The Provenance Check

Old pictures get reused for new events — a flood from five years ago, posted during tonight's storm. Checking a picture means finding its history: who posted it first, and when. A search tool can look for the same picture across the whole web and show you the earliest copy. If the picture is older than the event, the caption is false.

Provenance is an artifact's origin story: who created or first posted it, when, and where it was really taken. A viral image makes two claims — the pixels, and the caption — and the caption is the one that usually lies. Establishing provenance uses three outward moves: reverse image search to find the earliest appearance; metadata, when it survives, for camera, time, and place; and lateral reading — leaving the artifact to see what other, independent pages say about it.

Note what is not on the list: inspecting the image harder. Internal inspection can sometimes catch a clumsy edit. It can never certify origin — an authentic, unedited photograph miscaptioned by five years passes every pixel test and is still false.

Work the checks in order of cheapness. Save the artifact and its URL first — posts vanish mid-check. Then the earliest-appearance search across more than one engine. Then the tests the caption must survive: weather records for the claimed day, daylight and shadows for the claimed hour, signage, plates, and landmarks for the claimed place. Then laterally: have fact-checkers or local outlets already traced it? Each test works without trusting the poster — which is the point.

Metadata deserves calibrated trust. Intact EXIF can give camera model, timestamp, even coordinates — but major platforms strip it on upload, and it can be edited. Treat it as one witness among several, never as the verdict.

A photo cannot say when it was taken. Someone can share an old photo and call it new. The picture is real; the story about it is not.

So ask about any surprising picture: where did it come from first? Finding the first time it appeared is how news checkers test pictures too.

Verification practice formalizes this as five questions — provenance, source, date, location, motivation: are you looking at the original artifact; who captured it; when; where; and why is it being shared? The order matters because the later questions are unanswerable against a copy — a re-upload carries fresh metadata, a cropped frame hides landmarks. Establishing the original is the load-bearing first step, and folio 10 applies in full: a thousand copies of one photo are echoes of a single origin.

Synthetic media sharpens the method rather than changing it. Detection by artifact inspection is a contest the inspector eventually loses; provenance — where did this first enter the record, and does any independent trace corroborate the event it depicts — degrades far more slowly.

The infrastructure answer is cryptographic provenance: the C2PA standard binds capture-time assertions — device, time, edit history — into a signed manifest that travels with the file, shifting verification from forensic guesswork toward chain-of-custody audit. Its limits are adoption and the analog hole; for years yet, absence of credentials will remain uninformative.

For evidentiary-grade work, the Berkeley Protocol codifies open-source investigation: preserve originals with hashes, log the discovery path, geolocate and chronolocate against independent data, document uncertainty. The epistemology matches the reporter's method, held to a standard a court can audit — conclusions rest on converging independent traces, never on the artifact's own testimony.