gap, not just a media-literacy one
Right now, nothing legally requires a self-styled polling outfit to disclose who runs it, what methodology it used, what sample size or margin of error applies, or any track record before its numbers can be published, cited by a news outlet, or promoted by a campaign. Campaign ads carry disclosure requirements; campaign finance carries reporting obligations. A poll, arguably one of the most consequential pieces of “data” circulating in a race, currently carries none.
That gap is what made this particular episode possible, and what makes it repeatable. A basic disclosure standard requiring any poll cited in political coverage to name a real methodology, sponsor, and sample source, similar to the transparency standards professional polling organizations like AAPOR already ask legitimate pollsters to follow voluntarily, would give outlets and aggregators something concrete to check before running with a number, and would make it harder for a campaign to credibly promote a poll with no attributable source behind it. Several state legislatures already regulate deceptive campaign robocalls and AI-generated deepfakes in elections; unverified, anonymous “polling” operations that can demonstrably move real coverage arguably deserve the same scrutiny before they spread, not after.
Prakash himself, notably, framed his own findings as a warning rather than a boast: he told The Hill he hoped the episode serves as “a really good lesson” about what a single person could do and what a more serious, coordinated operation might accomplish next.
