As crucial elections loom in the United States, AI chatbots are dumping political misinformation onto the populace with astonishing confidence.
There’s evidence that the problem is widespread. According to recent reporting by the Washington Post, top AI chatbots are giving horribly flawed information to users asking for voting guides on the midterms. For example, when WaPo tasked ChatGPT with creating a voter guide for all the races and candidates relevant to a resident of Missouri, the chatbot added a Senate seat that isn’t even up for re-election. A similar test for a resident in Wyoming claimed that local politicians running in municipal elections were running for federal seats.
In all, WaPo took a look at 3,000 AI-generated voting guides, which frequently shared false voting information that usually traced back to other AI-generated websites. In other words, the AI chatbots are eating their own slop and passing it off as authoritative election information.
Overall, the paper found chatbots failed to find which congressional district an exact address was located in a whopping 48 percent of the time, while those that did got it wrong 22 percent of the time.
Luckily, the share of users actually querying the chatbots seems to be relatively low. According to a late-October survey by the Pew Research Center, just eight percent of US voters say they’ve used an AI chatbot like ChatGPT, Copilot, or Gemini to look up information relevant to the ballot-box.
While the ideal number would be zero voters in any nation receiving AI-generated election misinformation, it seems in 2026, we’ll have to take what we can get.
More on elections: Political Candidates Spending An Obscene Amount of Money on AI-Generated Election Slop