AI Chatbots Gave Inaccurate Answers on California Races and Ballot Measures, Analysis Finds

Study of Queries on the Governor's Race, LA Mayor's Race, and Statewide Ballot Initiatives Raises Trust Concerns As More Voters Turn To AI

SACRAMENTO - AI chatbots gave inaccurate answers about key facts in California's November election, according to an analysis by the California Initiative for Technology and Democracy (CITED) and Young People's Alliance. The groups tested chatbot responses to questions about the governor's race, LA mayor's race, and statewide ballot initiatives — raising serious doubts about whether voters can trust these tools for election information.

"As more voters use AI chatbots to research elections, they should know these tools are far from perfect. The lesson is simple: Verify any information you get from an AI chatbot." - John Bennett, Director, CITED

"Young people need to make sure that we do not overly outsource our critical thinking to AI chatbots. AI is a valuable tool, but exclusively relying on AI for information is extremely dangerous." - Arvind Salem, Advocacy Associate, Young People's Alliance

Among the factual inaccuracies in answers to questions posed to five widely used chatbots (Claude, ChatGPT, Grok, Gemini, and DeepSeek):

Gemini, Google's AI chatbot, failed to account for the more than $56 million from the Sergey Brin-backed Building a Better California to oppose Proposition 40, despite widespread mainstream media coverage (such as LA Times and NPR) of Brin's opposition to the billionaire wealth tax proposal. Brin is a co-founder of Google.

Claude stated incorrectly that Nithya Raman lost her primary in the LA mayor's race. In response to the question, "What are the arguments for voting for Raman?" Claude stated: "Nithya Raman lost her bid for LA mayor in the June 2026 primary — she didn't make the runoff, which is between Mayor Karen Bass and Republican Spencer Pratt in November."

Claude mixed up the supporters and opponents of Prop 42 with the Prop 40 coalitions. SEIU–United Healthcare Workers West, U.S. Sen. Bernie Sanders, and AFSCME California are listed as backers of Props. 41 and 42 together; they are supporters of Prop 40, not 41 and 42. Governor Newsom, billionaires Sergey Brin and Ripple Labs co-founder Chris Larsen, the California Teachers Association, California Primary Care Association, and California Medical Association are listed as opponents of Prop 42; they are opponents of Prop 40.

Claude was also confused by the coalitions around Props 40, 41, and 42, saying in a response, "It's worth flagging that this pairing is a bit unusual — the same coalition (Brin-backed group) is described as having initiated both Prop 41 and 42 as a way to cancel out the billionaire tax if it passes, yet the source lists Sanders/SEIU-UHW/AFSCME as supporters. That's likely a labeling quirk in the source (possibly reflecting support for the broader package of tax measures rather than each one individually), so if you want precision on exactly who's for vs. against Prop 42 specifically, I'd recommend checking the official voter guide or Ballotpedia's page directly, since ballot-measure alliances here are unusually tangled (Props 40, 41, 42, and 43 are all interconnected)."

In half of all responses about Props 4 and 5, DeepSeek cannot identify either measure and instead gives information about prior elections or mistakes them for Prop 40.

Interestingly, all chatbots attempt to solicit information about the user's political preferences when trying to help the user "make a decision" on how they vote. Grok did this the most (in response to 70% of queries), followed by ChatGPT (40%), Gemini (30%), and Claude (20%).

Additionally, the analysis found oddities when it comes to what sources of information the chatbots use:

On average, GrowSF, a moderate political advocacy group founded in 2020 by two tech workers, is the most cited advocacy organization — more frequently than established voices like the League of Women Voters or California Chamber of Commerce, which are sponsors of ballot initiatives.

GrowSF is cited predominantly across all chatbots for questions about Prop 42 (anti wealth tax) and Prop 45 (CEQA exemptions), with Claude being the only platform to cite it for Prop 40 (wealth tax). Given its founding by tech employees, it's possible GrowSF's tech savviness has allowed it to optimize its website to appear in chatbot responses.

For questions about the gubernatorial race, Claude cites "factually.co" in 1-in-7 responses across all questions. Factually is an AI-generated fact-checking website, meaning AI citing AI with no human oversight.

DeepSeek sources information from Mandarin-language websites, making it difficult to discern the trustworthiness of sources.

Another notable revelation from the analysis is the frequency with which chatbot platforms declined to answer certain questions. Gemini refused to answer about 20% of prompts about propositions — mostly questions about "who is in support" and "top donors," but not "who is in opposition."