Fishy Polls: Breaking Down the Median Strategies and TPSI Revelations
Last week was quite the week for the business of political polling.
First up, our good friends at The Guardian did what their American counterparts could not and were able to uncover the source of Median Strategies, the organization that released completely fabricated polls in three separate states. From their Friday exclusive:
A 21-year-old recent college graduate was behind an anonymous website that put out widely shared fake election polling in Wisconsin, Nevada and California, the Guardian can reveal.
The website, Median Strategies, abruptly shut down and withdrew its polling earlier this week following questions from a reporter from the Los Angeles Times. In an unsigned statement, it described a “short-term social experiment” to examine how fake polling “could enter and spread through the political information ecosystem without independent verification”.
“I wanted to see if fake polls could really penetrate the ecosystem that easily,” Rahil Prakash told the Guardian on Thursday when reached by phone. “And as it turned out, it could.”
Karen Bass, the mayor of Los Angeles, touted Median polling data after it appeared to put her ahead of her election challenger by more than 10 percentage points. The poll was also cited by the California Post. A widely followed polling aggregator on X also shared the group’s Nevada poll, although several credible polling organizations did not aggregate Median polling because they could not verify its methodology.
“Did I expect it to perhaps expedite in the way it did? Most definitely not,” said Prakash. “I did not expect the California political press to pick it up on their own.”
He added: “I’ll apologize straight out to the campaigns of Karen Bass, Nithya Raman and Francesca Hong for this.”
Raman is running against Bass in November’s LA mayoral election. Hong lost the Democratic primary for Wisconsin governor earlier this month, after Median suggested she had led the race by more than 20 percentage points. It also released a fake poll showing the Nevada governor, Joe Lombardo, up by five points before November’s general election.
In an interview, Prakash said he was motivated to start the fake polling project after a significant polling miss in the Michigan Senate primary, where polls had showed Abdul El-Sayed with a heavy lead over his opponent Haley Stevens before election day. In the event, El-Sayed wound up winning by a much smaller margin than polls had suggested.
“This is what inspired me to test my theory,” said Prakash. “I was a bit shocked, perhaps at how fast this sort of information can spread.”
As of Thursday morning, Median had about 20 followers on its X account. According to the Los Angeles Times, the account began posting on 9 August.
“I think my overall takeaway was that if you just make it look a bit pretty, it’s rather easy for it to spread,” Prakash said. “And I hope it gives a really good lesson going forward in that we need to be very serious about these things and very serious in how, if one person can do this, what can a serious operation do in the future – a real, you know, operation for an organization, country, whatever.”
TPSI admits that the firm changed the stated vote choice of certain respondents who favored a different candidate, current Lieutenant Governor Jay Collins, turning these respondents into supporters of James Fishback based on their belief that Fishback should have been polling higher. The report states:
Jay Collins was polling in the twenties in our data. That did not align with our read of the race, and we concluded it might reflect a data-quality problem specific to our text-message panel. Acting on that conclusion, we identified respondents who selected Collins but whose broader profile aligned with a competing coalition, and leaned those respondents toward James Fishback (emphasis ours).
This was the most consequential error in the project, and it was ours alone. … We reallocated real support away from a candidate (emphasis ours) on the basis of a suspicion that our own data contradicted. No survey methodology should permit a respondent’s stated preference to be reassigned because an analyst finds it implausible, and ours will not again.
This decision is an obvious violation of standard polling practices, which, as they themselves state, do not “permit a respondent’s stated preference to be reassigned because an analyst finds it implausible.” This decision is made all the more questionable by the undisclosed relationship between the pollster and the Fishback campaign.
Political polling falls under the category of "how is that still a thing?" It's done as pure filler between presidential elections. It measures vibes and nothing more. While it may have had value a half-century ago, that is clearly no longer the case. Technology has changed the game. Families no longer have home phones. Individuals can block unidentified numbers. While certain pollsters do allow for online polls, those themselves are inherently biased toward younger, more tech-savvy individuals, the exact same demographic that loves to overshare to complete strangers on the internet. To answer a call from an unknown number and to proceed to take the time to complete an automated survey requires an individual to have a certain vested interest in the poll's content. That is why we saw such a huge miss by Emerson in Michigan. That poll in particular failed to capture the steady support for Haley Stevens and instead skewed toward the "it" candidate of the moment, Abdul El-Sayed. Similarly in Wisconsin, other than the fraudulent Median Strategies poll, there was no true research after the last-minute candidate reshuffling on the Democratic side. A July 22-27 poll from Marquette Law School showed Francesca Hong at 38% of the vote with a massive 34% undecided. That pollster opted to end with that poll rather than run a final one the week before the election.
Polling has never claimed to be an exact science, but it shouldn't be even considered a science at all. Not in 2026, when we have fake polls, candidates manipulating polls, a generational gap in technology and access to taking the poll, and the introduction of prediction markets for the very first time. Like AI, prediction markets have crept up on us and remain largely unregulated. When you can place significant amounts of money on who will win elections, you can see how and why candidates like James Fishback might be tempted to overinflate their own numbers. While we don't yet have a specific instance of a campaign manipulating the betting markets, there remains credible concern that we will see an increase in misinformation throughout the fall that would greatly impact many House and Senate races, all of which will be available to bet on via sites like Kalshi and Polymarket. There's nothing stopping a dark money Super PAC from releasing a poll showing a particular candidate far behind, while simultaneously having the inside knowledge that this candidate is, in fact, a slight favorite. We've seen the GOP do everything under the sun to lie, cheat, and steal their way to election victories; why should the emergence of prediction markets be any different?
Going forward, we need to ignore every new poll released from now until November. We simply can't trust the process anymore. That's not to say we fly blind; rather, we rely on those on the ground rather than a survey of a few hundred voters based on their vibes of the current moment. We trust those putting in the work, especially our Black and Jewish voters who remain the base of the Democratic Party. Since the gutting of the VRA, Black voters are righteously angry, and they will be channeling that anger this fall, especially in places like Tennessee and Louisiana. We need to understand the makeup of these new gerrymandered districts: an R+5 district is undoubtedly in play this fall whereas an R+15 is not, no matter how inspiring a Democratic candidate in that district might be. Pollsters want to make news. They want to tell a particular story. They want to change the behaviors of those working on these campaigns. We cannot let them. Not now, not ever. Polls are done to create drama when none exists, and at a time when Democrats can and should crush the opposition, we know that pollsters will do everything in their power to create the impression of a horse race, right up until polls close on November 3rd. At the end of the day, we must come back to one simple fact: polls don't vote; people do.
And we need all of our people to understand that from now until November.