Guide · 9 min read · Updated September 2026

How to check a photo from breaking news before you share it

The most misleading images in circulation are usually real photographs. They are simply old, from somewhere else, or showing something other than what the caption claims — and recycling a genuine photo requires no AI at all. During a breaking story the supply of real imagery is thin and demand is enormous, which is precisely the gap that recycled and generated pictures fill. This is a method for the first ten minutes, before you forward anything.

By Ira Peoples

Start with the account, not the image

Before examining a single pixel, look at who posted it. Open the profile and check the creation date, the posting history, and whether the account has ever posted about this region or subject before. An account created three weeks ago that posts exclusively about one conflict, in several languages, at all hours, is an operation rather than a witness.

Look at what else they posted in the same hour. Genuine eyewitnesses post a messy sequence from one place: several angles, some badly framed, over a stretch of time. Accounts farming attention post one dramatic image from each of five unrelated events.

Run a reverse image search properly

Google Lens, TinEye, Yandex and Bing all index different corners of the web, and the answer is often in only one of them — so run at least two. TinEye is the most useful here because it can sort results by oldest first, which immediately shows whether a 'photo from this morning' has been online since 2019.

Crop before you search. Searching a distinctive fragment — a shopfront, a vehicle, a banner, an unusual building — often finds matches that a search on the whole frame misses, because cropping defeats the small alterations people make to evade exactly this check.

Geolocate using what cannot be moved

If a caption names a place, test it. Identify fixed features in the image: mountain profiles on the horizon, the shape of a distinctive roofline, road markings, the design of streetlights, the language and alphabet on signage, the side of the road traffic drives on, the style of licence plates, the species of trees.

Then open satellite and street-level imagery of the claimed location and look for the same arrangement. You are matching relationships, not objects: this tower to the left of that junction, this road curving in this direction. A single matching building proves little; three features in the correct spatial relationship is strong confirmation, and is how professional verification desks work.

Check the weather and the light

Historical weather records are free and public. If an image is claimed to be from a specific place on a specific date, compare the conditions in the frame — wet roads, snow, heavy cloud, bright sun — with what the record says. Contradictions are hard to explain away.

Shadows carry the time of day. Long shadows in a photo captioned as midday, or shadows falling in a direction inconsistent with the claimed hemisphere and season, place the image somewhere other than where you were told it was taken.

Look for the signs of a recycled picture

Watch for cropped-out watermarks along an edge, a hard border where a logo was cut away, and text or graphics with an aspect ratio that does not match the rest of the frame. Screenshots of screenshots accumulate these edges.

Seasonal and technological mismatches are easy wins: bare trees in a summer story, snow in a tropical location, a car model newer than the event, a phone or storefront brand that did not exist yet. Clothing suited to a different climate is another reliable one.

Then, and only then, consider whether it is generated

Provenance failures are far more common than synthetic images, which is why they come first. But when an image survives the checks above and still cannot be traced to any source, generation becomes a live possibility, especially for scenes that are conveniently dramatic and conveniently unattributed.

At that point run it through an authenticity check and read the reasons rather than the score. Our guide to spotting generated photos covers the visual tells, and the checker on this site will walk through the file-level ones — metadata, compression history, texture statistics — and tell you which came back inconclusive.

Decide what to do with an unresolved image

Most images will not resolve cleanly in ten minutes, and the correct action then is to do nothing. Not sharing is free; a wrong share during a live event reaches people who will never see the correction, and it is the corrections that travel slowly.

If you have already shared something that turns out to be wrong, delete it and post the correction separately rather than quietly editing. And if you find yourself certain because an image confirms what you already believed, that is the moment to run one more check, not one fewer.

The short version

Most misleading news images are real photographs shown out of context, not AI fabrications.
Check the account's age and posting history before you examine the image itself.
Run reverse image search in at least two engines, sorted oldest-first, and search cropped details.
Geolocate by matching three fixed features in the right spatial relationship, not one building.
If ten minutes leaves it unresolved, don't share. Silence costs nothing.