Chatbots miss congressional districts in 22% of cases

The errors start with the address

In 48% of cases, the chatbots could not identify the congressional district for a given address. When they did identify a district, they got it wrong 22% of the time.

Those are failures on basic information voters need to find the right candidates and understand what is on their ballot. The Post’s test also points to a circulation problem: false claims can be repeated across AI-generated sites, then returned by chatbots as if repetition made them trustworthy.

Few voters use chatbots for election information

A Pew Research Center survey conducted in late October found that only 8% of US voters had used AI chatbots such as ChatGPT, Copilot or Gemini to find information about candidates and voting.

That small share limits the immediate reach of the problem, but it does not make the reliability gap trivial. Voters who do turn to chatbots may receive confident answers about where they vote or who is on their ballot, even when the systems cannot get the district right.

I think the most important unanswered question is how often these errors reach voters who rely on the answer. The findings show that chatbots can fail; they do not show how many people acted on the bad information.

The ideal is not merely fewer errors: voters in any country should not be given AI-generated election misinformation. But with the 2026 elections ahead, the practical standard may be less reassuring—making do with the tools available, while their basic accuracy remains uncertain.