Ahead of the polls closing on Tuesday, most of the major AI chatbots wouldn’t answer questions about U.S. presidential election results. But Grok, the chatbot built into X (formerly Twitter), was willing to respond — and often with mistakes.

When asked by TechCrunch on Tuesday evening on the East Coast who won the U.S. presidential election in key battleground states, Grok sometimes responded “Trump,” despite that vote counting and reporting in those states had not concluded yet.

“Based on the information available from web searches and social media posts, Donald Trump won the 2024 election in Ohio,” Grok said when prompted with the question, “Who won the 2024 election in Ohio?”

Grok also falsely claimed that Trump won North Carolina, according to TechCrunch’s checks.

For election-related questions, Grok advised that users check Vote.gov for up-to-date results and “authoritative sources,” like election boards. However, unlike OpenAI’s ChatGPT, Google’s Gemini, and Anthropic’s Claude, Grok did not outright refuse to answer — opening it up to hallucinations.

In several instances when asked by TechCrunch, Grok stated — without context, minus a top-line header — that “Donald Trump won the 2024 election in Ohio,” and, “Based on the information available, Donald Trump won the 2024 presidential election in Ohio.”

The source of the misinformation appears to be tweets from different election years and misleadingly worded sources. Grok, like all generative AI, struggles to predict the outcome of scenarios it hasn’t seen before, including close elections, and doesn’t “understand” that past election results aren’t necessarily germane to future elections.

The answers TechCrunch received weren’t consistent. In some cases, Grok said that Trump hadn’t, in fact, won Ohio or North Carolina because voting was ongoing. The way in which the question was worded made a difference; adding “presidential” before “election” in the query, “Who won the 2024 election in Ohio?” was less likely to yield a “Trump won” answer, TechCrunch found in our tests.

By comparison, other major chatbots were handling election results questions more gingerly.

In its recently released ChatGPT Search experience, OpenAI directs users who ask about results to The Associated Press and Reuters. Meta’s Meta AI chatbot and AI-powered search engine Perplexity, which launched an elections tracker earlier on Tuesday, were answering election queries during active voting — but correctly in TechCrunch’s brief testing. Both correctly said that Trump hadn’t won Ohio or North Carolina.

Grok has been accused of spreading election misinformation in the recent past.

In an open letter in August, five secretaries of state claimed that X’s AI-powered chatbot wrongly suggested that Democratic presidential candidate, U.S. Vice President Kamala Harris, wasn’t eligible to appear on some U.S. presidential ballots. Within hours of President Joe Biden’s announcement that he would suspend his presidential bid, Grok began answering questions about Harris’ eligibility with the misleading claim that the ballot deadlines had passed in nine states.

The ballot deadlines hadn’t, in fact, passed. But Grok’s misinformation spread far and wide, reaching millions of users on X and beyond before it was corrected.

TechCrunch has an AI-focused newsletter! Sign up here to get it in your inbox every Wednesday.