Is That An AI Video? Here’s How Newsrooms Can Spot Them

Remember when we laughed at AI-generated images with six fingers? That was so last year. Today, video generators like OpenAI’s new Sora 2 model create clips so realistic they’re fooling millions daily, including newsrooms. A concerned community member or a prankster? The difference matters, and spotting it requires new verification protocols.

Columbia Engineering researchers recently developed DIVID, a detection tool achieving 93.7% accuracy identifying AI videos from platforms like Sora, Pika and Runway Gen-2. But journalists can’t wait for technology to catch up. The implications for newsroom operations are immediate and significant. 

Ways Newsrooms Can Identify AI-Generated Video

1. Check the video quality, especially if it’s bad.

A recent viral video of rabbits jumping on a trampoline got hundreds of millions of TikTok views. It looked like cheap security camera footage. It was completely fake.

“Grainy, pixelated footage should trigger immediate skepticism. Nearly everyone carries a smartphone with a 4K camera today, so something that looks low fidelity should give you zero confidence,” says Lindsay Stewart, CEO of the video marketplace and technology company, Stringr.

Low resolution hides AI’s subtle flaws: uncanny skin textures, shifting hair patterns, or background objects moving impossibly. AI creators deliberately downgrade quality to obscure these artifacts.

Stewart adds you can “reach out to the main provider and license it, use third-party services like Storyful to clear UGC. There are so many good ways to know if something is real. Your gut should never be driving your decision to clear a video.”

2. Watch for visual glitches and continuity errors.

AI-generated video struggles with consistency. Frame-by-frame analysis reveals these subtle mistakes. A polar bear “rescue” video with over 9 million TikTok likes showed the cute bear cub morphing into a dog and sprouting extra paws for several frames. People’s faces in backgrounds sometimes garble. Objects flicker in and out of existence. Arms and legs may blur or morph slightly when moving.

These glitches expose AI generation tools, particularly in current diffusion models that struggle with object permanence across extended sequences. If you are struggling with native video players, several free tools enable downloading videos from TikTok, LinkedIn and YouTube for frame-by-frame analysis.

3. Notice the uncanny, hyper-realistic look.

Sometimes AI videos look too good. Poreless skin, movie-quality weather and oddly timed blinking expose AI-generated content. Videos appear hyperrealistic, almost animated, because they are. Yet millions still fall for them.

“If it’s too perfect, then it’s probably a synthetic,” notes Kerry Oslund, VP of AI at The E.W Scripps Co. “This is a tough spot for both newsrooms and for responsible AI video creators. There are so many use cases for marketers, but happily so few for journalists. My advice: When in doubt, toss it out.”

4. Check the length and pacing.

Most AI videos run 6 to 10 seconds, much shorter than typical human generated social media clips. “Today’s AI video generators limit their outputs to just a few seconds,” says Michael Vamosy, founder & chief creative officer at Defiant LA. “Even if there’s character consistency between clips, be suspicious of any edit sequences that have an edit every few seconds or so.”

Generating longer videos costs more and increases error likelihood due to computational constraints.

Try searching your social network of choice for “cat saves baby from bear” and you’ll find dozens of viral videos of heroic cats that were all recently posted after Sora 2 was released. Besides all being very short clips, there’s no telltale sign they were AI-generated. 

5. Listen for audio problems.

Lip-sync issues are common giveaways. Watch closely to see if mouth movements match speech sounds. Does the timing feel slightly off? Do mouth shapes align with words?

Background audio matters too. AI-generated clips often lack natural ambient sounds or echoes. Some have no sound at all. As text-to-video models evolve, these inconsistencies become critical detection points.

6. Study the source and context.

Multiple credible outlets or witnesses should confirm real footage. If every link traces back to one viral TikTok post, it’s probably fake. Check the account posting it. Mystery handles with no bio, history or location should be treated as fiction until proven otherwise.

“Remember the saying, ‘I’ll believe it when I see it’? Maybe it’s time we retire that,” says Nick Monico, COO at Adams Multimedia. “Videos now require the same source verification as written claims.”

7. Look for Watermark Removal

Many AI tools add watermarks, but creators scrub them out. Look for blurry patches, smeared light or soft squares near corners or edges. Days after Sora 2 launched, tutorials flooded TikTok, Reddit and YouTube showing watermark removal techniques.

This cat-and-mouse dynamic between watermarking and removal presents ongoing challenges for platform-level authentication systems.

8. Is the lighting too perfect?

When a newsroom receives a “breaking” clip from an unknown source, the lighting can be a red flag. Real-world footage, especially eyewitness videos or security cameras, includes uneven light, inconsistent shadows and partially obscured faces.  If the scene appears perfectly balanced with evenly lit faces regardless of movement or environment, that may indicate a rendering engine rather than a camera lens.

AI models render subjects with soft, balanced illumination even when settings should be dark or unpredictable. Nighttime videos, surveillance clips or chaotic scenes almost never look pristine. If footage looks too perfect, it warrants closer examination.

9. Question the plausibility.

If it seems too funny, cute or amazing to be real, it probably is. Videos of babies walking runways or riding dangerous animals fooled many viewers, despite being physically implausible.

AI plays on emotions, especially with adorable, hilarious or shocking content. A fake video of a priest giving a surprisingly leftist sermon fooled even tech journalists. The emotional manipulation inherent in these videos exploits cognitive vulnerabilities that bypass rational analysis.

Why People Create Fake Videos

Research from the University of Amsterdam found people significantly overestimate their ability to detect deepfakes, with confidence remaining high even when accuracy is low. This overconfidence creates vulnerabilities that bad actors exploit through pranks, misinformation campaigns and sophisticated fraud schemes.

An employee at a multinational corporation sent fraudsters $25 million after receiving AI-generated videos of the company’s CFO. User-friendly apps now enable anyone to create video deepfakes, posing unprecedented verification challenges for all businesses with implications extending to platform trust and audience relationships.

Strategic Implications For Newsrooms

Here’s the bad news. All of the tips mentioned in this article have the shelf life of a bowl of tuna salad at a summer barbeque. AI video tools are evolving fast and these AI “tells” will all soon be eliminated. 

“We’ve now hit 95% photorealism with AI video,” explained PJ Ace, CEO of the viral AI ad company Genre.ai. “Within a few months, it will be nearly indistinguishable from real life.”

This timeline creates urgency for newsrooms to establish verification protocols now, before current detection methods become obsolete. Technology companies are developing new verification standards. Cameras could embed cryptographic information, like C2PA, proving content is real. AI tools could automatically tag synthetic content. But these solutions remain incomplete and face implementation challenges across both fragmented technology systems and our slow-moving media industry.

The strategic response requires multiple layers: technical tools, editorial protocols and audience education. Newsrooms need to invest in detection technologies like DIVID while simultaneously training staff on manual verification techniques. 

AI Video Detection: Action Items For Media Leaders

For now, skepticism is the new literacy. Assume every audience-submitted video is fake until credible sources verify it.

• Establish clear verification workflows for user-generated content
• Train all newsroom staff on AI video detection techniques
• Invest in technical detection tools as they become available
• Develop transparent disclosure policies for verification processes
• Build audience education into coverage of viral video stories

Look twice, think constantly and remember: Your journalistic instinct is the watermark that AI can’t erase. The preservation of newsroom credibility depends on maintaining verification standards even as new technology evolves to defeat them.

Disclosure: Adams Multimedia is a client of Ordo Digital.

The post Is That An AI Video? Here’s How Newsrooms Can Spot Them appeared first on TV News Check.