Mistake in AI-generated Community Note sparks controversy about anti-Trump protest

6 hours ago 2

Organisers reported that five million Americans took to the streets in towns and cities across the country on October 18 as part of the “No Kings” movement: a day of mobilisation against President Donald Trump

In Boston alone, organisers reported that 100,000 people joined the protests against the American president. An X account called “Republicans against Trump” published a video showing an aerial view of the large Boston crowd.

However, a number of pro-Trump social media users believe that this video is fake, basing their claims on a Community Note added to the post that casts doubt on the veracity of the video. 

A Community Note is a function on X that allows social media users to add fact checks to a post that they think is false or misleading, providing the sources that led them to this conclusion. Essentially, the Community Notes system is a crowd-sourced, community-based moderation system. 

As such, the Community Note added to the post featuring the video of the Boston protest casts doubt on its veracity:

“This video is from a 2017 Boston protest, not today’s ‘No Kings’ rally, which was indeed a large event with tens of thousands attending per reports.”

Pro-Trump social media influencer Laura Loomer claimed, based on this Community Note, that Democrats posted a fake video of protests in Boston. However, it turns out that the Community Note is erroneous. Pro-Trump social media influencer Laura Loomer claimed, based on this Community Note, that Democrats posted a fake video of protests in Boston. However, it turns out that the Community Note is erroneous. © X

“Wow. Democrats got caught faking their crowd sizes for No Kings Day. They lied and said 7 million people came out to protest President Trump,” wrote pro-Trump influencer Laura Loomer, sharing a screengrab of the post and the Community Note attached. 

Turns out, the video is indeed from October 18

However, this Community Note is inaccurate. 

To justify its claim that the video of the Boston protest is actually from 2017, this Community Note cites three sources: a Wikipedia page about a 2017 Boston protest, an article by the news agency Reuters and an article from the US daily newspaper USA Today.

However, none of these sources features the aerial view of the Boston protest posted by the Republicans Against Trump account. This means that the sources don’t prove that the video is actually from 2017, as some right-wing, pro-Trump social media users have been quick to claim. 

In reality, contrary to what the Note claimed, the video posted by the Republicans Against Trump is authentic. Another aerial view of the October 2025 Boston protest, this one filmed from another angle, was posted on the YouTube channel of the American television channel MSNBC.

Above is a screengrab of the video of the Boston “No Kings’ protest on October 18, 2025, posted on X by Republicans Against Trump. Below is a video of the same protest posted on MSNBC’s YouTube channel. Above is a screengrab of the video of the Boston “No Kings’ protest on October 18, 2025, posted on X by Republicans Against Trump. Below is a video of the same protest posted on MSNBC’s YouTube channel. © X

The MSNBC video, published on October 18, shows the same buildings, albeit at a slightly different angle, as the video posted on X. You can see one building with a screen, an annexe on the left side and a tent in the midst of the crowd. This indicates that the two videos show the same protest and were filmed on the same day.  

An erroneous Community Note written by AI

How does a moderation error like this happen?

The note indicates that its author is a user called “Zesty Walnut Grackle”. According to X, this author isn’t a real human but an “experimental AI contributor”. In other words, the author is what’s commonly called a “bot” or an AI robot.

The erroneous Community Note left on the video of the Boston “No Kings” protest was written by artificial intelligence. The erroneous Community Note left on the video of the Boston “No Kings” protest was written by artificial intelligence. © X

Starting on July 1, X added a new function as part of a “pilot programme". X now allows the chatbot Grok, X’s artificial intelligence, to write Notes. Before, Notes were only written by real users. Using AI means that now part of the process of Note writing is automated, though X users still validate them by assessing their utility.

According to the official Community Notes account, this new function is meant to “accelerate the speed and scale of Community Notes”.

This graphic shows the new way that Community Notes will work now that AI has been integrated. This graphic shows the new way that Community Notes will work now that AI has been integrated. © Research article: Scaling Human Judgment in Community Notes with LLMs

Social media users are already using Grok to verify images or videos, typically by calling on the AI in post comments or directly by the chatbot. In a previous article, we noted that AI often provides erroneous results first and later corrects them. This leads to false or misleading analysis in the comments.   

It seems as if, now, the same type of errors are appearing in the community notes system.

The history of Community Notes produced by Zesty Walnut Grackle shows that the AI produced inconsistent evaluations of the Boston videos. First, Zesty Walnut Grackle published an erroneous evaluation of the video posted by the account Republicans for Trump, falsely claiming that it was filmed in 2017 instead of on October 18, 2025. A few hours later, it actually verified the same video featured in another post correctly, but by that point, it had already published a number of Notes sharing the initial error.

The history of Community Notes shows that the AI produced inconsistent evaluations of the Boston videos. The history of Community Notes shows that the AI produced inconsistent evaluations of the Boston videos. © X

Our team asked Grok why it made the mistake: 

“In summary, this specific response on X was based on a partial analysis and incomplete information at that moment, leading to an error. My knowledge is constantly updated and, based on current proof (video analysis, web research and media confirmations), the video is from October 18, 2025. If responses vary, it is often because of the context or data available at the time of the interaction.” 

This article has been translated from the original in French.

Read Entire Article






<