{"slug": "how-ai-fact-checking-is-changing-politics-for-better-and-for-worse", "title": "How AI fact-checking is changing politics — for better and for worse", "summary": "A study of more than 15 large language models by Dr. Filippo Menczer of Indiana University Bloomington found that AI fact-checking tools are less accurate than human fact-checkers but can achieve 98-99% accuracy when given reliable sources like PolitiFact, raising the question of whether the trade-off between scale and accuracy is worth it. Meanwhile, USC student Risha Panigrahi created InTruth, a Google Chrome extension that uses AI to fact-check politicians in real-time during debates and speeches.", "body_md": "Artificial intelligence is quickly infiltrating every aspect of modern life, and politics is no exception.\n\nAI technology is now widely used to fact-check representatives and politicians, sometimes in real-time during debates and public addresses. Yet as the influence of AI expands, so too does the debate about the technology’s accuracy.\n\nChatbots and large language models are known for their hallucinations — a weakness that poses particular risk when the model is tasked with verifying or disputing statements made by some of the most powerful people in the country.\n\nThere is also concern about bias in AI, both in terms of [politics](https://news.stanford.edu/stories/2025/05/ai-models-llms-chatgpt-claude-gemini-partisan-bias-research-study) and other factors such as [religion](https://www.deseret.com/faith/2026/05/26/studies-find-religious-bias-in-ai-models/).\n\nAhead of the midterm elections, it’s a bad time for a politician’s reputation and platform to be put on the line because of a digital error. However, the question about whether or not to use AI as a fact-checking tool may be about more than just accuracy, experts say.\n\n“The question is,” said Dr. Filippo Menczer, a professor at the Luddy School of Informatics, Computing, and Engineering at Indiana University Bloomington, “what accuracy does the AI model have to achieve, even if it’s not perfect, so that it’s better to use it than not to use it at all?”\n\n### What do we know about using AI as a fact-checker?\n\nMenczer has been studying the accuracy and influence of AI as a fact-checking tool for years, and his research has revealed both advantages and drawbacks to implementing AI tools as “de facto fact-checkers.”\n\nIn a recent study of more than 15 large language models, the ability of the models to accurately identify true or false claims made by politicians was limited, even if the models were equipped with advanced reasoning and web search capabilities.\n\nMenczer and his team found the models were not seeking out verified sources to supply their generated answers on their own. To improve accuracy, Menczer and his team included links to PolitiFact, a fact-checking website, in their prompts. When the model was given reliable sources, it worked “really, really well.”\n\n“The accuracy can be very, very, very, very high, like 98-99%, if you give the right sources,” Menczer told the Deseret News.\n\nBut if ensuring accuracy requires human verification anyway, then why make AI the middle-man?\n\nWell, supporters say, accuracy isn’t the only factor to consider.\n\nThough verification done by professional fact-checkers is often more accurate, AI models outperform their human counterparts when considering their capacity to operate quickly and reach a broad audience. Human fact-checkers can only fact-check “a very, very tiny, tiny fraction of all the claims that are out there,” Menczer said. AI models, on the other hand, can operate on a greater scale.\n\n“So then the question is not so much which one is more accurate. Of course humans are more accurate,” he continued. “The question is, should we never use AI because it can be less accurate? The answer to that one is very delicate, because there’s an issue of scale, right?”\n\nThe low cost, constant availability and ability of AI to easily expand across all countries and languages is not something to ignore, Menczer said.\n\nHowever, he also noted that scale does not matter if accuracy is completely out the window.\n\n“It doesn’t matter that you get an answer very quickly if the answer is wrong, right?” Menczer said. “You have to consider the scale and the accuracy together and then decide whether the trade-off is worth it.”\n\n### Innovations in real-time AI tools are changing the field of political fact-checking\n\nRecently, a USC computer science student, Risha Panigrahi, created an AI program, InTruth, that fact-checks politicians in real-time during debates, speeches and other public addresses.\n\nInTruth is a Google Chrome extension that uses AI to evaluate claims and label them as “true”, “false” or “unverifiable”.\n\n“I want this to remediate the information environment that’s so fraught with, you know, misinformation and lies,” Panigrahi said in an interview with the Deseret News.\n\nThe product illustrates the global scale of AI fact-checkers referenced by Menczer. Though InTruth is only a few weeks old, it already has 23,500 average weekly users across the world, with concentrated use beyond just the U.S. and into France, India and Russia, Panigrahi said.\n\nThe instant verification that InTruth provides is also beyond the scope of what a human fact-checker can supply, but Panigrahi said InTruth is “not meant to be a total replacement for manual or traditional fact-checking.” Rather, it serves as an “instant assistant” that helps “democratize” access to fact-checking sources, demonstrating the continued balancing act of accuracy and scale as this technology develops.\n\n### AI models will likely be used for fact-checking ‘no matter what’\n\nThe “need to build up a better understanding of that trade-off,” Menczer said, becomes particularly important as AI becomes more prominent in modern society.\n\nRegardless of accuracy and scale, people will use AI for fact-checking purposes “no matter what” simply because it is easily accessible.\n\nStill, that isn’t necessarily a bad thing, Menczer added.\n\n“I’m not going to say ‘no, people should not use AI models,’” he said. “I’m going to say people should be very careful about when they use AI models because they should be aware that the answer that they get may not be true.”\n\nWhen asked whether AI will get to a place where accuracy is no longer a major concern, Menczer said it is hard to tell given there is little data available to explain how the platform performs, but achieving that accuracy is not out of the question.\n\n“Fact-checking could be one area of application of AI models that is hugely helpful because of the scale. We just need to make sure that the accuracy is high enough. And we know that technically, that high accuracy is possible.”", "url": "https://wpnews.pro/news/how-ai-fact-checking-is-changing-politics-for-better-and-for-worse", "canonical_source": "https://www.deseret.com/business/2026/08/11/college-student-develops-political-debate-fact-checker/", "published_at": "2026-08-11 16:00:00+00:00", "updated_at": "2026-08-11 16:21:43.834408+00:00", "lang": "en", "topics": ["artificial-intelligence", "large-language-models", "ai-ethics", "ai-tools"], "entities": ["Filippo Menczer", "Indiana University Bloomington", "PolitiFact", "Risha Panigrahi", "InTruth", "USC", "Google Chrome"], "alternates": {"html": "https://wpnews.pro/news/how-ai-fact-checking-is-changing-politics-for-better-and-for-worse", "markdown": "https://wpnews.pro/news/how-ai-fact-checking-is-changing-politics-for-better-and-for-worse.md", "text": "https://wpnews.pro/news/how-ai-fact-checking-is-changing-politics-for-better-and-for-worse.txt", "jsonld": "https://wpnews.pro/news/how-ai-fact-checking-is-changing-politics-for-better-and-for-worse.jsonld"}}