A few years ago, a video clip was strong evidence that something really happened. Today, that assumption is breaking down. Cheap, easy tools can turn a photo or a line of text into realistic footage, and social media feeds are filling up with clips that never happened at all.

A good example came during this summer's FIFA World Cup. Videos showing young women cheering in the stands drew hundreds of millions of views, according to an investigation by Belgian broadcaster RTBF. Its fact-checking service examined one of the most shared clips, which claimed to show a Brazilian fan at the Brazil versus Morocco match, and found dozens of similar videos that appeared to be made with AI. The original account behind the clip, which had more than 340,000 followers, appeared to be an AI-generated influencer.

The tell-tale signs

RTBF spotted several giveaways in that clip, and they are useful for anyone:

  • Details that do not change when they should, such as a match clock frozen in place.
  • Logos or branding that do not fit, like a broadcaster that never aired the event.
  • Flat, robotic sounding voices or commentary.
  • Accounts that post huge volumes of similar, highly polished clips.

A detection tool used by the fact-checkers rated the footage as 99.9% likely to be AI-generated. Detection tools are helpful but not perfect, so it is best to combine them with common sense and a search for the original source.

The problem extends to politics. Earlier this year, researchers identified 304 AI-generated political accounts on TikTok, with some posts reaching more than 500,000 views and each video estimated to cost just $1 to $3 to produce, according to The Decoder.

Platforms say they are responding. YouTube chief executive Neal Mohan has said AI-generated content on YouTube will be labelled and that the company is "actively building on our established systems" to reduce low-quality AI content. He has also argued that AI is "not a replacement" for creativity. At the same time, YouTube is adding more AI tools for creators, so the volume of synthetic video is only likely to grow.

The best defence is a habit: pause before sharing. If a clip seems designed to shock, flatter or enrage, check who posted it first and whether trusted news outlets are reporting the same thing.

Photo: Shixart1985, CC BY 2.0, via Wikimedia Commons