Thai PBS Verify Releases Guide to Spotting AI Fake Videos
Thai PBS Verify has published a comprehensive guide for identifying AI-generated fake videos, which are becoming increasingly sophisticated and widespread. The guide categorizes fake content into three types — fully synthetic, manipulated authentic footage, and real footage used out of context — and outlines practical verification steps including context checking, reverse image searches, and audio analysis. Experts warn that detection tools alone are insufficient as AI technology evolves rapidly, emphasizing the need for critical thinking before sharing sensational clips.
Fake news used to be crude text and poorly edited images, but now it arrives as realistic-looking clips that appear to be filmed on a bystander's phone with voices mimicking real leaders. Thai PBS Verify found a clip claiming Iran shot down a B-2 bomber and another alleging protests against a Philippine flood-control project, both AI-generated and widely shared before being debunked. Such clips are dangerous because the brain trusts visual evidence before critical thought, and fake content is engineered to trigger emotion and prompt sharing before verification.
AI-generated fake news is not one single type. Before fact-checking, you must classify the category because verification methods differ. The first type is video or audio created entirely by AI with no real event behind it. The second is authentic footage that has been manipulated — dubbed with new audio, edited to reverse meaning, or deepfaked with another person's face. The third type, more common than many realize and requiring no AI at all, is genuine footage used out of context. Fact-checkers once found a clip presented as a disaster on the Nepal-China border that was actually a reversed glacier collapse in Russia mixed with a 2021 Japanese landslide.
When encountering a shocking clip, always ask two questions first: "Did this really happen?" and "Is it happening right now, right here?" Verify the broader context before analyzing the clip itself. The most effective first step is comparing the clip against real-world context. In a clip claiming Iran attacked Israeli cities, posted on June 10, 2026, both sides had announced a ceasefire on June 9 and calm had returned; no weapons fell as the clip alleged. Check whether the date and time match real events, whether credible news agencies or official bodies have confirmed it, who the original poster is, whether the account is newly created or has a history of sensational posts, and whether other angles of the same event exist — major real events usually have multiple perspectives, while fakes often appear as a single, suspiciously perfect clip.
After context-checking, examine the clip itself. Grab key frames and reverse-image search for the original. Look for unnatural objects, unreadable signs or text, crowds with identical faces or repetitive movements, inconsistent shadows and lighting, and rapid cuts that prevent close scrutiny. Audio is harder to verify without visual cues. Be wary of "leaked" audio of unknown origin, speech that is too smooth, lacks ambient noise, and compare it with the person's full original footage. However, do not rely solely on watermarks or detection tools. Thai PBS Verify's 2025 review noted that AI has improved so fast that long-standing detection tools lag behind new technology. In the cases above, verifiers used Hive Moderation, which flagged a 99.1% probability of AI generation, as just one piece of evidence alongside timeline verification. Watermarks and provenance data like Content Credentials or SynthID carry high weight when present, but their absence does not prove authenticity because such data is easily lost through recompression or screenshots.
Ultimately, fake news spreads fast, but real news
