Meta says one of the year’s biggest fears about generative AI and elections did not materialize at scale on its platforms. According to the company, AI-generated election misinformation remained a small share of the misinformation it fact-checked across Facebook, Instagram, and Threads.
The company’s review covered major elections in the U.S., Bangladesh, Indonesia, India, Pakistan, the EU Parliament, France, the U.K., South Africa, Mexico, and Brazil. Meta said there were confirmed or suspected uses of AI in election-related content, but that the overall volume stayed low.
What Meta says it found
At the start of the year, concerns were widespread that generative AI could be used to interfere in global elections by spreading propaganda and disinformation. By the end of the year, Meta said those concerns had not played out in a major way on its own apps.
The company said its existing policies and processes were enough to reduce the risk around generative AI content during the elections it reviewed. Its central claim was specific: ratings on AI content related to elections, politics, and social topics represented less than 1% of all fact-checked misinformation during those election periods.
That does not mean AI was absent from election misinformation. Meta acknowledged that there were cases where AI was either confirmed or suspected to have been used. But its conclusion was that, on Facebook, Instagram, and Threads, the amount of such content remained limited compared with the broader pool of misinformation that fact-checkers rated.
How Meta tried to limit deepfakes
Meta also pointed to controls inside its Imagine AI image generator. In the month leading up to election day, the company said the tool rejected 590,000 requests to create images of President-elect Trump, Vice President-elect Vance, Vice President Harris, Governor Walz, and President Biden.
The purpose of those rejections was to prevent users from creating election-related deepfakes. In Meta’s telling, this was one part of a broader effort to keep generative AI tools from being used to create misleading political images during a sensitive period.
The number is notable because it shows demand for political image generation was not theoretical. Users did try to create images involving major political figures. Meta’s position is that its systems blocked those attempts often enough to keep the impact limited on its own services.
Covert influence campaigns still mattered
Meta’s report was not only about individual AI-generated posts. The company also discussed coordinated networks of accounts that were trying to spread propaganda or disinformation.
According to Meta, those networks made only incremental productivity and content-generation gains from generative AI. In other words, the company said AI may have helped some operators produce or manage content more efficiently, but it did not fundamentally change Meta’s ability to find and remove those campaigns.
Meta said its enforcement approach focuses on account behavior rather than only on the content being posted. That matters because a network can still be detected by how it operates, even if some of its posts are generated with AI. The company said AI use did not prevent it from taking down covert influence campaigns.
Meta also said it removed around 20 new covert influence operations around the world to prevent foreign interference. It added that most of the networks it disrupted did not have authentic audiences, and that some used fake likes and followers to appear more popular than they really were.
Where the company pointed next
Meta also said false videos about the U.S. election linked to Russian-based influence operations were often posted on X and Telegram. That claim shifts part of the discussion beyond Meta’s own platforms, while still keeping the focus on how election-related falsehoods move across online services.
The company did not present its findings as the end of the issue. It said it would keep its policies under review and announce any changes in the months ahead.
For now, Meta’s message is narrow but important: generative AI was present in election-related misinformation, yet it did not dominate the misinformation landscape on Facebook, Instagram, and Threads during the major elections the company examined. The bigger enforcement picture still included propaganda networks, fake engagement, foreign interference concerns, and false videos moving across other platforms.
That leaves a practical takeaway. AI election misinformation remains a risk, but Meta says the most visible impact on its apps was limited, and that behavior-based enforcement remained central to detecting covert influence operations.