ChatGPT Cited Fake News in Half of Election Queries — While Tech Giants Adopt Trump's AI Label
How this was made Verified AI
Every Intellegix briefing is generated from that day's broadcast and run through automated checks before it publishes — with a human paged on any flag. Here is the trail for this edition.
NPR posed 37 questions about competitive House races to ChatGPT and found that in roughly half of those cases, the model cited fake news websites as sources. Those fabricated sites had a consistent directional lean, favoring Democratic candidates. Google's Gemini, given the same questions, cited none of the fake sites. With millions of voters potentially using AI assistants to research candidates, the finding represents a direct misinformation vector at exactly the moment it can do the most damage.
The partisan slant of the fake sites does not necessarily indicate political bias in OpenAI's systems; it may instead reflect coordinated influence operations that built more of these sites targeting specific races, which ChatGPT's retrieval systems then surfaced without adequate verification. That distinction matters to engineers fixing the problem but not to the voter who received bad information. OpenAI has implemented election-specific guardrails this cycle; the NPR test suggests they are insufficient for granular down-ballot races where independent fact-checking infrastructure is thin.
The performance gap between ChatGPT and Gemini likely reflects architectural differences. Google has spent roughly 25 years building search-quality and source-verification systems at the core of its business; that institutional knowledge appears to inform how Gemini's citation systems were constructed. OpenAI built a language model first and added web retrieval later — a different design philosophy with measurable consequences in this kind of controlled test.
While one AI tool was spreading misinformation about elections, both Elon Musk and Mark Zuckerberg separately adopted President Trump's preferred term 'super intelligence' for advanced AI — a linguistic convergence between two men who are otherwise fierce rivals across multiple product categories. The alignment is less likely intellectual agreement on AI taxonomy than strategic positioning: in an environment where the administration controls AI regulation, export controls, and procurement decisions, adopting the President's preferred framing is a form of lobbying. When a President's terminology migrates into legislative drafts and regulatory guidance, the companies that helped define the term gain an early advantage in shaping its meaning.