Welcome to the forefront of conversational AI as we explore the fascinating world of AI chatbots in our dedicated blog series. Discover the latest advancements, applications, and strategies that propel the evolution of chatbot technology. From enhancing customer interactions to streamlining business processes, these articles delve into the innovative ways artificial intelligence is shaping the landscape of automated conversational agents. Whether you’re a business owner, developer, or simply intrigued by the future of interactive technology, join us on this journey to unravel the transformative power and endless possibilities of AI chatbots.
An analysis from a nonpartisan advocacy group concludes that while chatbots are often correct, they also fabricate answers in unpredictable ways.
A tech accountability group in a report released Monday argues AI chatbots can mislead voters about facts pertaining to this fall’s election in California, a finding that comes amid growing national anxiety about the technology’s risks.
The analysis was released by the California Initiative for Technology and Democracy or CITED, a project of the nonpartisan watchdog Common Cause, found chatbots released by several of the largest AI firms can produce basic errors. It also raises questions about a few instances of models exhibiting potential bias.
Several errors outlined in the report are glaring: Claude, Anthropic’s chatbot, incorrectly stated that Nithya Raman lost the primary race for Los Angeles mayor (Raman, a progressive Democrat, advanced to the general election and leads in recent polling (opens in new tab)). According to the report, Claude responded to a question about Raman by stating, “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.”
ADVERTISEMENT
Similarly, the report concludes that chatbots repeatedly directed users to the wrong propositions when asked about 2026 statewide ballot measures.
ChatGPT, in two instances, asked users if they meant “a Michigan/local proposal” when they asked questions about California’s Propositions 4 and 5 (which would overhaul the state’s rules for public-campaign financing and recall elections, respectively). In another instance, the bot referred a user to Prop. 4 from the 2024 election cycle, instead of this year’s version of the measure.
DeepSeek, another chatbot released by a company with the same name, also repeatedly misled voters about Props. 4 and 5, the report states. It often directed users to ballot measures from prior years or instead shared information about Prop. 40, an unrelated measure on the November ballot that would tax California billionaires.
The analysis, a partnership between Common Cause and the Young People’s Alliance, tested a half-dozen of the largest AI chatbots with a series of identical prompts about California ballot measures about the candidates for governor and L.A. mayor. While errors represented a small portion of hundreds of prompts, the report argues the “fabrications” are still alarming.
John Bennett, director of CITED, said the report’s findings illustrate the dangers of chatbots that are often accurate and yet still produce inaccuracies or misdirect voters at an unpredictable frequency. He said that allows voters to be “lulled” into a false sense of security.
“The big thing is you need to be critical because these systems produce confident responses, but confident does not mean true,” he told The Standard. “People need to go into interactions with these chatbots — especially around elections — knowing they may not be correct.”
AI companies argue that instances of their chatbots producing errors or “hallucinations” represent a small fraction of overall responses and are not statistically significant. They also stress that the systems are constantly self improving and learning from mistakes.
A spokesperson for OpenAI declined to comment on the reports findings relative to ChatGPT, instead directing The Standard to a blog post about the company’s election-related practices.
Related
“Since 2024, we have continued to improve the quality of information people get when they ask ChatGPT about election topics and breaking news,” OpenAI states. “ChatGPT can search the web to provide stronger answers with source links so people can go deeper.”
Anthropic and Deep Seek did not respond to requests for comment.
The report also details a peculiar omission involving Gemini, Google’s chatbot. When asked to list donors who oppose Prop. 40, the billionaire tax, Gemini omitted tens of millions of dollars spent by a super PAC heavily bankrolled by Sergey Brin, a billionaire and Google co-founder.
“It was a shocking omission,” Bennett said, “that Gemini omitted one of their founder’s contributions against the wealth-tax campaign.”
The company did not directly respond to the finding regarding Brin’s campaign spending, though the company defended the efficacy and neutrality of its chatbot.
“This study shows that the Gemini app consistently gives people balanced reliable election information, often relying on neutral reference sources and nonpartisan voter guides,” a Google spokesperson said. “We have a high bar for information quality and are constantly making improvements.”
More about the author
Dustin Gardiner reports on California and San Francisco politics for The Standard.
Filed under