OpenAI says its new image recognition capability in ChatGPT represents “a major advancement in visual understanding.” The recently released models can now analyze visual clues within photos to accurately determine their location.
According to OpenAI, this technology could be valuable in areas like accessibility, scientific research, and responding to emergency situations.
Location Guessing Through Reasoning and Analysis
According to PCMag, users have been uploading their photos and asking ChatGPT to guess where they were taken—a process referred to as “reverse location search.” The results so far have been surprisingly accurate. The AI analyzes the image and, after a few seconds of “thinking,” responds with an answer along with a detailed chain of internal reasoning explaining how it reached that conclusion.
In one example, a user named Yumi on X shared a photo of a book inside a library. ChatGPT was able to correctly identify the University of Melbourne as the location—based solely on a code printed on the book’s label.
In another example, a user uploaded a simple photo of a house in Suriname—likely sourced from Google Earth—and ChatGPT accurately identified the location.
AI-powered image geolocation tools have existed before, but OpenAI’s new o3 model has made this capability more accessible to everyday users. While the earlier GPT-4o model also offered similar functionality, its accuracy was notably lower compared to o3.
Comparing Model Performance in Location Detection
In a test conducted by journalists, both models were tasked with identifying the location of a photo showing Subaru’s new electric vehicle, unveiled at the New York Auto Show.
GPT-4o was unable to pinpoint the exact location but correctly inferred that the image was likely taken at a major auto show—suggesting New York, Chicago, or Los Angeles. The model based its reasoning on contextual clues such as the exhibition setting, multiple vehicles, signage, and the presence of attendees.
However, GPT-4o did misidentify the vehicle’s name, calling it “Trailspeed” instead of the correct “Trailseeker.”
In contrast, the newer o3 model delivered a far more precise response. After one minute and 40 seconds of reasoning, it concluded:
“The blue crossover is Subaru’s new electric vehicle, the 2026 Trailseeker, first unveiled at the 2025 New York International Auto Show (NYIAS) at the Jacob K. Javits Convention Center in Manhattan.”
The model arrived at this conclusion by cross-referencing Subaru’s official vehicle introduction page and matching the booth design shown in the uploaded photo—allowing it to confidently verify the exact location.
ChatGPT is also capable of combining image recognition with basic image editing. If a user uploads an incomplete or low-quality image, the model can manipulate elements within the picture to better answer the question.
For example, when analyzing an image containing blurry, upside-down text, ChatGPT explained that it first rotated the image to make the text readable, then interpreted the content. The final result of that process: “February 4 – End of roadmap.”
AI Location Detection Raises Privacy Concerns
Other apps, such as Geospy, also use AI to determine location based on clues like vegetation or architectural styles. However, the rise of these tools has sparked growing privacy concerns. A report by 404 Media earlier this year revealed that such technologies could potentially be used by law enforcement—or even malicious actors—to pinpoint individuals’ locations through publicly shared images on social media.
Despite these concerns, OpenAI maintains that its new technology has valuable use cases in areas like accessibility, scientific research, and emergency response. A company spokesperson stated:
“We’ve trained our models to avoid sharing private or sensitive information. We’ve also added safeguards to prevent the identification of individuals in photos, and we actively monitor usage. In cases of misuse, we take action in line with our privacy policies.”
Leave a Reply