37 Fake News Sites Cited in ChatGPT Answers Favor Democratic Candidates, Firm Finds
A cybersecurity firm's findings illustrate how AI-generated answers can serve as an unintentional distribution channel for partisan content dressed as local journalism, raising questions...
Digital security firm VisionHeight identified 37 websites presenting themselves as independent local news outlets that were promoting Democratic candidates in content that appeared in ChatGPT responses to questions about congressional races, according to a report cited by NPR on October 30, 2026. The findings arrived four days before the November 3, 2026 midterm elections.
NPR described a number of the sites as "thinly veiled political ads made to look like reporting from local news outlets," according to VisionHeight's findings as reported by the New York Post. The exact methodology VisionHeight used to attribute partisan intent to each site, and the full list of the 37 URLs, had not been published in full as of this writing. What document would answer that question: the complete VisionHeight technical report, which the firm had not released publicly as of October 30, 2026.
The mechanism involved in the concern is specific to how large language models like ChatGPT retrieve and surface information. When a user asks a question about a candidate or race, the model may draw on text indexed from websites that appear credible. If a site mimics the format of local journalism but carries partisan content without disclosure, the model has no inherent mechanism to flag the site as non-journalistic. OpenAI has not publicly detailed what domain-level filters, if any, it applies to sources before they inform ChatGPT responses. OpenAI did not respond to a request for comment included in the NPR report, according to the New York Post's account.
The Federal Election Commission requires political committees and individuals who pay for public political communications to include a disclaimer identifying the payor. Whether websites that carry partisan content but do not solicit contributions or coordinate with a campaign meet the FEC's definition of a reportable expenditure is a legal question the commission has addressed inconsistently in prior advisory opinions. The FEC's most recent guidance on internet communications, Advisory Opinion 2006-20, exempted most unpaid online activity from disclaimer requirements, a standard that has not been updated through formal rulemaking to address AI-era content distribution.
No equivalent Republican-aligned network of fake local news sites was identified in the VisionHeight report as described in available coverage. Whether VisionHeight searched for right-leaning sites using the same methodology, and what it found if so, is unknown. That information would be contained in the full VisionHeight report.
The broader fake local news phenomenon predates this election cycle. The Stanford Internet Observatory and the NewsGuard misinformation tracking service have each documented networks of low-quality or politically motivated sites mimicking local news formats since at least 2022. NewsGuard's 2023 tally identified more than 1,200 such sites across the political spectrum, though its partisan breakdown varied by measurement period.
For voters, the practical effect depends on behavior. A user who asks ChatGPT which candidate to support in a specific House race and receives an answer drawn partly from a partisan site formatted as local news may not receive a disclosure that the underlying source carries a political viewpoint. Whether that constitutes deceptive electioneering under any existing federal statute has not been tested in court.
Congress has introduced but not passed several bills addressing AI and elections in the current session, including the Protect Elections from Deceptive AI Act, introduced in the Senate in 2023 and reintroduced in 2025. As of October 30, 2026, neither chamber had brought a comprehensive AI election integrity bill to a floor vote, according to the congressional record maintained at Congress.gov.
The Federal Communications Commission licenses broadcasters and can enforce sponsorship identification rules for broadcast political content, but its jurisdiction does not extend to websites or AI platforms under current statute. The FTC has authority over deceptive commercial practices but has not issued a formal rule specifically governing AI-generated or AI-surfaced political content.
What remains unknown: the identities of the individuals or organizations that registered and funded the 37 websites, whether any of them filed as political committees with the FEC, and whether OpenAI has taken action to remove or flag the sites since VisionHeight's report was shared. Domain registration records, FEC committee filings, and a formal OpenAI statement would each address one of those open questions.