Case study: verifying social media photos in the Netherlands

In January 2026, a series of photos began circulating on social media platforms with alarming claims. One post showed a flooded street with partially submerged cars, captioned: “Massive flooding in central Utrecht — authorities are hiding the extent of the damage.” The images were shared thousands of times within hours, and several international outlets picked up the story without independent verification. A Dutch regional newsroom decided to get to the bottom of it. This is the story of how they used GeoPin to separate fact from fiction.

Note: this case study is a composite narrative based on realistic scenarios. Names and specific details are fictionalised, but the investigative workflow reflects authentic open-source verification techniques.

The initial report

A journalist at the newsroom, whom we will call Sarah, first saw the flood images on a Monday morning. The photos looked convincing: water reaching car door handles, reflections of Dutch-looking brick buildings, and a street sign partially visible in one of the frames. Comments below the post ranged from outraged demands for government action to sceptical replies pointing out that it had not rained heavily in Utrecht recently.

Sarah checked the KNMI weather records for the previous week. Rainfall in Utrecht province had been unremarkable. No flood warnings had been issued by Rijkswaterstaat. This discrepancy made verification urgent: either a real emergency was going unreported, or the images were misleading.

Step one: reverse image search

Sarah started with a standard reverse image search. The earliest instance she could find was posted three days earlier on a different platform by an anonymous account. The post used vague language and no precise location beyond “Utrecht.” The image did not appear in any news archive, which was unusual for a genuine flood event of this scale.

The reverse image search was inconclusive. The photos did not match any previously indexed images, meaning they were either original or had been modified enough to evade detection. Sarah needed to determine where the photos were actually taken.

Step two: GeoPin analysis

Sarah uploaded the clearest of the three images to GeoPin. Within two seconds, the system returned its top prediction: a street in Dordrecht, not Utrecht, with a confidence score of 0.89 and an estimated accuracy radius of 200 metres.

This was a significant lead. Dordrecht, located in South Holland, has a well-documented history of urban flooding due to its low-lying geography and proximity to multiple rivers. A flood photo from Dordrecht would be unremarkable local news, not evidence of a hidden disaster in Utrecht.

Sarah uploaded the remaining two images. One returned a prediction in the same Dordrecht neighbourhood at 0.91 confidence. The third, showing a more generic scene, returned a lower confidence of 0.62 but still pointed to South Holland rather than Utrecht.

Step three: ground-truth verification

Armed with GeoPin’s coordinates, Sarah switched to Google Street View and navigated to the predicted location in Dordrecht. The match was striking. The same brick facades, the same style of bollards along the quay, and the same distinctive arched bridge were visible in the Street View imagery. The street sign partially visible in the original photo matched the Street View sign: it was a street in the historic centre of Dordrecht.

Sarah then searched local Dordrecht news archives and found coverage of a minor flooding event in the area from November 2025, caused by a combination of high river levels and spring tide. The photos matched the timeframe and location perfectly.

Step four: building the story

With the location verified, the story became clear. Someone had taken real photos from a routine flooding event in Dordrecht months earlier and repackaged them with a false caption claiming they showed Utrecht. The motivation was unclear — whether it was a deliberate disinformation attempt or simply a carelessly reshared post that snowballed — but the factual record needed correcting.

Sarah’s newsroom published a verification article walking readers through the process step by step:

1. The original claim placed the images in Utrecht.
2. Weather data showed no significant rainfall in Utrecht during the claimed period.
3. GeoPin’s visual geolocation identified the actual location as Dordrecht with high confidence.
4. Street View comparison confirmed the match.
5. Archival news reporting explained the original context.

The article included side-by-side comparisons of the social media images and Street View captures, along with a map showing the actual location. It was widely shared and cited by fact-checking organisations covering the same viral post.

Why visual geolocation matters for verification

This case illustrates a pattern that verification professionals encounter regularly. The most effective misinformation often uses real images in a false context. The photos were real, the flooding was real, but the claimed location and timing were not. Traditional tools like reverse image search struggle with this category of manipulation because the images themselves are unedited.

Visual geolocation fills that gap. By analysing the visual content of the image — architectural styles, street patterns, signage, vegetation — and matching it against a comprehensive reference index, GeoPin can establish where a photo was actually taken, regardless of what the caption claims.

For journalists and researchers working in the Netherlands, this is particularly valuable because Dutch cities share many visual similarities. Canal-lined streets, brick row houses and flat landscapes occur across multiple provinces. The subtle differences that distinguish Dordrecht from Utrecht, or Leiden from Delft, are precisely the kind of fine-grained visual features that trained models like CosPlace excel at.

Lessons for newsrooms

Several practical lessons emerged from this investigation.

Speed matters. The viral post accumulated thousands of shares in the first hours. Being able to verify a photo in seconds rather than hours changes the calculus of whether a newsroom can publish a correction while the story is still alive.

Multiple images strengthen the case. Uploading all available images rather than just one gives you multiple independent predictions. When three images all point to the same neighbourhood, the combined evidence is much stronger than any single result.

Confidence scores guide editorial judgement. A high confidence score does not mean the result is infallible, and a low score does not mean the result is wrong. Treating the score as one input alongside others — combined with weather data, local knowledge and Street View verification — produces the most reliable conclusions.

Document the process. Readers trust verification articles more when they can follow the reasoning step by step. Showing your methodology, including the tools you used and the confidence levels you observed, builds credibility in a way that simply asserting “this photo is fake” does not.

Looking ahead

As social media continues to outpace traditional editorial gatekeepers, automated and semi-automated verification tools are becoming essential infrastructure for newsrooms, fact-checkers and researchers. GeoPin is built with these workflows in mind, and we are actively developing features like batch verification and integration with existing fact-checking platforms.

If you work in journalism or open-source investigation and want to explore how GeoPin fits into your verification workflow, get in touch at press@geopin.nl. We offer personalised support and extended trial access for newsroom teams.