Two AI assistants nailed voter-election accuracy in one audit — do the others ask you to trust do just as well, or does it depend on who you ask?

This explores whether AI assistants other than ChatGPT and Google's AI, which posted strong accuracy numbers in one US audit, are just as reliable when voters ask them about elections. The short answer from the corpus: on simple factual questions the leading assistants have gotten quite good. Once the question becomes about which candidate or party fits you, or about current events, the problems start to look the same across every assistant tested. That is the more interesting part. The assistants don't fail in different ways. They tend to fail in the same way.

Start with the good news. States United found that ChatGPT and Google's AI reached zero verifiable factual errors on voter questions by early 2026 Do AI chatbots give voters accurate election information?. Being accurate is not the same as being useful, though. Fewer than half of their answers pointed voters to official state election websites, candidate information was often incomplete, and the assistants leaned on sources anyone can edit, like Wikipedia. The corpus has no comparable item-by-item audit of Claude, Copilot, Meta AI or other assistants on the same factual questions, so a direct head-to-head on accuracy isn't available here.

What the corpus does have is evidence that looks across several assistants at once, and it points to shared weaknesses. When the Dutch Data Protection Authority tested four chatbots as voting advisors, the assistants recommended the same two parties (PVV and GroenLinks–PvdA) in more than half of cases, even when the test user's stated views matched a different party. Centrist parties showed up less than 2% of the time Do AI chatbots systematically bias voters toward extreme parties?. The regulator traced this to the chatbots drawing on loose internet text instead of structured data on party positions Do chatbots steer Dutch voters toward the same parties?. On news questions more broadly, a study run by 22 public broadcasters found significant errors in nearly half of more than 3,000 responses, and the problems appeared on every major platform and in every language tested Do AI assistants reliably answer questions about news?.

The likely reason the assistants behave alike comes from outside election research. A study of 70+ models found an "Artificial Hivemind" effect: different LLMs, built by different companies, independently produce strikingly similar answers because they share much of their training data and alignment methods Do different AI models actually produce diverse outputs?. For voters, this means switching to a different chatbot may not get you a real second opinion. If one assistant funnels voters toward the same two parties, a competitor trained on similar web text may well do the same. A related line of work suggests one fix. Assistants tend to fill gaps in what they know about the user with confident guesses, and making them explicitly track what they don't know cut hallucinations roughly in half Do language models know what they don't know about users?. Applied to elections, an assistant that says "I don't have your state's current deadline, check here" is more useful than one that is merely fluent.

How much this matters depends on how many people rely on these tools, and right now not many do. Only 15% of US voters planned to use a chatbot for 2026 election information, mostly younger voters and voters of color, and a majority said they distrust AI accuracy Will voters actually use AI chatbots for election information?. People also tend to open an assistant after they've already searched, not before Do people use AI assistants before or after searching?, so for many voters chatbots act as a second step after search. The takeaway: the gap between assistants on basic facts may be closing, but they seem to share blind spots, and testing one model doesn't tell you the others are safe.