When our team was selected for the JournalismAI Fellowship Programme, we were clear on one thing: we wanted to create a multilingual AI tool that could quickly detect potential repeated false claims made by politicians on social media.
We knew why we wanted to do this: to save our fact-checking teams time and effort. By using this tool, journalists and editors in our newsrooms would be able to quickly check certain claims, thereby improving accuracy and enhancing their reporting capabilities. But the how was more difficult to figure out: there was an ocean of possibilities and a lot of uncertainties.
This guide aims to share our path in making this tool so that others can learn from what we believe worked and what didn’t.
Table of contents
<aside> 🚀 Who are we
ClaimCheck is a tool created from the JournalismAI Fellowship Programme composed by journalists and engineers:
Gina McKeon, journalist and producer at ABC Australia
Irene Larraz, fact-checking team coordinator at Newtral
Rubén Míguez, CTO at Newtral
Gareth Seneque, machine learning engineer at ABC Australia
</aside>
Our initial idea was to create a multilingual tool for journalists that uses AI to quickly detect false claims by setting up an alert system through an interface in Teams or Slack. This system would become a central hub for newsrooms to quickly check specific claims with previous fact-checks. For the Fellowship, we decided to focus our prototype on Twitter and monitoring tweets from all sitting politicians in Spain and Australia as a sample group to test.
<aside> 📌 Our case: what, why, and how
Why did we choose politicians as the sample group? Politicians may repeat potential false claims. This can take time and resources from fact-checkers and journalists.
What is the problem? Politicians may rephrase or reformulate potential false claims, making it difficult for fact-checking technology to detect when they are making a similar claim.
How do you plan to solve this problem? We created ClaimCheck, a multilingual tool for journalists and editors that uses AI to quickly detect repeated potential false claims in three ways:
</aside>
Interest in fact-checking tools has gained traction at several organisations around the world. Fact-checking organisations are looking for ways to cross-reference corrections, such as Full Fact, while others are underlining the problems of repeated false claims, for instance, The Fact Checker, which established a new rating, the Bottomless Pinocchio, to highlight them. The New York Times and Washington Post have also reported on repeated false claims.
We found it essential to define the scope of what we expected to achieve from the Fellowship and how we planned to do this. We wanted to tackle potential repeated false claims from politicians for three reasons: to save time and effort; to scale the reach of verifications; to detect potential repeated misinformation strategies. We planned to do this by: