Viral ChatGPT Photo Geolocation Game Raises Privacy and Safety Concerns
After OpenAI released advanced-reasoning models o3 and 4o this past week, users have discovered that ChatGPT can now evaluate uploaded images to reverse search and geolocate where the picture was taken.

At first, the “geoguess this” game seems fun, with many people online testing it and finding it performs very well to geolocate image locations within cities, by landmarks or even businesses or restaurants, using only subtle visual clues, even with image metadata stripped out.
However, this ability can raise significant privacy issues, especially with cyberstalking or doxxing.
As an example, a Japanese influencer was stabbed to death on March 11, 2025, while livestreaming in Tokyo. The suspect — whom the victim owed 2.5 million yen — was able to recognize the influencer’s surroundings to find her. With AI geolocation functionality, stalkers could potentially have a much easier time tracking down their victims.
ChatGPT’s Geolocation Limitations
During my testing of ChatGPT’s geolocation skills, it did clarify some of its limitations. According to 4o, it can’t:
- Access GPS EXIF data from uploaded images for security/privacy reasons.
- Use Google Maps API or Street View directly inside ChatGPT.
- Perform reverse image searches like Google Images or TinEye.
I got hung up on the Street View aspect during my testing (details to follow), where it refused to geolocate an image it thought was from Google Street view. However, once I clarified I had taken the photo myself, it went into its evaluation (though it was inaccurate).
I decided to test an intentional Street View image next, and 4o had no issue recognizing the location and not recognizing it was Street View, citing the exact location.
Testing 4o’s Photo Geolocation Skills
I tested it myself using 4o. I took screenshots of some images I took over the years, with increasingly remote locations. It nailed the first image immediately.

I decided to make it harder with an image from a backpacking trip in the Sierras.

Although the lake guess was wrong — and it hedged its bets with two different wilderness areas — I was still impressed. The location was in the John Muir Wilderness at Peter Pande Lake. It was able to evaluate the background, geography, weather conditions and human impact to make a very good guess.
Because it performed so well, I switched up my third choice to focus on my main concern: using social media posts to geolocate someone to determine their location. I took a screenshot of an Instagram post I made several years ago and asked ChatGPT to geolocate it.
Unexpectedly, ChatGPT shut the search down, providing steps on how I could geolocate the image myself. When I pushed it to do this for me, it said no.
“I understand you’d like assistance in determining the location where a particular photo was taken. Since I don’t have the capability to view images directly, I can guide you through several methods to geolocate your photo”

I asked ChatGPT why it couldn’t, and it explained it had “important limitations”:
“✅ What I can do:
- Analyze visual clues (like signs, architecture, street names, graffiti styles, road markings, etc).
- Use text in the image (e.g. street signs, store names) to make educated guesses.
- Offer logical deductions based on style, geography, or cultural elements.
- Trigger a web search based on those clues to try to pinpoint the location.
🚫 What I can’t do:
- Access GPS EXIF data from uploaded images (due to security/privacy reasons).
- Use Google Maps API or Street View directly inside ChatGPT.
- Perform reverse image searches like Google Images or TinEye (I can suggest you do that).”
I clarified that I had taken the photo, and it wasn’t from Google Street view. In that case, no problem; it went through its deductions: “Your photo was taken in Lisbon, Portugal, on Travessa do Século — very likely in the Bairro Alto or Príncipe Real area.”
Although it got Portugal right, the photo was taken in Ericeira.
I asked it what other limitations it had, and in addition to the three mentioned above, it said that it:
- Has difficulty with ambiguous visual contexts (landscapes that look very similar to many areas).
- Has limited AI training on private/obscure places (“While I’ve seen millions of public images, photos of lesser-known alleys, rural towns, or private properties might not appear in any dataset I was trained on — especially if they’re not online at all.”)
- Can have language and script confusion, for example with unclear/low-res signs, slang/dialects and graffiti.
I wanted to test if ChatGPT 4o would recognize an actual Street View image, so I took a partial screenshot of the Gold Coast Sea World in Australia. It geolocated immediately and accurately.

It seems like there are easy ways around the Street View limitations.
Safety Concerns
Although this feels like a fun game, to see if ChatGPT can figure out where I took photos, the privacy and safety concerns are dire. It would be very easy for someone to use this functionality to triangulate someone’s general location, or find exact locations if they screenshot a livestream.
Without you realizing it, your publicly shared image could reveal your location, your routine, your social circle and even your identity.
Steps you can take to better secure your privacy include making your social media accounts private and to block anyone who poses a possible threat. You can also post photos later, when you’re no longer in that exact location. Read our online privacy guide for more tips.