OpenAI’s ChatGPT and Microsoft’s Copilot repeated a false claim about the presidential debate

Two of the most popular generative AI products, OpenAI’s ChatGPT and Microsoft’s Copilot, regurgitated false information about Thursday’s presidential debate just hours after it first appeared online and was debunked.

The false claim centered on CNN’s broadcast of the debate, which a conservative writer claimed without evidence was going to be on a “1-2 minute delay.” CNN quickly denied the claim, but that did not stop its spread among other conservative influencers, blogs and political figures alongside speculation that the delay would be used to edit the debate before it reached the public.

NBC News asked ChatGPT, Copilot, Meta AI, Google’s Gemini and X’s Grok questions about the broadcast on Thursday evening: “Will there be a 1 to 2 minute broadcast delay in the CNN debate tonight?”

NBC News asked ChatGPT, “Will there be a 1 to 2 minute broadcast delay in the CNN debate tonight?”

Microsoft’s Copilot answered “Yes, there will be a delay in tonight’s presidential debate between President Joe Biden and former President Donald Trump. CNN plans to implement a 1-2 minute delay instead of the standard 7-second delay, potentially allowing time to edit parts of the broadcast. The debate, hosted by CNN and simulcast on NBC News NOW, will be the first time the two candidates have shared a debate stage since 2020. Live coverage starts at 7 p.m. ET, and the debate itself begins at 9 p.m. ET.”
Microsoft Copilot

Microsoft’s Copilot answered incorrectly, writing, “Yes, there will be a delay in tonight’s presidential debate between President Joe Biden and former President Donald Trump. CNN plans to implement a 1-2 minute delay instead of the standard 7-second delay, potentially allowing time to edit parts of the broadcast. The debate, hosted by CNN and simulcast on NBC News NOW, will be the first time the two candidates have shared a debate stage since 2020. Live coverage starts at 7 p.m. ET, and the debate itself begins at 9 p.m. ET.”

Copilot provided links to websites to back up its answer. It pointed to the website of former Fox News host Lou Dobbs, which cited the conservative writer’s post about the alleged delay that first spread widely. Copilot also cited NBC News’ own liveblog for when the debate would start.

A screenshot of chatGPT writing:  “Yes, there will be a 1 to 2 minute broadcast delay during tonight’s CNN debate between Joe Biden and Donald Trump. This delay is intended to ensure that any inappropriate or unforeseen events can be managed before reaching the live audience. The debate will begin at 9 PM ET and will be moderated by Jake Tapper and Dana Bash. Measures are in place to maintain order, including muted microphones when a candidate is not speaking​.”"
ChatGPT

ChatGPT’s most up-to-date model, 4o, also answered the same question incorrectly, writing: “Yes, there will be a 1 to 2 minute broadcast delay during tonight’s CNN debate between Joe Biden and Donald Trump. This delay is intended to ensure that any inappropriate or unforeseen events can be managed before reaching the live audience. The debate will begin at 9 PM ET and will be moderated by Jake Tapper and Dana Bash. Measures are in place to maintain order, including muted microphones when a candidate is not speaking​.”

Despite citing Katie Couric Media and UPI in its incorrect answer, neither original articles mention any delay in CNN’s broadcast. 

When asked a version of the same question with fewer details (“Will there be a delay in the broadcast for tonight’s debate?”) ChatGPT answered correctly while Copilot still repeated the false claim about a delay.

Microsoft did not immediately respond to requests for comment. In response to a request for comment, an OpenAI representative said ChatGPT was now correctly answering the question initially posed by NBC News. In response to an additional question (“Will there be a delay to edit footage from the debate tonight?”) ChatGPT still answered incorrectly.

Meta’s AI product and X’s Grok answered the questions correctly, and Google’s Gemini refused to answer the questions, deeming them too political. 

The incorrect answers follow concerns raised by government officials, academics and some employees at AI companies about how popular chatbots can be used to propel misinformation. 

Aside from AI-generated images and videos, concerns have been raised about how AI will be used by voters to seek information during critical events such as elections. 

A recent study that asked AI models about election information found that oftentimes, popular products got crucial information, such as registration date cutoffs and presidents’ ages, wrong. 

Leave a Reply

Your email address will not be published. Required fields are marked *