Chouhan said journalists are increasingly relying on artificial intelligence to assist their fact-checking efforts. He noted that key technologies being employed include machine learning algorithms, natural language processing, and image, audio, and video recognition.
Chouhan, who was speaking at WAN-IFRA’s Bangalore AI Forum, outlined several reasons why AI is becoming essential in fact-checking:
- Sheer volume of claims and misinformation
- Breaks down complex patterns easily
- Eventually more accurate than the human eye: “It is nearly impossible for a human being to keep a tab on everything. Eventually, when properly trained, AI will be more accurate than the human eye – it already is in many aspects and will continue to be so in the future.”
- Real-time fact-checking speed: “We now have the capability to fact-check even as TV news is live – as the speaker is on air,” he explained.
- Tireless, unbiased work
- Endless formats and ways to verify information
Real-world use cases of AI in fact-checking
Chouhan outlined three key examples of how newsrooms are deploying AI to tackle misinformation head-on. These cases demonstrate how AI is being utilised to monitor conflict zones, track statements in real time, and enhance fact-checking efficiency – addressing the growing misinformation in news.
Faktisk Verifiserbar, Norway
Chouhan cited the Norwegian fact-checking cooperative Faktisk Verifiserbar, which is leveraging AI to monitor conflict zones.
“They focus on fact-checking conflict areas using OSINT techniques,” he said, pointing to their work tracking troop movements and artillery during the Ukraine war and later the Israel-Gaza conflict.
The organisation employs a combination of open-source tools such as Google Reverse Image Search, Google Earth Pro, and Google Maps. In one instance, they debunked a false claim about a US missile interception in Iran.
“They looked at Google Earth Pro, various reverse image searches, and combined it with reporter verification to provide a quick update that the story was fake,” he explained.
Faktisk also uses commercially available services like GeoSpy, which matches unique features from photographs against geographic databases.
Additionally, Chouhan said, the organisation has partnered with AI researchers at local universities to develop tools such as a Tank Classifier – capable of identifying and classifying tanks and artillery – and a video verification dashboard that automates the detection and processing of battlefield imagery.
Full Fact, UK
Another example Chouhan shared was Full Fact, a UK-based non-profit that received a Google grant to expand its AI fact-checking efforts.
Their system monitors live TV, online news, and social media to identify statements that could be fact-checked.
“The system looks at claims or sentences which can be called claims and flags those to journalists for verification,” Chouhan explained.
The tool cross-references statements against an extensive database of previous fact checks and data from the UK’s Office for National Statistics, helping to speed up verification without starting from scratch.
“They have a live page where they keep checking what’s happening in Parliament constantly, which is very good for democracy actually,” he added.
Factly, India
Chouhan also highlighted Factly, a Hyderabad-based fact-checking and public data research company that has been leveraging technology since 2016.
“They claim to have published around 10,000 fact checks in English, Telugu, and Kannada,” he said, noting that the organisation focuses heavily on data journalism, supported by a dedicated in-house tech team.
Factly has developed two AI-based products trained on its own journalism and public datasets: Dataful and Tagore AI.
Dataful is a portal containing over 20,000 high-value public datasets relevant to various Indian sectors, drawn from government and other public sources. It is offered as a paid tool.
Tagore AI is a chatbot still under development, designed to answer verification queries based solely on Factly’s own verified content. “There is very little chance of hallucination because they are using their own journalism for training,” Chouhan said.
AI as support, not replacement
Despite its promise, AI-driven fact-checking still faces significant limitations, Chouhan cautioned.
- Contextual understanding: AI struggles with satire, sarcasm, and nuanced political claims.
- Language bias: Most AI fact-checking tools are trained on English and Western datasets. “That’s a big problem for language markets like ours,” Chouhan noted. For instance, tests by independent organisations, such as Rest of World, revealed poor performance by systems like ChatGPT when responding in languages such as Bengali, Swahili, Urdu, and Thai.
- Bias in image recognition also remains a concern. Chouhan pointed out that even image-based generative AI systems show tendencies toward stereotypes and struggle to accurately identify non-white skin tones.
- Further, many AI fact-checking tools operate as black boxes, lacking transparency and accountability.
Chouhan emphasised that while AI offers significant support, it should not replace human fact-checkers. “First-level filtering can be done through AI. It is being done all over the world, and I think we can also do that,” he said.
He stressed the importance of continuous improvement of AI training datasets to reduce bias and maintain quality. “We need to keep a close watch on where it’s going,” he added.
