Image and video verification checklist

What to take away

• Check the attached claim as well as the file.
• Preserve the best available version and all relevant publication context.
• Record independent findings for source, place, time, and editing.
• Treat provenance and detector output as evidence with defined limits.
• Review safety, consent, identification, and credit before publishing.

Use this checklist for photographs, video, thumbnails, frame grabs, screenshots, composites, illustrations, and generated media. Add the reviewer, date, evidence location, and result beside each item. Marking "unresolved" is better than checking a box without proof.

Intake and preservation

• Exact URL, account, caption, upload time, and access time saved
• Highest-quality lawful file collected
• Original received file preserved and working copies created
• Platform preview, thumbnail, download, and camera-native file distinguished
• Redirects, deleted pages, edits, and archive captures recorded
• File hash recorded when the risk or workflow warrants it
• Graphic content and sensitive identifiers access-controlled

Claim and source

• Claimed creator, uploader, place, date, event, and people separated
• Camera-recorded, staged, edited, composite, simulated, and generated possibilities considered
• Creator or uploader contacted through a confirmed route where feasible
• Source explains how the file was captured, obtained, and transferred
• Full sequence, original file, and adjacent material requested
• Source access, motive, safety, consent, and conflicts assessed

Origin and prior circulation

• Full-frame reverse searches run
• Distinctive crops and video keyframes searched
• Exact captions, watermarks, signs, and usernames searched
• Mirrored, rotated, cropped, and higher-resolution variants compared
• Earliest confirmed public appearance recorded
• Page dates distinguished from file creation and upload dates

Metadata and signed provenance

• Metadata extracted from the best file without modifying it
• Time zone, clock accuracy, coordinates, software, and encoding checked
• Fields compared with visual evidence and source account
• Missing metadata not treated as proof of manipulation
• Content Credential or other signed provenance validated when present
• Signer identity, assertions, edits, ingredients, and asset binding reviewed

The C2PA Content Credentials explainer describes a cryptographically bound structure for recording asset provenance. It also states that provenance alone cannot tell whether visual content is true or factual and may be incomplete. A valid credential can support origin and edit-history findings without replacing scene verification.

Location

• Language, signs, roads, utilities, terrain, buildings, and skyline inventoried
• Observations kept separate from geographic inferences
• Several stable features matched against independent imagery
• View direction, camera height, lens effects, and later changes considered
• Alternative candidates tested and rejected with reasons
• Precision limited to country, area, street, or viewpoint as evidence allows
• Publication of location reviewed for safety

Time

• Earliest circulation and any source timestamp compared
• Local time zone and daylight-saving rules checked
• Daylight, shadows, weather, vegetation, clothing, and clocks assessed
• Construction, signage, event, transport, and archive changes used to bracket time
• Other independent recordings from the event compared
• Result stated as exact time, range, consistency finding, or unresolved

Video and audio

• Full duration reviewed at normal and slow speed
• Keyframes extracted from meaningful intervals and cuts
• Speech and visible text transcribed and translated carefully
• Audio continuity, synchronization, and environmental sound assessed
• Cuts, loops, speed changes, overlays, and missing lead-in noted
• Longer and alternate-angle recordings sought

WITNESS's metadata capture guidance identifies date and time, geographic location, event description, and the person who recorded a video as key contextual information to capture at creation. A newsroom receiving footage later should seek those fields from the creator and explain any gaps rather than reconstructing them as facts.

Editing and forensic review

• Crop, color, stabilization, redaction, compression, and captioning distinguished
• Material effect of each edit assessed
• Technical tool, version, settings, input, and output saved
• Detector limitations and suitable comparison data recorded
• Specialist review obtained for a high-risk technical conclusion
• No single artifact or model score presented as conclusive

Publication and safety

• Caption says what is confirmed and what remains unresolved
• Illustration, simulation, reenactment, composite, and generation labeled
• Creator credit and license checked at the exact asset page
• Consent, dignity, identification, minors, victims, and location risk reviewed
• Headline, thumbnail, alt text, social crop, and video poster frame checked
• Evidence package and final publication version preserved
• Correction plan covers every distributed copy

Common questions

Is every item required for every file?

No. Record why an item does not apply and complete the checks that affect the claim and risk.

Can a screenshot be verified?

It can be linked to a source or event, but it lacks much of the file and page context needed for stronger conclusions.

Should metadata be published?

Only fields needed for the story and safe to disclose. Coordinates and device details can expose people.

What does a final status look like?

Use component findings such as "source confirmed, location confirmed, date narrowed to two days, editing unresolved."