Learn how to check if a viral video is altered, why pixel detectors fail, and how tracing the earliest appearance uncovers the truth behind doctored clips.
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.
AI-generated illustration. A computer monitor displaying a video timeline with a magnifying glass icon hovering over a freeze frame.
You can verify an altered video by checking its earliest documented publication date and verifying the claims made within the audio. Pixel detection is often unreliable, so tracing the source and cross-checking the actual quotes provides the strongest evidence of authenticity.
Mid-September 2026 has brought a massive wave of altered videos to social media timelines. As Journo News reported, professional fact-checking organizations are straining under a volume of misinformation they simply cannot keep up with. The gap between claims requiring verification and available human capacity keeps widening. This means you have to verify footage yourself before you share it.
Recent weeks have provided several clear examples of how video manipulation spreads. In one instance, a video circulated online purportedly showing a reporter questioning Ministry of External Affairs spokesperson Randhir Jaiswal about a viral clip of him eating snacks during the BRICS Summit. As Newschecker reported, this was an AI-altered video. No reporter asked such a question during the press briefing.
Similarly, another altered video appeared to show Ambassador Harish Parvathaneni, India's Permanent Representative to the United Nations, dismissing Amnesty International's allegations concerning arms exports to Israel. As Newschecker confirmed, this footage was manipulated to put false words into the envoy's mouth.
Sometimes the manipulation is purely visual. A video recently went viral showing a woman bathing in the Ganga river in a bikini. Alt News investigated the clip and found it was created by altering an original Instagram video, digitally changing the woman's yellow clothing into a bikini.
Conversely, genuine footage is frequently dismissed as fake. Social media users recently flagged a video of actor Mel Gibson criticizing politicians as an AI-generated deepfake. However, Lead Stories verified that the footage is completely authentic and was originally uploaded to the internet in the summer of 2025. The claim that it was AI-generated even appeared in a community note on X, demonstrating how easily the crowd can get it wrong.
When faced with a suspicious video, people often want a tool that outputs a percentage of how "AI-ish" the footage looks. That number is the exact harm the wyper AI Spotter exists to prevent. The tool is a provenance report, never a detector, and it does not run deepfake pixel forensics.
The AI Spotter builds its report using layers of verifiable data: content credentials (C2PA), file metadata, reverse image search, earliest documented publication, and quote verification. For video verification, the tool takes still frames and searches each one separately. A hit on any frame is considered a hit on the video.
When checking a video taken off a social feed, the content credential layer will almost always find nothing. A content credential sits in its own box beside the picture data. Any repackaging drops it entirely, including a lossless repackaging that leaves every pixel untouched. Since platforms automatically repackage files upon upload, a missing signature is the normal case, not a suspicion. The report explicitly says so, and we advise checking the original file instead where the signature can still be intact.
What actually carries the weight for video verification is the earliest appearance combined with the event-versus-post date gap. That is what catches the most common real video deception: old footage relabelled as new. A pixel detector cannot see that at all, which is why tracing the origin date is critical for verifying older clips like the Mel Gibson video.
Because a provenance report calls three paid services on every single run, there is no cheap first look to give away. The AI Spotter is not in the free tier for images or video. It costs 1 credit per image and 4 credits per video, available on paid plans only. This is entirely separate from the text fact-check, which remains free.
If the visuals are authentic, you still need to verify the claims spoken in the audio. The free wyper web app handles text and audio claims by generating a truth score of 1 to 10 alongside an evidence chain of real, linked sources.
When checking a video, the system uses subtitles when they are reachable. When they are not, an AI model actually watches the video within a strict time window. On the free tier, the model can watch up to 10 minutes on YouTube. The Pro tier extends this to 30 minutes. For X videos, the limit is up to 2 minutes on the free tier and up to 7 minutes on Pro. Longer videos on X are refused outright, not partially checked.
Keep in mind that the web app cannot see your logged-in user session on YouTube, so subtitles are often unavailable there. The browser extension runs in your active tab and can reach more subtitles. For X, both the web app and the extension take the exact same route, so there is no extension advantage there.
When the system extracts the text or audio claim, it runs a standard check. The standard check uses a Dual-AI cross-check powered by Gemini and Grok. Grok searches for itself there, so the two do not share one evidence set. The system then returns an evidence chain. Sources are never filtered by origin, meaning there is no mainstream bonus or preference for official sources. The reader decides what to trust.
Our published measurements show the effectiveness of this method. In our quote checking tests spanning 633 real citations re-fetched, 73 percent of quotes were found verbatim on the page. Meanwhile, 19 percent of pages were unreachable for automated readers due to paywalls or PDFs, and 6.4 percent were not findable.
In our source reach tests covering 24 claims with the root named in advance, 63 percent were primary sources, 14 percent were legacy media, and 1 percent were fact-checkers, with all 6 non-English roots reached successfully. Our published measurements also include AI-text detection, with 240 texts measured during our evaluations.
The tool provides a truth score of 1 to 10, but you should never consider this score as proof of trustworthiness. The system does not judge intent, and the final decision always stays with the reader. If you disagree with a result, wyper maintains a public dispute register. This register is append-only and hash-chained, meaning entries cannot be silently edited or removed.
For users who require a deeper investigation, the Full Spectrum tier offers a different approach. This tier exclusively features 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. Therefore, a disagreement is a different reading of the same material, not a different search result.
Our own published measurements based on 30 claims with frozen sources show that two models trained on overlapping text are correlated, not independent. Gemini and Grok agreed in 29 of 29 test cases, even on deliberately contested claims. Consequently, confirmation by two AI models carries little information. However, a third model of different origin disagreed in 3 of 26 cases, every time with a machine-verified quote. This independent third voice is the strongest thing we have, and it argues against the idea that confirmation by two AI models is enough.
| Manual checks | wyper web app | wyper extension |
|---|---|---|
| Free to use, keeping all search data strictly in your own browser history. | Offers a free tier (5 text checks per day) with no signup required. | Offers a free tier (5 text checks per day) directly in your active tab. |
| Requires no installation. | Requires no installation. | Requires the Chrome extension to be installed. |
| You must manually transcribe and search for video claims. | Uses subtitles if public; AI watches up to 10 min on YouTube (2 min on X). | Reaches logged-in YouTube subtitles; AI watches up to 10 min on YouTube (2 min on X). |
| Requires manually reverse-searching video frames for provenance. | AI Spotter available on paid plans only (4 credits per video). | AI Spotter available on paid plans only (4 credits per video). |
Who needs which: Casual readers who want to quickly verify a suspicious quote on their phone should use the web app, as it requires no installation and handles text checks immediately. Heavy researchers and daily news consumers are better served by the extension, which seamlessly reads subtitles from logged-in YouTube sessions and operates directly in the active tab. Privacy purists who prefer not to use third-party tools can stick to manual checks, keeping all search queries strictly within their own browser history.
Verifying an altered video requires looking past the pixels to trace the earliest publication date and checking the actual claims spoken in the audio. Tools can surface the evidence chain and find the original source, but they cannot judge intent. The final decision on what to trust always remains with you.
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.
wyper