The Impact of AI Video Generators on the Credibility of Human-Made Videos

An AP Research Academic Paper (4,655 words) examining how AI video generators have affected the perceived authenticity of real, human-made videos. The paper opens by framing AI through Cole Stryker's IBM definition and the Federal Reserve finding that 34% of working-age adults aged 18–34 already use generative AI, then narrows in on AI video apps like Sora, whose realism is now close enough to real footage to create genuine confusion about authenticity.

The literature review is organized around three lenses. Celebrity portrayal covers Sora 2's 1M downloads in five days, viral deepfakes of Martin Luther King Jr. (which prompted OpenAI to pause his likeness), and Zelda Williams' public plea to stop generating videos of her late father Robin Williams. The entertainment section uses Grand View Research and Morgan Stanley data — a $25.98B AI media market in 2024 projected to $99.48B by 2030, and ~10% potential cost reduction for studios — alongside The Brutalist and Emilia Perez as examples of AI-assisted Oscar-caliber work. The social media section draws on Coolbear, Majerczak & Strzelecki, and Granicus to argue that authenticity is now a survival metric for creators.

The methodology is an original survey study distributed to peers and a wider audience, with results presented across multiple figures and tables — including how often respondents encounter AI-generated video, where they most often see it, whether they have ever mistaken AI video for real footage, and a side-by-side image-identification task where 82.1% correctly picked the real image over the AI one. Open-ended responses surface concerns about deepfakes in politics, blurry-edge tells viewers now actively look for, and recommendations for clearer AI-content labeling.

The discussion ties the survey results back to the literature: AI video is already eroding default trust in human-made content, and the burden of proving authenticity is shifting onto creators, platforms, and public figures. The paper closes by recommending stronger provenance signals and continued public education on the visual cues that distinguish AI-generated video from real footage.