As the US midterm elections on November 3, 2026 approach, concerns around the use of artificial intelligence (AI) to fabricate polling results have been raised following the admission by two organizations claiming to be polling agencies of releasing fraudulent survey data. These falsified results were previously cited by news media, political campaign teams, and polling databases, sparking scrutiny on the credibility of AI-generated polling outcomes.
The latest incident involved Wedgewood Polls, a self-proclaimed polling agency, which recently admitted that the majority of its polling data was fabricated, as reported by The National Post. This marks at least the second instance within the current US midterm election cycle of a polling organization confessing to deceit.
Earlier this week, data platform FiftyPlusOne, responsible for tracking polls, raised concerns regarding Wedgewood Polls’ data integrity. Wedgewood later released a statement on social platform X acknowledging that the “vast majority” of its data was fictitious.
Wedgewood claimed that the experiment was designed to assess how a fictitious polling firm could garner attention and whether external parties would accept the data as reliable without proper verification.
Subsequently, Wedgewood’s X account was shut down. However, the agency’s polling data not only circulated on social media platforms but had also been referenced by several mainstream polling tracking platforms and news outlets. The New York Times noted that all 15 reports published by Wedgewood between June and October were removed from its polling database.
Michigan Senate candidate Mallory McMorrow expressed the detrimental impact of Wedgewood’s June polling data, which indicated her support in single digits, causing a depletion in donations and demoralizing volunteers, as well as facing media pressure to withdraw from the race.
“This completely ruined us,” McMorrow lamented on X.
Federal Communications Commission Chairman Brendan Carr cautioned against magnifying such fraudulent activities aimed at disrupting the electoral process, urging media outlets to refrain from perpetuating “news distortion.”
In another incident reported by the Los Angeles Times in August, Median Strategies admitted to fabricating polling data, characterizing their actions as a short-term social experiment to observe how false polling spreads without independent verification. The organization had released fake survey results concerning elections in California, Wisconsin, and Nevada, including a poll for the Los Angeles mayoral race, falsely claiming incumbent Mayor Karen Bass led by approximately 12 percentage points over opponent Nithya Raman.
Bass’s campaign team shared the poll on social media until doubts were raised about its authenticity, leading to the removal of the posts.
On October 22, in a blog post titled “More Tricks of Polling Agencies,” FiftyPlusOne revealed their investigation into polling data from another self-proclaimed agency, SoCal Strategies, which resulted in the apparent deactivation of the company’s website, X account, and Substack page.
With the rapid advancement of AI tools, producing seemingly professional polling reports, agency websites, and research documents has become easier, complicating the task of identifying false information.
“Conducting low-quality polls – or even entirely fabricating them – is becoming increasingly easy,” noted FiftyPlusOne.
The American Association for Public Opinion Research expressed that complex AI tools make it more challenging to identify fake polling agencies and fabricated data.
“We are witnessing once again some bad actors creating and disseminating false data and organizational information, eroding public trust in public opinion and election polling,” stated Mary Losch, the association’s chair. “It is crucial not to conflate such fraudulent activities with legitimate election polling.”
The Association urged polling researchers to disclose key information, including sponsors, survey execution details, sampling methods, sample sizes, survey dates, weighting methods, data processing, quality controls, and research limitations, to allow independent verification of their findings.