iVerify Pakistan investigated two narratives about visuals from the recent violence and found that the truth is not always what everyone is saying it to be, yet conclusively determining it is not always possible. An image said to be AI-generated was found to be apparently real, while a video being used as evidence of a “staged” stunt does not appear to be the case.

On July 28, just a few hours after polling in Azad Jammu and Kashmir (AJK) kicked off, a video began doing the rounds on social media, particularly X. The clip shows bloodied bodies lying in the middle of a road in the Mirpur district. “Exposed,” read the caption of the post, shared by an X user whose account features pro-state content. “They were just lying down with red liquids thrown over them for a photo session. When the police neared and kicked, they got up. Fraud unveiled,” it added.

The video quickly crossed 100,000 views, with shares multiplying by the thousands every hour. Several fact-checking accounts that are considered to be aligned with the state’s narrative also shared the clip with similar captions. Most of them alleged the clip was staged. At the same time, an image, said to be showing the same incident, also began circulating on X, with the claim that the visual was AI-generated. The accounts sharing it alleged that the image was being spread to portray a false picture of the situation in AJK.

When iVerify Pakistan was alerted to the video and the image, both were already being contested from multiple directions, accused of being staged, AI-generated, or both, largely by accounts aligned with the state’s narrative. Untangling that took two separate lines of inquiry.

On the image, the trail led backwards. Several fact-checking accounts had pointed to visual distortions — blurred objects, odd number plates — as proof the photo was AI-generated. But a search turned up an earlier, blurrier version of the same image, circulating nearly half an hour before the viral one, shared by several users, including PTI’s official account and its chapters such as PTI Faisalabad, PTI Mardan and PTI West Punjab.

Running that older image through a set of AI-detection tools returned no indication it was synthetic, and a check for Google’s SynthID watermark came back clean. When iVerify ran the older image through an AI enhancement tool, the output matched the sharper, more “suspicious” version that had gone viral — down to the same distorted details fact-checkers had cited as proof of fabrication. In other words, the image wasn’t AI-generated evidence of a staged protest; it was a real photo that had likely been run through an enhancement tool, which introduced its own artefacts along the way.

The video, meanwhile, took a different route. A keyword search surfaced the same clip published by BBC Urdu on July 29, described as clashes in Rawalakot and Mirpur amid “conflicting claims regarding casualties”.

A reverse image search then turned up a clearer version of the same moment, along with footage of the same incident from other angles. None of it showed what the “fraud unveiled” caption described: no one stood up and ran off when police approached. Instead, officers could be seen dragging two people toward a police vehicle, both visibly limping and unable to walk unassisted.

Other angles picked up the sound of gunshots and people fleeing in panic, with one clip showing a police vehicle positioned in the direction the crowd was running from. Taken together, the visual evidence points away from the “staged” narrative that crossed 100,000 views.

None of this amounts to a full account of what happened in Mirpur that day because restablishing whether the visuals actually show what was claimed requires something forensics alone can’t provide: confirmation from someone who was there. At iVerify, a key element of the investigation that goes into any and every fact-check involves confirmations, testimonies, and on-the-ground reportage, especially in times of conflict. This time, though, the team was unable to reach a single source inside Mirpur to confirm the authenticity of the video and the claim. Two days later, we are still struggling to get hold of any information at all.

For over a month, it has been challenging to reach locals or correspondents in the region; WhatsApp messages to correspondents are on a single tick, and calls are barely getting through. As it votes in its most “contested” election in years, AJK has also been cut off from the outside world — a combination that has turned fact-checking from a routine editorial task into something closer to a rescue mission.

The government has not addressed the allegations or reports of casualties at substantive length as well, while it accuses the JAAC of instigating the violence and harming security personnel.

The Mirpur video is one of several claims iVerify Pakistan has been unable to verify since polling began — each testing the limits of what fact-checking can do when a region goes dark.

Another example is videos appearing to show civilians in AJK opening fire on police. Some claimed they showed members of the now-proscribed Joint Awami Action Committee — a socio-political civil rights coalition at odds with the government since before the elections. Meanwhile, others alleged the men opening fire were security personnel dressed to look like ordinary citizens.

Both claims rest entirely on interpreting the same footage, and neither can be tested without someone on the ground to say who the men in the videos actually were. The claim remains unverified, not because iVerify hasn’t tried, but because the tools of open-source verification run out exactly where testimony would need to begin.

Fortunately, not every claim iVerify has encountered since polling began has run into the same wall. Some have been resolvable without a single call to the ground, serving as a reminder that the blackout, as severe as it is, hasn’t stopped fact-checking entirely; rather, it has just narrowed what kind of fact-checking is still possible.

An image circulating widely on X claimed to show a large number of bullet casings recovered after security forces opened fire on peaceful protesters in Rawalakot. We did not require a source in the district to take it apart.

A closer inspection of the clip revealed that several of the people in the frame had anatomically implausible hands — missing or extra fingers, unnatural shapes — and one face appeared partially melted, a distortion common in AI-generated crowd scenes. The bullet casings themselves were oddly oversized and inconsistent in scale relative to the people around them, more like props than spent shells from small-arms fire.

Taken together, the artefacts were consistent enough with AI generation that the image could be assessed on forensic grounds alone, without anyone on the ground to vouch for it.

Similarly, a video claiming to show Punjab Police firing on citizens in Kashmir followed a similar pattern, though by a different route. It wasn’t new: we had already fact-checked the same clip once before, tracing it back to a police clash with protesters near the US consulate in Karachi in March 2026. Its reappearance now, recontextualised as footage from Rawalakot, showed that old material can be recirculated to fit a new crisis, and that a claim debunked once doesn’t stay debunked in the timeline.

Given the nature of the unrest, this information isn’t just circulating — it’s actively feeding a divisive narrative on either side of an already polarised online discourse. In such a climate, there is no shortage of actors with a reason to keep the information war going, and every unresolved claim becomes fuel for both.

Verification, in this environment, isn’t a formality; it’s the only thing standing between a claim and its unchecked spread. But as these cases show, it is also becoming harder to do at exactly the moment it matters most, with the ground truth cut off by a blackout. Until that access is restored, the responsibility shifts to the people encountering this content before anyone else does. If a claim’s authentication is dubious — if it cannot be traced, confirmed, or corroborated — it should not be shared.