Content Verification

Content verification is the open-source practice of testing whether a photo, video or news clip shows what its caption claims, by tracing it to its original source, place and date. Newsroom desks such as BBC Verify and AFP Fact Check and fact-checkers such as Full Fact, Misbar and Newschecker do this work, often in the hours after a real attack, when fake footage can spread faster than the genuine record.

Old footage from unrelated fires and flights attaches itself to fresh events, from Moscow to Riyadh. Generative AI adds deepfake news reports that borrow trusted brands, synthetic images and video, and AI "enhancement" that invents detail. Fact-checkers combine reverse image search, geolocation, comparison with primary footage and AI detectors, which catch only part of the problem.

As of October 2026, fact-checkers had debunked at least six fake or miscaptioned clips tied to a single Flydubai hijacking attempt, Google's public SynthID Detector read watermarks from four AI companies, and Google had withdrawn a Google Earth tool that generated synthetic satellite imagery after open-source researchers objected.

Recycled footage after real attacks

A real attack opens a gap before verified imagery emerges, and old footage fills it within hours. Fire videos from a 2020 blaze in Ajman, UAE, and a March 2026 fire in Egypt spread as footage of Ukraine's September 2026 drone strikes on Moscow. The Flydubai hijacking attempt of 30 September 2026 drew at least six fake or miscaptioned clips by October 2026, among them old videos from unrelated flights, such as a July 2025 EasyJet incident recaptioned as the Flydubai co-pilot shouting "Death to Trump".

Houthi strikes on Riyadh's King Khalid International Airport in October 2026 drew a run of fake and recycled clips. A fabricated video of a missile hitting a Saudia airliner, which iVerify traced to a 2017 post on X, drew 3.6 million views on one repost and outran real documentation of the 8 October strike, which shows only the aftermath. After a further attack on the airport on 10 October, Misbar traced a viral clip claiming to show the damage to footage of Dubai's airport after an Iranian strike; it traced clips claiming strikes on a Riyadh festival and on Dammam's airport to a November 2025 fire at a Riyadh attraction.

Deepfakes and AI-altered images

Fabricated news reports borrow the look of trusted outlets. After the Flydubai attempt, a deepfake styled as a BBC report blamed an Indian passenger and the pilot; the suspect Israel identified was an Omani national. Another clip invented a NewsGuard report, which NewsGuard said appeared tied to a Russian campaign imitating trusted media brands.

Synthetic images and video fabricate the scene itself. An AI-generated image of Iran's President Masoud Pezeshkian at the UN added a pink school bag he never held, which AFP said appeared to exploit a symbol of a February strike on a school in Minab. An AI-generated video of Ukrainian soldiers changing into civilian clothes spread to Russian propaganda outlets; Ukraine's Center for Countering Disinformation said it aimed to falsely accuse Ukrainian forces of violating international humanitarian law.

AI "enhancement" predicts detail rather than recovering it. After five men were arrested near RAF Fairford in September 2026, AI-sharpened versions of a villager's grainy photo showed different faces for the same men, and some users on X presented them as proof the men were Muslim, though police had disclosed no names or religion.

Verification methods

AFP's debunk of the Moscow fire videos rested on standard open-source techniques rather than specialised access: it matched keyframes of the Ajman video to a 2020 Facebook post and confirmed the location in Google Street View. Lead Stories used Google Lens to match the Riyadh missile video's backdrop to a 2016 stock photo, and BBC Verify geolocated the original Fairford photo against Street View, confirming it as authentic with the faces pixelated.

Comparison with primary footage can settle a case without forensic tools. AFP checked the Pezeshkian image against UN video of the speech, which showed no school bag. On the fake BBC report, independent teams converged by different routes: France 24 ran it through a detector that scored it 99% likely AI-generated, Newschecker.in traced it by reverse image search to a genuine BBC video, and Full Fact spotted a mismatched banner font and an AI watermark.

Detection tools and their limits

AI detectors add a further check but give uneven results. DeepFake-O-Meter scored the Ukrainian soldiers video 86.1 to 99.6 per cent likely synthetic, and PA Media's check with OpenAI's own tool found an OpenAI watermark on the sharpened Fairford image. On the Riyadh missile video, detectors returned no conclusive signal, which Lead Stories said does not rule out AI origin.

Google opened its SynthID Detector to the public in October 2026. The free site reads watermarks from Google, OpenAI, Nvidia and Kakao, but misses Meta's watermarking standard, open-weight models and anything stripped of metadata.

The reference layer itself is exposed. On 31 July 2026, about a day after adding it, Google rolled back a Google Earth tool that generated synthetic satellite, aerial and 3D imagery from text prompts, after Bellingcat's Jake Godin and the Alan Turing Institute's Sam Stockwell warned it risked corrupting the imagery analysts use to debunk fakes.