{"id":232561,"date":"2026-07-01T04:45:03","date_gmt":"2026-07-01T11:45:03","guid":{"rendered":"https:\/\/www.mcafee.com\/blogs\/?p=232561"},"modified":"2026-06-30T11:56:44","modified_gmt":"2026-06-30T18:56:44","slug":"ai-travel-photo-location-scams-geolocation-research","status":"publish","type":"post","link":"https:\/\/www.mcafee.com\/blogs\/mcafee-news\/ai-travel-photo-location-scams-geolocation-research\/","title":{"rendered":"AI Can Find Your Location 91% of the Time Using Just One Photo"},"content":{"rendered":"

How AI\u00a0<\/span>uses simple details in your photos to pinpoint where you are and why\u00a0<\/span>that\u2019s<\/span> a gold mine for scammers<\/span><\/span><\/h4>\n

McAfee Labs<\/span>\u00a0Safer Summer Travel<\/span>\u00a0Report | Summer 2026<\/span><\/span>\u00a0<\/span><\/p>\n

A Photo Is Worth a Thousand Data Points<\/span><\/span>\u00a0<\/span><\/h2>\n

You just got back from a week in\u00a0Central America. You posted a few shots:\u00a0the\u00a0colorful\u00a0streets of\u00a0Tulum, a\u00a0picture of the ancient ruins of Tikal, a close-up of your\u00a0shrimp\u00a0tacos. No location tag. No caption naming the city. Just a good photo.<\/span>\u00a0<\/span><\/p>\n

A few days later, you get a message. It references your bank. It mentions suspicious activity “while traveling internationally.” It feels oddly specific, with details about where you were and when.\u00a0It feels real.<\/span>\u00a0<\/span><\/p>\n

These types of personalized scam messages are a growing tactic. And your own photos may have helped write it. <\/span><\/p>\n

McAfee Labs set out to understand exactly how much location information exists inside an ordinary travel photo, and what that means for the\u00a0roughly 244\u00a0million Americans who travel each year.\u00a0<\/span>\u00a0<\/span><\/p>\n

What we found should change the way you think about what you share online:\u00a0<\/span>Some AI models have a more than 90% accuracy rate at detecting the location a photo was taken based on the visuals in the photo alone.\u00a0<\/span><\/b>And critically, that level of accuracy is now achievable using tools that are free and widely accessible.<\/span>\u00a0<\/span><\/p>\n

That’s why we’ve built tools like <\/span>McAfee\u2019s Scam Detector<\/span><\/a> that are designed to help spot these kinds of highly targeted, convincing messages before they lead to costly mistakes.<\/span>\u00a0<\/span><\/p>\n

What We Tested\u00a0And\u00a0Why<\/span>\u00a0<\/span><\/h2>\n

The question McAfee Labs wanted to answer was deceptively simple:\u00a0<\/span>Can AI look at a travel photo and figure out where it was taken, even without GPS data or location tags?<\/span><\/i>\u00a0<\/span><\/p>\n

Not metadata. Not embedded coordinates. Just the image itself:\u00a0the background, the architecture, the signage, the light;\u00a0the visual context that any photo naturally captures.<\/span>\u00a0<\/span><\/p>\n

To find out, we built an automated testing pipeline and ran it against a dataset of\u00a0<\/span>21,236 travel images<\/span><\/b>\u00a0sourced from publicly available image sets. We also conducted a separate, more controlled review of\u00a0<\/span>102\u00a0additional\u00a0images<\/span><\/b>\u00a0to pressure-test our findings.<\/span>\u00a0<\/span><\/p>\n

We tested two publicly available, large-scale AI vision models\u00a0that\u00a0are\u00a0both\u00a0freely available. Neither required\u00a0special access, proprietary data, or advanced technical\u00a0expertise\u00a0to run. We used the same tools a scammer could access today.<\/span>\u00a0<\/span><\/p>\n

Each image was\u00a0analyzed\u00a0using a consistent automated prompt asking the model to\u00a0identify\u00a0the location depicted\u00a0(city, country, or region)\u00a0based solely on visual content. Results were then reviewed by human analysts to\u00a0validate\u00a0accuracy and flag edge cases.<\/span><\/p>\n

What We Found<\/span>: AI Has a Whopping 91% Accuracy Rate<\/span><\/span>\u00a0<\/span><\/span><\/h2>\n

The results were striking.<\/span>\u00a0<\/span><\/p>\n

Gemma3 27B correctly\u00a0identified\u00a0the city and country of a travel photo 87% of the time.<\/span><\/b>\u00a0Qwen3 VL 30B performed even better, reaching\u00a0<\/span>91% accuracy<\/span><\/b>\u00a0across the same dataset.<\/span>\u00a0<\/span><\/p>\n

That means in\u00a0roughly 9\u00a0out of 10 cases, an AI model\u00a0that\u2019s\u00a0available for free, to\u00a0anyone, could\u00a0look at an ordinary travel photo and correctly name where it was taken.\u00a0This kind of analysis is also how AI tools understand images more broadly, shaping not just\u00a0scams, but how information shows up in AI-powered answers.<\/span>\u00a0<\/span><\/p>\n

And when the exact city\u00a0wasn’t\u00a0identified, the country alone was\u00a0almost always\u00a0correct. For a scammer,\u00a0that’s\u00a0more than enough.\u00a0It\u2019s\u00a0also enough to turn a vague, generic\u00a0scam\u00a0into one that feels specific,\u00a0timely, and believable.<\/span>\u00a0<\/span><\/p>\n

What Makes a Photo Easy to Place?<\/span>\u00a0<\/span><\/h3>\n

Certain types of images were identified with even higher confidence:<\/span>\u00a0<\/span><\/p>\n

    \n
  • Photos featuring\u00a0<\/span>famous landmarks<\/span><\/b>\u00a0or recognizable skylines<\/span>\u00a0<\/span><\/li>\n
  • Images taken in\u00a0<\/span>popular tourist destinations<\/span><\/b>\u00a0with distinctive visual signatures<\/span>\u00a0<\/span><\/li>\n
  • Photos with\u00a0<\/span>visible signage<\/span><\/b>,\u00a0unique\u00a0street markings, or local architecture<\/span>\u00a0<\/span><\/li>\n
  • Images that captured\u00a0<\/span>cultural context:<\/span><\/b>\u00a0transportation, storefronts, food stalls<\/span>\u00a0<\/span><\/li>\n<\/ul>\n

    Less recognizable scenery, like\u00a0a generic beach, a rural road,\u00a0or\u00a0a hotel room,\u00a0lowered accuracy. But even in those cases, country-level identification remained high.<\/span>\u00a0<\/span><\/p>\n