agentage.news / Articles / Checklist

Checking the Provenance of Images and Video Before Publishing

A practical checklist for journalists and editors to trace the origin of visual media and identify deepfakes before publication.

By Zoltán Varga2026-11-054 min read

Ask for the Original File and Inspect Content Credentials

Request the original, unedited file from the source before making any judgment. Social media platforms often strip metadata and cryptographic signatures during compression, rendering verification impossible on the posted version. A file shared via email or secure transfer retains the necessary digital fingerprints for analysis.

Look for the small 'cr' icon in the corner of the image or video player. This symbol indicates the presence of Content Credentials, a standard developed by the Coalition for Content Provenance and Authenticity (C2PA). Clicking this icon reveals a signed history of creation, editing, and device information.

Use the official Content Credentials Verify tool to check the cryptographic signature. This process confirms whether the file has been altered since the camera captured it. If the signature is missing or invalid, the file lacks a verifiable chain of custody and requires further scrutiny.

Understand that signing currently outpaces verification on many platforms. Major social networks like Facebook and X often remove these credentials upon upload. Always demand the raw file directly from the photographer or agency to access the full provenance data.

• Check for the 'cr' icon indicating C2PA credentials.
• Verify the signature using the official online tool.
• Ensure the file is the original, uncompressed version.
• Review the edit history for unauthorized modifications.

Read Metadata and Perform Reverse Image Searches

Examine the EXIF data within the file to identify the camera model and capture date. This information helps confirm if the device existed at the claimed time and location. Inconsistencies here often signal manipulation or the use of stock imagery repurposed for a new event.

Run the image through reverse search engines like Google Lens or TinEye. These tools map where the image has appeared online previously. A photo claiming to show a current conflict might actually be from a decade-old archive or a different geographic region entirely.

Apply geolocation techniques to match visual landmarks with satellite imagery. Verify street signs, building architecture, and vegetation against known locations. This manual step remains essential when digital metadata is missing or has been scrubbed.

Be aware that metadata can be easily spoofed by malicious actors. While a timestamp might match the event, the camera serial number could be fabricated. Treat metadata as a clue rather than definitive proof without corroborating visual evidence.

• Extract EXIF data to check camera model and date.
• Use reverse search to find earlier instances of the image.
• Match visual landmarks to known geographic locations.
• Cross-reference timestamps with local weather or events.

Understand Why AI Detectors Are Not Proof

Do not rely on AI detection tools as the sole basis for your editorial decision. Current detectors struggle to distinguish between human errors and machine generation, often flagging non-native speakers as AI-generated content. False positives occur frequently, penalizing legitimate creators.

Research indicates that humans perform no better than a coin toss when identifying deepfakes visually. A 2025 study found that only a tiny fraction of participants correctly identified all synthetic media. This limitation extends to automated systems, which face similar accuracy ceilings.

The prevalence of AI-generated content creates a feedback loop that confuses detection models. As AI trains on AI output, the distinction between human and machine text or image blurs further. Detection accuracy drops significantly as the volume of synthetic media increases.

Avoid citing detector scores as evidence of authenticity or fabrication. These tools provide probabilistic guesses, not forensic facts. Their limitations are well-documented, and their results should never override direct provenance verification or human investigation.

• Detectors have high false-positive rates for non-native speakers.
• Human visual identification is statistically unreliable.
• AI training on AI output degrades detection accuracy.
• Detector scores are probabilities, not forensic conclusions.

Document the Provenance Decision Process

Record every step taken to verify the media, including tools used and results obtained. This documentation protects your publication if the content is later challenged. It demonstrates due diligence and a commitment to transparency in the reporting process.

Note the absence of Content Credentials or the failure of signature verification. Clearly state whether the file was sourced from a social platform or received directly. Explain how you traced the image back to its earliest known appearance online.

Include links to the verification tools and the original source files in your internal records. This allows other editors to replicate the check or update the assessment if new information emerges. A clear audit trail builds trust with your audience.

Publish a brief explanation of your verification method alongside the story when appropriate. Readers appreciate knowing how you confirmed the visual evidence. This practice aligns with the principles of the Starling Framework for data integrity in journalism.

• Log all verification steps and tool outputs.
• Specify the source of the original file.
• Archive links to verification results and source images.
• Explain the verification method to your readers.

AI-assisted: drafted with a language model from the author's research notes, then edited and checked before publication. Author: Zoltán Varga, AI visibility researcher and AI ecosystem architect, Budapest.