Can Google Lens Find a Location From a Photo?

Google Lens is the first thing many people try on a mystery photo. It is built into Chrome, the Google app and Google Photos, and in May 2025 Google said more than 1.5 billion people were using it every month. So it is natural to drop in a holiday snap, a screenshot or a picture from a news post and ask where it was taken.

Lens can often help, but not in the way the question assumes. It does not look up a location for a photo. It recognizes what is in the frame and searches for matching pictures on the web, so a place only appears when something in the results names it. This guide sets out what Google itself says Lens does, where that works, where it fails, and what to use instead.

Google’s help page lists four kinds of results a Lens search can include: AI Overviews, search results for objects in the image, similar images, and websites that carry the image or a similar one. A location is not on the list. You get one when a result happens to name the place, such as a landmark Lens recognizes, a street sign it reads, or a travel blog that shows the same view with a caption.

Lens is one of several reverse image search engines (the main ones compared), and it runs at a huge scale: Google said in October 2024 that it handled nearly 20 billion visual searches a month. For finding a location, though, popularity is not the point. What matters is whether the place in your photo has been photographed and posted before.

Google’s longer explanation of how Lens works used to live at lens.google. That address now redirects to a product page, but an archived copy from September 2026 is still readable. It says Lens “compares objects in your picture to other images” and ranks them by similarity and relevance. It adds that Lens can also weigh the text, language and other metadata of the page where an image appears. That is the web page around a matching picture, not data stored inside your file.

• Landmark recognition. Google’s help says a photo of a landmark or building can bring up historical facts or opening hours. Its Lens product page lists landmarks, murals and artwork among the things Lens can tell you about.
• Visual matches. Lens shows similar images and the websites that carry the same or a similar picture. If one of those pages is a news story, a listing or a travel post, its caption may name the place.
• Text. Lens can copy text and translate it in over 100 languages, so a street sign, shopfront or menu becomes something you can search. On Android and iPhone, selecting an address in an image offers a Maps action.

Ranking also means Lens can sound surer than it is. Google’s own example is a dog that Lens judges 95% likely to be a German shepherd and 5% likely to be a corgi: it may show only the German shepherd. A place works the same way. The top result is the most similar picture, which is not proof that it shows the same spot.

Then there is location. When you agree to share yours, Google says Lens uses it to return more accurate results, for example when it identifies places and landmarks. In Google’s example, someone in Paris gets the Eiffel Tower rather than a look-alike elsewhere. That is the searcher’s location, though, not the photo’s. For a picture from last year’s trip, where you are sitting now says nothing about where it was taken.

Lens is also no longer a pure lookup. Its results can include an AI Overview, and in Chrome a follow-up question from the Lens side panel opens AI Mode. In Bellingcat’s August 2025 retest of 24 models, Google’s AI Mode was the most capable geolocation tool overall. Even so, Bellingcat found it sometimes named the wrong place with confidence.

Bellingcat’s tests are a useful public benchmark. In June 2025 the investigations group ran 500 tests: 20 AI models each looked at the same 25 travel photos, none of which had been published online. Lens was checked on the first 10 results of its visual match feature for every photo. Only three ChatGPT models beat it, and only narrowly. The same tests from ChatGPT’s side are covered separately.

Because none of those photos was online, Lens could not have found the pictures themselves. It found places that look the same in other people’s photos. That is the real requirement: not that your photo is published, but that the place in it has been photographed and posted many times. A photo that is published adds a second route, because Lens lists the websites that use the image.

Landmarks are the easiest case. If Lens names the building, the hard part is done, although the landmark’s position is not where the photographer stood. Identifying a landmark from a photo covers that last step, from naming the building to matching the exact view.

Two of Bellingcat’s photos show the pattern. A field near Zurich, with only distant mountains to go on, did not immediately lead Lens to Zurich. For an alley on Circular Road in Singapore, Lens spread its results across Singapore and Malaysia. The models that got it right read the address on a mailbox in the foreground.

Lens can read text like that too, but you have to point it there. Google’s tip is that selecting a smaller area of the image gives more specific results. A whole street scene gives Lens a dozen competing things to match. A single sign gives it one.

A private photo of an unremarkable spot is the worst case. Nothing like it has been posted, so there are no matching pages, no captions and no landmark to recognize. Lens still ranks the most similar pictures it can find, so you get look-alikes from other places. They are easy to mistake for an answer. If Lens draws a blank, Yandex is worth a second try before you give up on image search.

EXIF is the block of data a camera or phone writes into a photo file, and it can include GPS coordinates when location is switched on (what EXIF data records). A GPS tag states a position. Everything Lens looks at only suggests one.

Google’s page on image details lists where the file information it shows comes from. Creator, credit and copyright come from IPTC fields, license details from IPTC or structured data, and signs of AI generation or editing from C2PA and SynthID. Location is not among them. Google also says SynthID watermarks can surface in Search and Lens, and that Lens shows AI-edit details under About this image. That panel is available only in some regions.

About this image is still useful for location work, just not through GPS. It shows when Google may first have seen an image or similar versions of it, and which other pages use them. An early page with a caption can name the place, or show that a photo is older than the post claims.

So check the file first. If the photo came straight from a phone or camera, its coordinates may still be inside, and reading them takes a minute. If it came from a social feed, the tag is probably gone, because publishing platforms re-encode uploads (why photos lose their location).

The three fail on different photos, which is why they combine well. A screenshot of a famous square has no GPS tag, but Lens may well name it. A phone photo of a quiet lane has no match online, but its tag may still be there. A reposted picture of a village street has neither, and the clues in the frame are all that is left. How AI geolocation works explains how a model reads those clues.

1. Step 1Check the file for GPSOpen the original in your phone’s photo app or your computer’s file properties and look for coordinates. If they are there, you have your answer, and the rest is confirmation.
2. Step 2Run Lens on the whole imageOn a computer, go to google.com, click Search by image and upload the file. In Chrome you can also right-click an image on a page and choose Search with Google Lens, which opens the results in a side panel. Read the pages that carry the same or a similar picture, not just the first label.
3. Step 3Crop and search againDrag the selection box around one detail at a time: a tower, a sign, a shopfront, a mountain ridge. Each crop can lead somewhere different, and Google says a smaller area gives more specific results.
4. Step 4Read the textUse Lens to copy or translate signs and menus. A business name or street name is often the fastest route to a place. Search it, or select an address in Lens on Android or iPhone to open it in Maps.
5. Step 5Get an AI estimate for what is leftIf Lens returns only look-alikes, the scene has probably not been photographed and posted much. An AI estimate reads the clues instead. The GeoSpy extension adds Find where this photo was taken to Chrome’s right-click menu and shows the place, a confidence percentage and a map in the side panel.
6. Step 6Confirm on the mapTreat every answer, from Lens or from AI, as a candidate. Find the spot in Google Maps and Street View and match several independent details. Street View is dense in some countries and thin in others, so check where it has coverage first.

GeoSpy – Photo Location Finder is free to install from the Chrome Web Store. Photo analysis needs a geospys.com account with GeoSpy Premium, at US$7.99 a week or US$29.99 a year. If a JPEG still carries GPS data, the answer comes from that. Otherwise you get an AI estimate from the visible clues, with a confidence percentage. The extension can only talk to geospys.com and reads nothing until you right-click an image. The extension page has the details.

An estimate is a lead, not a fact, and GeoSpy is no exception. When a photo shows no usable clues, GeoSpy on geospys.com says that no location was found.

As for your own searches, Visual Search History, the setting that saves the images you search with to your Google Account, is off by default. When you use Lens on a web page in Chrome, a screenshot of the page and page data go to Google, which says it keeps them only temporarily to answer the query and does not use them for training.

Faces are a separate question. In its help page for Lens in YouTube Shorts, Google says Lens does not use biometric facial recognition, although well-known public figures can appear in results. Lens is a tool for places and objects. That is exactly why the background of a photo can matter more than the face in it, as a selfie taken at home shows.