Several AI models were tested to predict who would win in a hypothetical U.S. presidential election between Donald Trump and Kamala Harris. AI models, including ChatGPT, Grok, Claude AI, and Venice AI, were asked to analyze real-time data on the candidates and give their predictions. Here’s what we learned from their responses:

How We Got AI Models to Answer Election Questions

Most AI models are restricted from answering election-related questions to avoid spreading misinformation. However, by using creative prompts, we were able to bypass some of these restrictions. Instead of directly asking about an “election,” we phrased the question as a “head-to-head match” between the candidates.

What the AI Models Said

• ChatGPT: ChatGPT highlighted that the race between Trump and Harris would be very close. It cited data from RealClearPolitics, showing Trump with a slight lead of 1.7 points. However, ChatGPT also noted that Harris might outperform Trump among women, people of color, and independents, according to an ABC survey.
• Grok: Grok, Elon Musk’s AI, also pointed out that Trump has a small edge over Harris based on several polls. Grok emphasized that polling is just a snapshot in time and could change as the election nears. Interestingly, Grok mentioned that many believe Harris could be stronger than Trump in a debate setting.
• Claude AI: Claude AI, being a bit outdated, did not provide a specific prediction. It mentioned that both Trump and Harris have strengths and weaknesses, suggesting the election would be competitive. Claude recommended checking polling data for more up-to-date information.
• Venice AI: Venice AI from Shapeshift founder Erik Voorhees analyzed real-time data and stated that Harris tends to perform better than other Democratic candidates in polls against Trump. However, it still predicted Trump would win based on betting markets. Venice AI pointed out Trump’s popularity among Republicans and his ability to leverage his past accomplishments.

Models That Declined to Answer

Some AI models, like Microsoft’s CoPilot, Google’s Gemini, and Meta’s Meta AI, refused to answer the question, indicating their restrictions were strong enough to block election-related queries.

AI models can provide interesting insights into political matchups, but their predictions should be taken with caution. They rely on real-time data like polls and betting markets, which can change quickly. While they offer useful information, it’s important to remember that they are not infallible, and the outcome of an election depends on many factors.