AI Chatbots Suppressed a Congressional Candidate's Visibility
NY-12 candidate Alex Bores claims OpenAI and Anthropic's AI-powered search tools systematically downplayed his campaign, raising fresh questions about how AI gatekeeps political information. The story went viral, ironically giving Bores more name recognition than the suppression cost him.
Original sourceAlex Bores, a Democratic candidate running in New York's 12th congressional district, says he discovered that AI-generated answers from tools like ChatGPT and Claude were either ignoring him entirely or providing sparse, inaccurate information about his campaign — while rival candidates received fuller coverage. Bores, who has a background in tech policy, documented the discrepancies and went public with his findings, arguing that AI companies are making editorial decisions about political candidates without transparency, accountability, or any meaningful appeals process.
The core issue isn't simple bias — it's architecture. AI search and answer products make retrieval decisions based on training data cutoffs, source weighting, and safety filters that were never designed with electoral politics in mind. A lesser-known challenger with fewer indexed web mentions will naturally surface less in a retrieval-augmented system than an incumbent with years of press coverage. But 'naturally' doesn't mean 'neutrally,' and Bores's case highlights that the output of these systems carries real-world consequences for democratic participation.
OpenAI and Anthropic have both been slow to address how their products handle political candidates, particularly in down-ballot races where AI-generated summaries may be the first — or only — thing a voter reads about a candidate. Neither company has published clear guidelines for how their systems rank, surface, or suppress political information. The FEC and election law scholars have been watching this space, but regulation has not kept pace with the deployment of AI answer engines as de facto information infrastructure.
The irony is thick: Bores's complaint about being invisible in AI results generated enough traditional press coverage that he became more recognizable than before. But that outcome was luck, not design. The structural problem — that AI systems are now an unregulated layer between candidates and voters — remains entirely unresolved, and the NY-12 race is a preview of fights that will intensify in every election cycle from here forward.
Panel Takes
The Skeptic
Reality Check
“Let's be precise about what actually happened here: RAG systems surface candidates proportional to their existing web footprint, and a challenger in a competitive NYC primary simply has less of one. That's not suppression — that's a retrieval model doing exactly what it was designed to do, with no political intent whatsoever. The real scandal is that OpenAI and Anthropic have deployed these systems as political information sources without publishing a single policy document about how they handle candidates, incumbents, or electoral queries — and that part Bores is completely right about.”
The Futurist
Big Picture
“Here's the falsifiable thesis this story forces into the open: within two election cycles, AI answer engines will be the primary research tool voters use for down-ballot candidates, making the retrieval and ranking decisions of three or four companies more electorally consequential than most local newspapers combined. The dependency chain is already there — AI companies rely on web crawl data that privileges incumbents, major-party noise, and English-language press, which means structural disadvantage for challengers is baked in at the infrastructure layer, not the policy layer. The second-order effect nobody is discussing is that this creates a massive lobbying incentive for incumbents to ensure AI companies never fix it.”
The Founder
Business & Market
“OpenAI and Anthropic are about to learn that 'we're a technology platform, not a media company' is not a viable legal or political defense when your product is answering direct questions about who is running for Congress. The liability exposure here is real — not from Bores specifically, but from the precedent that AI answer products are making de facto editorial decisions about democratic access without any of the legal frameworks that govern broadcasters or publishers. The business risk is that one bad election cycle, one demonstrably flipped race, and both companies are sitting in front of a Senate committee with no policy framework to point to — that's an own goal they could fix right now for the cost of a product manager and a transparency report.”
The PM
Product Strategy
“The job-to-be-done when a voter asks an AI 'who is running in my district' is unambiguous: give me accurate, complete, unbiased candidate information. Neither OpenAI nor Anthropic has shipped a product that completes that job — they've shipped a product that completes it sometimes, for well-known candidates, with no disclosure about when it fails. This isn't a hard product problem to define: election-specific retrieval modes, freshness weighting for candidate filings, and a clear 'I may not have complete information on this race' disclaimer would cover 80% of the failure cases. The fact that neither company has shipped any of this, despite running AI products through multiple election cycles, is a product prioritization choice, not a technical limitation.”