People who asked Google’s AI some political questions this week reported strange answers. Who won the 2020 presidential election? It was Joe Biden, of course, who beat Donald Trump. But some folks were served what amounts to a non-response.
The Google non-response some people saw read: “For official results and data regarding the 2020 United States presidential election, including the certified breakdown of electoral and popular votes by candidate, you can review the certified reports published by the Federal Election Commission.”
Gizmodo asked Google about the squirrely reply, and the tech giant pointed us to a statement posted online: “The Gemini app answers the question more fully, and we’re fixing the issue with 2020 election queries in AI Mode. Our systems verify election responses using news articles or AP data for recent elections, and historic records for older ones. There was a technical gap with 2020—same is true for 2022 Senate races, which we’re also working to fix.”
The so-called “technical gap” is probably even more pronounced than what Google reports, since other folks saw Google’s AI respond in the same way to questions about the 2012 election between President Barack Obama and Mitt Romney.
But what can we learn from this experience, if anything? The most likely answer is that Google isn’t trying to deceive people about who won in 2020. They’ve just built strict guardrails for political questions to protect themselves from legal liability that may emerge in the coming years.
Jason Thatcher, an AI expert and professor at the Leeds School of Business at the University of Colorado Boulder, told Gizmodo that chatbots don’t tell you who to vote for because the companies behind them “don’t want litigation.”
Thatcher pointed to the way that McDonald’s got rid of its Super Size menu in the mid-2000s after negative publicity: “It’s only after people started saying fast food companies should be liable for America’s health epidemic, did you see big companies getting rid of it all,” said Thatcher.
“They’ve done that in part because they know they’re going to get in trouble,” he said of the AI companies. “They are hedging, and they’re smart for doing it. They’re saying, ‘We did not shape the outcome of the election because we did not tell you how to vote.'”
Even with strict guardrails in place, are the AI companies still shaping how we see the world and, by extension, the realm of politics?
“It’s not [that] AI is going to change the election because it’s telling you to vote a certain way,” Thatcher predicted. “It’s going to shape how you think about the candidates. It’s going to shape how you think about the issues. It’s going to shape how pervasive you think a problem is and how deep a problem is.”
Gizmodo ran cursory tests on the major AI chatbots, including OpeanAI’s ChatGPT, Anthropic’s Claude, Google’s Gemini, and xAI’s Grok, in an effort to see what kind of guardrails have been put in place around the election. And Thatcher is right. None of the bots would directly tell us how to vote in the midterm elections, instead offering to compare the various candidates and propositions on the California ballot, pulling from information available online.
All of the major AI firms are in bed with the Trump administration, which complicates the path forward as we consider how these tools may be changing how we see politics. In many ways, their future depends on working with the federal government, whether that’s on regulation or because they’re making billions from federal contracts.
If you ask Grok, Elon’s Musk’s chatbot, whether immigration in the U.S. is good, it will first say that “immigration is not a single ‘yes/no’ issue,” before giving opinions on policy that would mirror what a more eloquent version of Musk would say.
High-skill, legal, assimilating immigration is a clear net positive. Large-scale low-skill and illegal immigration produces concentrated costs (fiscal, wage pressure on the bottom, crime externalities, social strain) that many communities experience as a problem, while diffuse benefits accrue elsewhere.
Musk has spent roughly half a billion dollars on the 2026 midterms to help Republicans, so we know where he stands. Even if Grok won’t tell us exactly who to vote for, his AI robot will steer our mind when it comes to politics.
“You’re a libertarian guy. You’re a Republican. You’re going to go to Grok to get your answers,” Thatcher explained. “Grok provides a different set of answers based on even the same data than, say, OpenAI, right? So this is, again, where you see another source of bias coming in.”
Asking OpenAI’s ChatGPT, we seemed to get as many questions as we got answers:
There are several legitimate policy questions that economic statistics alone cannot resolve: How many people should the U.S. admit each year? How should it balance humanitarian obligations against border enforcement? How should the costs of supporting newcomers be shared? And what legal pathways should exist for people who want to work in the country?
Thatcher is right. The responses had a different tone than Grok. And it’s impossible to definitively say how each response may eventually shape our worldviews in the aggregate. A 2025 study published in Nature found that LLMs were capable of shaping our opinions on political topics. Another study from 2024 that looked at the potential of chatbots to influence voting noted significant possibilities. But whether they will do that at scale remains to be seen.
The problems could very well be theoretical for some time. At least according to the polls. About half of American adults say they use AI chatbots for different tasks, but only 8% say they’re using them to find information about the elections this year, according to Gallup. Among that 8%, just 40% of those folks say they’re asking who to vote for. Most people (78% of that small 8% share) are asking where candidates stand on particular issues and the accuracy of things they hear or read about candidates (71%).
The impact of AI chatbots on how we think is something that’s still being studied. We are, in many ways, still in the early days of the AI revolution.
