What happens when you search by image
A reverse image engine does not compare pixels one by one. It reduces each picture it has crawled to a compact description of shapes, colours and textures, does the same for the one you submit, and returns the indexed pictures whose descriptions sit closest. That is why a resized, recompressed or lightly cropped copy still matches, and why a new photo of the same scene may not.
Two styles exist. An exact-match engine such as TinEye looks for the same image in altered forms and can sort results by when it first saw them. Similarity engines such as Google Lens, Bing and Yandex also return pictures that merely look alike, with guesses about the object, place or product shown. For tracing a source you want the first kind. For identifying a thing, the second.
Getting a direct image address
The URL box needs the address of the image file itself, not of the page it sits on. In a desktop browser, right-click the picture and choose Copy image address or Copy image link, depending on the browser. On a phone, press and hold the picture for the same option. Many addresses end in .jpg, .png or .webp, though plenty of sites serve images without an extension.
Some addresses will not work. Anything starting with blob: or data: exists only inside your browser. Pictures behind a login, on a company intranet or in a private message cannot be fetched by an outside server, and some sites refuse requests that do not come from their own pages. In those cases save the picture and use the device upload card.
Searching with a file from your device
Choosing a file here does one thing: it shows a preview with the file name, pixel dimensions and size in kilobytes, so you can confirm you picked the right picture and that it is large enough to match well. The file stays in your browser. Search engines do not accept uploads passed on by other websites, which is why local files have no single Search button.
Click Upload there next to an engine and its own search page opens in a new tab. Drag the file onto it or use its upload control. The largest, least edited version you have works best. Crop away borders, captions and collage neighbours so the engine sees one subject, and if a screenshot is all you have, trim the interface around the picture first.
Reading the results across engines
Start with the oldest and the largest copies. The earliest appearance is the best candidate for the original, and a full-size version usually predates the thumbnails made from it. Check the surrounding page: a photographer's portfolio or an agency credit carries more weight than an aggregator. When two sites both claim the picture, WHOIS lookup shows which domain was registered first.
For verification, look for the same photo attached to a different story. A picture said to show yesterday's event that also appears in an article from years earlier settles the matter. For a suspicious profile, matches under another name, on stock photo sites or in a model's portfolio are the usual sign of a borrowed identity. If the picture contains writing, image to text extracts it for a text search.
Limits of the method
No result does not mean the picture is original. Engines only know pages they have crawled, and most of what people post to private accounts, messaging apps and closed groups is never indexed. Mirrored, heavily filtered or redrawn copies may escape an exact-match engine, and a generated image has no earlier source to find.
Searching for a person carries responsibility. A match shows where a photo has been published, which can identify someone who never agreed to be found, and data protection or harassment laws apply in many countries. Keep it to checking that a profile, seller or source is what it claims to be.