“Who is the lady in the red dress,” an X user asked xAI’s chatbot Grok on a photo showing filmmaker Pooja Bhatt, and actor Alia Bhatt with journalist Rana Ayyub—the latter being the ‘lady in the red dress’.
Grok’s response erroneously called her Jyoti Rani Malhotra, a YouTuber arrested on accusations of spying for Pakistan during Operation Sindoor—India’s retaliatory strikes in Pakistan in response to the attack on tourists in Kashmir’s Pahalgam.
Go to X and type in “Is it true (@grok)”—you will find hundreds of users asking the chatbot to verify some information or the other.
It is evident that Grok has become a destination for information verification, and tech policy researcher Prateek Waghre believes that this, “misplaced belief in their (AI chatbots) ability is adding to the chaos in an already dysfunctional information ecosystem.”
When Grok Got It Wrong On Operation Sindoor
On May 7, 2025, as India launched Operation Sindoor, news outlets and social media alike were awash with misinformation. Bereft of reliable sources of information during the fog of war, social media users turned to Grok for fact-checking.
While many of its ‘fact-checks’ were accurate, BOOM found a number of inaccurate responses by the chatbot that severely challenged Grok’s credibility as a go-to tool for verification.
Apart from misidentifying Rana Ayyub as Jyoti Malhotra (archived here), Grok also claimed that two digitally altered photos of Congress leader Rahul Gandhi, purportedly posing with Malhotra, were genuine (archived here). BOOM fact-checked this claim, and found that one of the photos showed Gandhi with Uttar Pradesh MLA Aditi Singh, while the other showed him with a supporter during the Bharat Jodo Yatra.
A deepfake video of the Director General of Inter-Services Public Relations (ISPR) of the Pakistan Armed Forces, went viral with the claim that Pakistan has admitted to losing two fighter jets.
Professor Hany Farid, a forensic expert in synthetic media at UC Berkeley confirmed to BOOM that the viral clip is a deepfake. However, when a user tagged Grok in one of the posts (now-deleted) that shared the video, the chatbot responded (archived here) saying, "There is no evidence suggesting it is AI-generated."
Similarly, it misidentified a video game clip of a dogfight between fighter aircraft as a real footage from Operation Sindoor (archived here), and an old video of wildfires in Chile as India’s attack on the Pakistani city of Sialkot (archived here).
Why Grok's Fact-Checks Cannot Be Trusted
While Grok can get many of its responses right, Waghre believes that this is incidental. He notes, "The way LLMs works, there isn't a concept of adherence to the truth or facts, or even that the responses have to necessarily be meaningful."
The above examples from Operation Sindoor support Waghre's statement—highlighting that Grok can definitely get it wrong.
The fact-checking industry at large rely on the use of consistent and transparent methodologies for fact-checking. This entails not relying simply on 'reliable sources' and independently verifying information through such replicable methodologies. BOOM's fact-checking methodology can be viewed here.
In contrast, Grok's responses rely entirely on sources available on the internet, and lacks any form of independent verification.
Waghre adds that LLMs are useful in low-risk, easily correctable settings, but becomes problematic when deployed at population scale without proper understanding of how it arrives at its responses. "It is not possible for them to generate responses with reliable facts where they don't exist, which is often the case in emergent, real-time scenarios," he points out.
And even Grok agrees with him. We queried the chatbot on its fact-checking methodologies, to which it clarified that its "accuracy depends on available online data, which may include errors or biases."
"For complex topics, I may miss nuances, so I encourage users to verify critical information independently," the chatbot responded to our query.
