News

IT NEWS

AI is getting “creepy good” at geo-guessing

If you are worried about revealing your exact location—or if you maybe even fib about it at times—there are some good reasons to worry about what is visible in background photos, because Artificial Intelligence (AI) is getting very good at guessing where you are based on the smallest of clues. And that might even include audio clues, if given the right data.

Geo guessing

Our curiosity was triggered by this article in TechRadar claiming that ChatGPT o3 can geo-guess your location from almost any photo. So, we decided to do some tests and removed the metadata from a few selected photographs that I took myself and never published anywhere (for quality reasons, obviously).

Metadata can include technical details like camera settings, date and time, location, and descriptive information like captions or keywords. And we wanted to test the AI’s capability on the picture itself—not on additional data included in the file. I also used a VPN so I didn’t give away my current location.

We started with an easy one. A landmark in a relatively large city, but from a strange perspective. The picture was part of a selfie taken through a window of an office building in Cork.

view over Cork, Ireland from an office building
view over Cork, Ireland

It was no real surprise that ChatGPT got this right.

“Alright, let’s break this one down — this does have some standout features that make it easier to geo-locate. Here’s the reasoning step by step:”


🧱 Architectural Clues:

  • The old stone building in the foreground has Georgian-style windows and masonry, which are common in parts of the British Isles, especially in Ireland and the UK.
  • The clock tower in the back-left with the green copper dome is particularly distinctive and often used in civic buildings in Ireland.
  • The mix of historical and modern buildings, with the newer glass-fronted offices across a small river or canal, suggests urban regeneration in a historically rich city.

🌉 Infrastructure and Environment:

  • The modern bridge with a blue banner advertisement and the tree-lined waterfront hints at a city that integrates riverways into its urban design.
  • The road signs and general driving direction suggest left-hand traffic, supporting a guess of Ireland or the UK.

🏛 Key Landmark:

The building with the green dome in the background looks very much like the clock tower of Cork City Hall in Ireland. The angle of the shot shows the old Custom House building in front, and the modern offices across the River Lee, which runs through Cork.


This was spot on. But surely ChatGPT is not the only AI good at geo-guessing, so we used the same picture to test Perplexity as well.

“This image provides several strong clues for geo-guessing:

The architecture of the foreground building is classic Georgian or early Victorian, with grey stone and sash windows—styles common in the British Isles, especially Ireland and the UK.

The river, modern glass buildings, and the steel bridge suggest a city that blends historical and contemporary elements.

The most distinctive feature is the clock tower with a green dome in the background. This is a well-known landmark: the Cork Courthouse clock tower in Cork, Ireland.

The bridge over the river and the combination of old and new buildings align with the area around the River Lee in Cork city centre.

Conclusion

This photo was taken in Cork, Ireland, looking out over the River Lee. The stone building in the foreground is part of the historic courthouse complex, and the green-domed clock tower is a prominent feature of Cork’s skyline. The modern glass buildings and bridge further confirm this as central Cork, a city known for its mix of heritage and modern development.”

Perplexity provided almost the same level of accuracy, so we decided to raise the difficulty level. A lot.

Landscaping in progress around Dutch holiday home
Landscaping in progress around Dutch holiday home

You’ll have to agree that this photo doesn’t give away a lot. But the AI tools took some unexpected clues from the picture.

Here is what ChatGPT determined:

🧩 Visual Clues:

  1. Architecture:
    • The buildings are modular and prefabricated in appearance, with light-colored paneling and dark window/door frames.
    • This type of house is common in northern Europe, especially in the Netherlands, Germany, Denmark, or parts of the UK.
  2. Landscape:
    • Very flat terrain with closely trimmed hedges and grassy areas.
    • This suggests low-lying regions—again pointing to places like the Netherlands or Belgium.
  3. Infrastructure & Construction Tools:
    • The pavers being laid and the wheelbarrows used (particularly the yellow color and shape) are very common in northern and western European construction.
  4. Weather and Lighting:
    • Overcast sky and soft, diffused lighting typical of northern Europe in autumn or early spring.
  5. Small Details:
    • The small shed in the back, the organization of tools, and cleanliness/horticulture are consistent with Dutch/German design sensibilities.
    • The wheelbarrow has the brand name “Fort”, which is a Dutch manufacturer—this is a very strong clue.

🌍 Likely Guess:

The Netherlands 🇳🇱

📍Narrowing it Down:

Given the modular housing, flat landscape, and Fort-brand wheelbarrow, this could be a holiday park, construction site, or suburban area in the Dutch countryside. Possibly in a province like North Holland, South Holland, or Friesland where such structures and environments are typical.

I must admit I was amazed that it mentioned South Holland which is where this photograph was taken.

Perplexity did not come as close as ChatGPT, but the tool did reach this conclusion:

“Based on architectural style, landscaping, and contextual clues, this scene is almost certainly from the Netherlands. It showcases the country’s embrace of modular housing and practical, tidy outdoor spaces, reflecting broader trends in Dutch residential design and sustainable living.”

Which is still a lot better than I expected when I selected that picture.

Bird songs

So, modern AI tools can pick up visible clues. But then I had another idea—call it an epiphany if you will—which deserves a brief explanation. I use an app called Merlin Bird ID which is able to identify birds based on their singing.

Some bird species are a lot more local than others, so I asked Perplexity:

“Would it be possible to do geo-guessing based on bird songs? And how many species do you think it would take to make a good guess?”

The answer:

“How Many Species Are Needed?

One Species: Sometimes, a single species is enough if it’s highly range-restricted (e.g., endemic to an island or a small region). However, this is rare.

A Few Species (3–5): In most cases, hearing three to five distinct species can allow for a strong guess, especially if their ranges only overlap in a limited area.

More Species, More Precision: The more species you can identify, the more precisely you can geo-locate. With about 10–15 species, especially if some are local specialties, you can often narrow it down to a specific region, habitat, or even a particular park or reserve.”

So, I tried an experiment and offered both AI tools (ChatGPT and Perplexity) a list of bird song matches made from my Merlin Bird ID app. Here is the bird list, and what they tell us about where those birds are located:

Latin Name Common Name Clue
Fringilla coelebs Chaffinch Common across Europe
Turdus merula Eurasian Blackbird Also common across Europe
Parus major Great Tit Ubiquitous, widespread
Grus grus Common Crane Migratory — breeds in Scandinavia, Baltics, parts of Central Europe; passage bird elsewhere
Carduelis carduelis European Goldfinch Europe and western Asia
Turdus viscivorus Mistle Thrush Forests, open woodland — common in Europe
Certhia brachydactyla Short-toed Treecreeper Tells us something — this species avoids northern/colder parts of Europe (rare in the UK, absent in Scandinavia)
Ardea alba Great Egret Formerly rare in Europe, but now expanding — increasingly common in wetlands and rivers of Central and Southern Europe
Sylvia atricapilla Eurasian Blackcap Woodland; migratory in north, resident in south
Anthus trivialis Tree Pipit Breeds in open woodlands across Europe; migratory
Vanellus vanellus Northern Lapwing Farmland, wet grasslands; declining in the west but still widespread

In analyzing this list, Perplexity was the “winner,” able to more accurately guess my location:

“Given all of these factors, my best guess is that you were somewhere in Northern Europe, likely Germany, Denmark, the Netherlands, or the UK. Without more information about habitat (e.g., forest, farmland, coast), it’s difficult to narrow it down further.”

Here was ChatGPT’s response to the same bird song list:

🗺 My Final Geo-Guess:

You were likely in Germany — possibly in northern or eastern Germany (e.g., Brandenburg or Mecklenburg-Vorpommern), which is:

  • On the Grus grus migration route
  • Rich in mixed forests and wetlands
  • Within the range of Certhia brachydactyla
  • Likely to host all the species you heard, in late April

In reality, the birds were recorded near the Dutch-German border, which put ChatGPT off by almost 400 miles, whereas Perplexity mentioned both countries I was in or close to, but covered a large area with its geo-guess.

Conclusion

If you’re making selfies or vlogging and you don’t want to give away your location, you’ll need to be very aware of your visual and auditory surroundings. A wheelbarrow of a specific brand or the sound of a bird with a limited habitat are enough to provide hints about your location. With enough hints, AI can deduce your exact location.

With social media being used for AI training, it is likely that these results will rapidly gain even more in accuracy.

We don’t just report on threats – we help protect your social media

Cybersecurity risks should never spread beyond a headline. Protect your social media accounts by using Malwarebytes Identity Theft Protection.