Home › Blog › Fact-checking

How to check if a viral breaking news video is AI?

A guide to verifying AI-generated footage of breaking news events using publication timelines, content credentials, and text-based fact-checking.

Fact-checking · · 8 min read · 1,730 words

W
wyper Fact-Check Team Builders of the wyper Fact-Check extension. Every claim in this article links to its source.
AI transparency

This article was researched and drafted with AI, then checked and released by a person before it went live. The illustration is AI-generated. Every factual claim links to its source, so you can verify it yourself.

A laptop screen displaying a social media video feed next to a timeline showing publication dates and metadata blocks. (AI-generated illustration)

AI-generated illustration. A laptop screen displaying a social media video feed next to a timeline showing publication dates and metadata blocks.

The short answer

You can check if a viral breaking news video is AI-generated by tracing its earliest publication date and looking for an intact content credential. A provenance check reveals whether the footage existed before the event actually happened.

When a major news event happens, social media feeds immediately fill with dramatic footage. Recently, the visual evidence following breaking news has increasingly included AI-generated clips designed to look like raw, on-the-ground recordings. Following a rare US aviation ground stop on November 13, 2026, AFP reported that videos circulated showing exhausted passengers packed tightly together and sleeping on airport floors. The footage was synthetic. A few days later, after an attack on Riyadh on November 18, 2026, Lead Stories debunked a viral video claiming to show severe damage to the PIF Tower. The clip was generated with AI, and no established media outlets had published it.

These cases highlight a shifting reality in digital information. As generation tools become cheaper and more accessible, relying on your eyes to spot visual errors is no longer enough. You need a structured verification workflow to determine where a video came from and what it actually shows.

Why pixel forensics fall short

People often try to spot AI videos by looking for weird hands, blurry text or impossible physics. While the fake US airport video did contain visual irregularities, relying entirely on your eyes is a failing strategy. Generation models improve constantly, smoothing out the obvious glitches that used to give them away. Furthermore, AI-driven disinformation is now cheap enough to produce and distribute at scale, as the Konrad-Adenauer-Stiftung noted in a recent guide for journalists.

The key question is no longer just whether a video looks fake, but where it came from and when it first appeared. A pixel detector cannot see the most common form of video deception: taking completely real, unedited footage from an old event and labelling it as a new one. Instead of staring at pixels, you need to establish the provenance of the file and compare its publication timeline to the actual event.

Tracing the earliest publication

The strongest piece of evidence against a fake breaking news video is time. If a video depicting the aftermath of a November 18 attack was uploaded to a forum in October, the footage cannot be from the event. As a verification guide from thenews.club explains, a viral video can be authentic and still be misleading if it is old, edited, translated incorrectly, or paired with a false location.

To find this timeline gap, you must run a reverse image search on still frames from the video. Pause the clip when a clear, distinctive landmark, face, or object is on the screen. Take several screenshots from different scenes. Upload these still frames to search engines to find the earliest documented publication. For videos, the provenance check takes still frames and searches each one separately. A hit on any frame is a hit on the video.

This event-versus-post date gap carries the most weight in video verification. It catches both AI-generated clips that were prepared in advance and recycled authentic footage that has been pulled out of context. If you locate a matching upload that predates the news event, the verification is complete, and the caption attached to the viral post is false.

Checking content credentials and metadata

The technology industry has developed a standard called C2PA to attach cryptographic content credentials to digital files. This acts as a digital signature, proving the origin of an image or video. A robust provenance report will check this signature against official trust anchors. A signature only counts if it comes from a trusted anchor, because a merely "valid" signature can be self-signed by anyone.

However, there is a counter-intuitive reality about video verification on social media that you must understand. A content credential sits in its own box beside the picture data. Any repackaging drops it. This includes a lossless repackaging that leaves every single pixel untouched. Because platforms repackage files automatically upon upload, a video taken off a feed will almost always lack a credential. That is the normal case, not a suspicion. To check the signature, you must track down the original file where the credential might still be intact.

Verifying the claim with a text check

A video never exists in a vacuum. It is always accompanied by a caption or a spoken claim, and you can verify these claims independently of the video file itself. If a video claims to show a specific building damaged on a specific date, you can check news reports, official statements and public records to see if the event actually occurred.

When running an automated fact-check, tools search for evidence related to the textual claim. In our own published measurements of 633 real citations re-fetched, we found that 73 percent were verbatim on the page. We also found that 19 percent of pages were unreachable for automated readers due to paywalls or formatting, and 6.4 percent were not findable. Reaching primary sources is critical. In a separate test of 24 claims with known roots, our system reached primary sources 63 percent of the time and legacy media 14 percent of the time, successfully reaching all non-English roots.

If a video has a spoken voiceover, automated tools can use the subtitles if they are reachable. The web app cannot see your YouTube session, so subtitles are often unavailable there. The extension runs in your browser and can reach more. When subtitles are unavailable, the AI model actually watches the video within a time window. It watches up to 10 minutes on the free tier and 30 minutes on Pro for YouTube. For X videos, it watches up to 2 minutes on the free tier and 7 minutes on Pro, while longer ones are refused outright rather than partially checked. It does not run deepfake pixel forensics, and it does not judge the intent of the creator.

For a deeper analysis, the Full Spectrum check uses a Triple-AI reading: three models, one evidence set. The third voice does not search. It receives exactly the sources the main model received and reads them independently, so a disagreement is a different reading of the same material, not a different search result. We added this because two models trained on overlapping text are correlated. In our tests with frozen sources, Gemini and Grok agreed in 29 of 29 claims. A third model of different origin disagreed in 3 of 26, every time with a machine-verified quote.

Comparing your verification options

You can verify videos manually, or you can use tools to speed up the process. The free wyper web app allows you to run text-based fact-checks on the claims attached to a video without installing anything. For video provenance, the AI Spotter tool uses still frames to search for earliest appearances and checks metadata. The AI Spotter is a paid feature because a provenance report calls three paid services on every single run, costing 4 credits per video.

Below is a breakdown of how the different approaches compare.

FeatureManual checkswyper web appwyper extension
Text fact-checkingFree, requires manual searchingFree tier available (5 checks a day)Free tier available (5 checks a day)
Video provenance (AI Spotter)Free, requires reverse searching framesPaid feature (4 credits per video)Paid feature (4 credits per video)
Access to browser timelineN/ANo, requires pasting links or textYes, checks videos where you read them
Subtitle extractionRequires manual transcriptionLimited by platform blocksHigh, runs in your logged-in browser
Evidence chain providedYou build your ownYes, real linked sourcesYes, real linked sources
Privacy of dataNo data leaves the browserData sent to servers for checkingData sent to servers for checking

Who needs which: Manual checks are best for users who want total control over their research and prefer not to send data to external verification services. The web app is ideal for readers who want a quick text fact-check or a provenance report without installing any software. The browser extension is built for users who want to check videos and posts directly inside their social media feeds without switching tabs.

Does a missing content credential mean a video is fake?
No, a missing content credential is the normal state for videos downloaded from social media. Platforms automatically repackage files during the upload process, which strips the signature completely. To find an intact credential, you must locate the original file before it was uploaded to a social feed.
Can AI detect if a video is a deepfake by looking at pixels?
Pixel forensics cannot reliably prove a video is a deepfake or authentic. AI generation models improve constantly, making visual artifacts harder to spot. Furthermore, a pixel detector cannot see the most common form of deception: taking completely real footage from an old event and labelling it as new.
How does wyper fact-check a video without subtitles?
When subtitles are unreachable, the wyper AI model actually watches the video within a set time limit. It watches up to 10 minutes on the free tier and 30 minutes on Pro for YouTube. It evaluates the visible claims against real evidence without running pixel forensics or judging intent.
Why do two AI models sometimes agree on a false claim?
Two models trained on overlapping internet text are correlated, meaning they share the same blind spots and biases. In our tests, Gemini and Grok agreed in 29 of 29 contested claims. A third model of different origin is required to provide a truly independent reading of the same evidence.
Does the wyper free tier include the AI Spotter?
The AI Spotter is not included in the free tier for images or video. Running a provenance report calls three paid external services on every single run, requiring 4 credits per video. The text-based fact-checking tool remains completely free for up to five checks a day.

The short version

Verifying a viral breaking news video requires tracing its earliest publication date rather than searching for visual glitches. By checking the gap between the event and the upload, you can quickly spot AI-generated fakes and recycled footage. Tools can automate the provenance research, but the final judgment on trustworthiness always remains with the reader.

Useful? Pass it on: 𝕏 Post it Telegram

Check posts where you read them

The wyper Fact-Check extension runs these checks on the post itself: truth score, evidence chain, and the date gap that catches recycled footage.

The web app installs to your home screen in one tap. No store, no account.
Already using it? ★★★★★ Rate it on the Chrome Web Store. Reviews decide what others get shown first.

Keep reading

Fact-checking · How do I verify where a video was filmed? Learn how to verify video location using visual clues like street signs, architecture, sun angles, and map comparisons to spot relabelled footage. Fact-checking · How to verify a viral emergency warning video? Learn how to verify viral emergency alerts, blackout warnings and manipulated news clips using reverse image searches and provenance checks.

Guides and tools

AI Fact-Check A 1 to 10 truth score for any claim, page, image or video, with the sources. Fact-checking tools compared Snopes, Ground News, Verafy and wyper: which does which job. How to fact-check a video or claim The method, step by step, before you share. Quote study: 633 real cases How often the quoted sentence really appears in the credited document.