Verifying Claims, Photos and Video

Verification is the work of checking that something is what it claims to be before you repeat it. A claim gets checked against records and independent sources. A photo or video gets checked for where it came from, where it was taken, and when. Speed is the enemy. Most false images spread because people shared first and looked second.

Claims

The method for a claim is the one from the primary records lesson: go upstream to the original. Add one standard. Before you state something as fact, you want either a document that shows it or two independent sources who know it firsthand. Independent means they didn't get it from each other or from the same third person.

If you can't meet that standard but the claim matters, you can still report that someone made it. Say who, say what they base it on, and say what you couldn't confirm. "Commissioner Park said the contract was never put out for bid. The district had not provided bid records by Friday" is honest. "The contract was never put out for bid" is a fact you don't have yet.

Photos

Step 1: reverse image search

The most common fake photo is a real photo from somewhere else or some other year. A reverse image search finds earlier copies. Google lets you search with an image from the camera icon in the search box or by right-clicking an image in Chrome (Google help). Bing Visual Search works similarly. Try more than one engine, since they index different parts of the web.

Look for the earliest date. If the "storm this morning" photo shows up in a 2021 news story from another state, you're done. If you find nothing, that doesn't prove it's new. It means you move to the next step.

Tips that help:

• Crop to the most distinctive part (a building, a sign) and search again.
• Flip the image horizontally. Recyclers sometimes mirror photos to dodge matching.
• For video, grab a few still frames and search those. The free InVID-WeVerify plugin pulls keyframes and runs searches from them.

Step 2: metadata

Cameras and phones store data inside each photo, usually in a format called EXIF: date and time, camera model, settings, and often GPS coordinates. You can read it with ExifTool, a free program, or with many photo viewers under "properties" or "info."

Cautions before you trust it:

• Missing metadata proves nothing. Many social platforms and messaging apps strip most of it when a photo is uploaded. Ask the person who took it to send the original file by email or a file link.
• Metadata can be edited. It's easy to change. Treat it as one clue.
• Camera clocks are wrong all the time. Wrong time zone, never set, off by an hour after daylight saving.

Some images now carry Content Credentials, a signed record based on the C2PA standard showing what device or software made the image and what edits followed. You can upload a file to Content Credentials Verify to read it. Most images don't have credentials yet, and stripping them is easy, so absence tells you little. Presence is useful evidence.

The same caution runs the other way. Photos you take carry your own GPS location. Before you publish your own photos or send them to anyone, strip the metadata. AVPrep has a metadata strip feature for this.

Step 3: geolocation

Geolocation means finding where a photo was taken from what's visible. You're looking for things that don't move.

• Signs: store names, street signs, billboards, phone numbers with area codes.
• Buildings: distinctive rooflines, church steeples, water towers, grain bins.
• Terrain: the shape of a mountain ridge or a river bend.
• Road details: lane markings, highway signs, which side traffic drives on.

Then match them using map satellite views, street-level imagery and Google Earth. Line up two features that should sit in a certain relation, like a steeple to the left of a water tower, and see if a spot on the map produces that view. It's slow at first. You get faster.

Step 4: chronolocation

Working out when is harder. Clues: weather (was it actually snowing that day?), leaves on the trees, store signs that changed, a car model year, and shadows. SunCalc shows where the sun sits at any place, date and time, so you can check whether shadow length and direction match the claimed time.

AI-generated images and audio

Generated images are now good enough that looking for six fingers isn't a reliable test. Treat a striking image with no traceable original source as unconfirmed, whatever it looks like. The workflow above still applies: find the first poster, ask for the original file, check for Content Credentials, and geolocate. A real scene can be found on a map. A generated one usually can't.

Watch sources before the story breaks

Much verification is easier if you've been watching the right sources all along. You'll know a photo is new if you saw the agency post about the flooding an hour earlier. OSINT Monitor collects public feeds into watch lists so you can keep an eye on agencies, news outlets and topics in one place, and pull up what was posted on a subject in a given time window.