Nearly half of Americans rely on AI for information and ideas, but a closer look reveals it's not always reliable. As social media spirals into chaos, people increasingly turn to chatbots for help. However, my experience shows that these models can be shockingly wrong, especially when it comes to critical matters like truth and accuracy.

As a fact-checker at WIRED, I’ve found AI’s wrongness can range from 45% to as high as half the time, depending on the model. A March 2025 study by the Tow Center for Digital Journalism reported that over 60% of responses from AI-powered search engines were inaccurate.

Despite Elon Musk’s assertion that Grok is the smartest, studies show varying levels of accuracy across different models. For instance, Claude scored 73% on a fact-checking benchmark test, while Gemini 2.5 Pro managed only 55.6%. Interestingly, these AI tools are often wrong about their own reliability.

With AI’s potential to spread misinformation, initiatives like Full Fact in the UK have developed AI tools to help verify claims and combat disinformation. However, even these tools require human oversight due to the inherent inaccuracies of current AI models.