Quick Buzz Feed
AI imagery is no longer a hypothetical factor, but at the same time, we've been able to use AI in new ways ourselves to confront the challenge.
The article opens by illustrating a significant new challenge for fact-checkers: the proliferation of AI-generated content in political campaigns. It cites a specific example from the Scottish parliamentary elections where an independent candidate shared AI-generated campaign videos on Facebook, explicitly labeling them as “illustrative” of future aspirations rather than actual events. These videos, set to a soulful pop ballad, depicted the politician meeting voters, chatting with schoolchildren, and visiting a hospital – all entirely fabricated by artificial intelligence. This real-world instance from 2026 underscores that AI imagery is no longer a theoretical concern but an active factor impacting the integrity of election information, raising fresh and complex questions for both journalists and fact-checking organizations. This scenario highlights how easily sophisticated, yet entirely fictional, narratives can be constructed and disseminated, making it difficult for the public to discern reality from AI-created simulations, thereby emphasizing the urgent need for advanced detection and verification mechanisms.
In response to the evolving digital landscape, Full Fact, the independent U.K. fact-checking nonprofit, has positioned itself at the forefront of combating misinformation through technological innovation. The organization, comprising 34 individuals, includes a dedicated AI team of data scientists and software engineers who collaborate closely with its editorial team of eight journalists. Full Fact has been an early adopter of machine learning, integrating AI into its fact-checking processes since 2016. This long-term commitment has led to the development of "Full Fact AI," a sophisticated suite of tools designed to monitor and analyze claims from a diverse array of sources. These include traditional online news sites, popular social media platforms like Facebook, TikTok, X (formerly Twitter), and YouTube, as well as video platforms. The primary functions of these tools involve identifying new claims that warrant verification and efficiently detecting repeats of claims that have already undergone fact-checking. This technological backbone allows Full Fact to process an immense volume of information, such as approximately a third of a million sentences on a typical weekday, significantly enhancing its capacity to address misinformation. The efficacy and scalability of Full Fact AI are further demonstrated by its adoption by over 40 fact-checking organizations operating in three different languages across 30 countries, establishing it as a crucial resource in the global fight against disinformation.
Full Fact rigorously deployed its advanced AI tools during the recent busy election periods in England, Scotland, and Wales, significantly bolstering its monitoring capabilities. Leveraging data from Democracy Club, the organization initiated the monitoring of more than a thousand social media accounts across Facebook, TikTok, X, YouTube, and Instagram that were linked to various candidates. The AI tools played a critical role in this process by providing live transcripts of claims made in videos, which were then systematically matched against Full Fact’s extensive database of previously published fact-checks. This automated matching process allowed journalists to swiftly search for relevant claims and receive real-time alerts through an internal Slack channel, streamlining the workflow and ensuring maximal utilization of the gathered data. A particularly innovative application was the scanning of social media posts for evidence of SynthID, Google’s invisible digital watermark, indicating AI creation or editing. During the May elections, Full Fact scanned 16,514 images and videos attached to candidate posts, successfully identifying 136 instances with apparent watermarks. While many of these were non-controversial, like AI-generated images of future construction projects or infographics, some, such as the Glasgow candidate’s "illustrative" video, warranted deeper investigation and would likely have gone unnoticed without AI detection. Beyond direct fact-checking, this comprehensive AI monitoring provided Full Fact’s small editorial team with unprecedented visibility into online discourse, surfacing a broad spectrum of claims and posts that might otherwise have been missed. Furthermore, post-election, generative AI tools were employed to rapidly analyze over 33,000 posts from Scottish and Welsh parliamentary candidates, yielding unique insights into the campaign’s dominant topics, with the economy being a major focus and independence a significant issue in Scotland.
The experiences of Full Fact during these elections offer valuable strategic insights for other news organizations grappling with the pervasive challenges of AI-generated content, especially as journalists in the U.S. prepare for upcoming midterm elections. The core principle highlighted is the critical importance of integrating AI monitoring capabilities directly into editorial workflows. This integration allows smaller teams to significantly expand their coverage and monitoring capacity across the vast online landscape, making it feasible to track a much larger volume of information than traditional methods would permit. Full Fact’s success underscores the necessity of minimizing friction in these processes, ensuring that AI tools seamlessly assist human journalists rather than complicate their tasks. Crucially, the model emphasizes that while AI can amplify monitoring and detection efforts, human judgment remains indispensable. Journalists must stay "very much in the loop" to evaluate the relevance and veracity of AI-flagged content and make final decisions on what is worthy of attention and further investigation. This approach enables newsrooms to harness the power of AI to combat disinformation at scale while maintaining the essential human element of journalistic integrity and critical analysis.