Newsguard debuts new automation tools for tracking election-related misinformation

Although the company has used aspects of automation before now, previous tactics have been much more analog, according to Matt Skibinski, NewsGuard’s general manager. New elements include both internally built features using open-source tools and data from external partners including media monitoring startups like Meltwater and NewsWhip. Skibinski said NewsGuard also is seeing a higher percentage of misinformation made from manipulated media like doctored images and audio clips.

“For each false narrative, we have data such as search terms that are designed to find content with the false narrative and avoid false positives,” Skibinski said. “We assemble a bunch of text excerpts and media excerpts of content conveying the false narrative … If we see we have a known false narrative that we’re tracking and we see a new publisher or network of publishers start to pick it up, that tells us we need to look at them.”

As NewsGuard identifies more misinformation, it’ll be added to the company’s “Misinformation Fingerprint” database, which is used to help track the sources of harmful content and using that data to see where similar content appears online. NewsGuard says it already is monitoring harmful content across hundreds of accounts associated with Russian, Chinese and Iranian disinformation operations. 

Features in the tracking center also include exclusion lists for advertiser that want to avoid election misinformation or hyper-partisan content, as well as inclusion lists for advertisers to appear alongside high quality news websites. NewsGuard’s data also includes credibility ratings for political news sources across podcasts, streaming platforms and linear TV.

NewsGuard also is tracking more than 700 AI-generated websites it’s classified as “unreliable AI-generated news,” or UAINs. Another category is what the company calls “pink slime” websites, which are funded by partisan groups but pretend to be local news outlets. So far, NewsGuard’s team has identified more than 1,000 websites funded by left- and right-leaning groups without disclosures — including dozens that are AI-generated.

“It’s undebatable that this will come up in a major way this election,” Skibinski said. “And the question is just how and where and who. And so that’s why we have to monitor kind of everything.”

To help advertisers navigate the election year, IPG-owned media agency Magna Global recently distributed an election-specific brand safety playbook to clients to help them prepare for the election year landscape. According to Sara Tehrani, vp of impact investment at Magna Global, election-related disinformation and AI-generated misinformation are both requiring advertisers to take a “deeper look” at their ad placements and partnerships.

“Social is a place where advertisers don’t want to pull away dollars from,” Tehrani said. “It’s the place they know they need to be [where] their consumers are; it’s just [about] finding ways to make sure they’re not showing up in that sort of dicey sort of content.”

AI-generated misinformation isn’t just coming from malicious actors. Popular chatbots have been found to give inaccurate info about questions related to voter eligibility, polling locations and identification requirements. In a study published this week, researchers examined election-related answers from five top AI language models — Anthropic’s Claude, Google’s Gemini, OpenAI’s GPT-4, Meta’s Llama 2, and the French startup Mistral’s Mixtral — and found more than half of the chatbots’ answers were inaccurate or harmful. The report was published by AI Democracy Projects in collaboration with dozens of state and local election officials, AI researchers, the nonprofit Proof News and the Princeton, NJ.-based Institute for Advanced Study.

Election-related misinformation is increasingly a concern in the U.S. and around the world. European officials recently accused Russia of spreading disinformation across social media and other internet platforms — and it’s been a theme around the U.S. elections for several years. AI-generated misinformation related and unrelated to elections also adds additional layers of risk. Earlier this month, nearly two dozen tech companies agreed this month to do more to address harmful election-related content around the world generated by AI. Signatories include tech giants like Google and Meta and startups like OpenAI and Anthropic.

In addition to hyper-targeted ads directing people toward disinformation, other concerns revolve around the chance of using hyper-targeted ads with unrelated creative. That could allow nation states to slip by online platforms’ current policies around political content, according to Mike Lyden, vp of threat intelligence at the Trustworthy Accountability Group, who spent most of his career in U.S. intelligence. 

“You [could] advertise a toolbox towards white men in Missouri over the age of 55, who are a highly likely demographic [to like] MAGA information,” said Lyden. “That click isn’t necessarily nefarious, doesn’t look like you’re going to a political site, but then you get them to a landing page through that.”