Live photo-only benchmark
How accurate is AI at geolocating photos?
We record the AI detective’s first guess before it sees any clue cards, then compare that pin with the verified answer. This page updates as each daily case moves into the public archive. It uses real, human-curated images and excludes today’s mystery photo.
Current sample: 22 past daily cases. Last case: 2026-07-11.
Results at a glance
The useful lesson is not a single accuracy score. Image geolocation is uneven: recognizable landmarks can be easy while visually similar streets, landscapes, and artworks can send the same system to the wrong region or continent.
The five largest photo-only misses
- Machu Picchu, Cusco Region, Peru2026-06-2916,586 km off
- Mesa Arch, Canyonlands National Park (Island in the Sky), San Juan County, Utah, USA2026-06-2811,820 km off
- Reither Spitze, Austria2026-07-118,584 km off
- São Paulo, Brazil2026-07-108,408 km off
- Iiyama, Japan2026-07-087,683 km off
All links lead to already-revealed archive cases. The live daily answer is never included.
Methodology
Each measurement uses clue count zero: pixels only, with no EXIF, GPS, filename, answer, or human clue card. Error is the great-circle distance between the AI pin and the verified answer. We report the median because a few continent-scale misses can distort the average. This is a transparent product benchmark, not a claim that the sample represents every country, camera, or geolocation model.
Read how visual geolocation works, inspect every revealed case in the archive, or play today’s human-vs-AI duel.