How to Verify AI Media with Content Credentials and SynthID

Problem

Anyone who scrolls through social feeds or reads a news article now faces a familiar doubt: Is this piece created by a human, or by an AI? The uncertainty fuels misinformation, erodes confidence in digital media, and makes it hard for publishers to protect their brand. OpenAI acknowledges the gap and, on May 19, 2026, announced a suite of provenance tools—Content Credentials, SynthID, and a verification service—designed to let users identify the source of AI‑generated media (OpenAI Blog, 2026-05-19). Without a clear way to trace content back to its creator, readers and platforms remain vulnerable to deepfakes, misattributed quotes, and other forms of manipulation.

Prerequisites

Before you start, gather the following:

• The AI‑generated file you want to check (image, audio, video, or text).
• Access to OpenAI’s verification tool, which was introduced in the May 19 announcement. The tool works as a web interface or API endpoint that can read embedded provenance data.
• A modern browser or a command‑line environment capable of sending HTTP requests to the verification service.

All three items are mentioned in the OpenAI blog post that launched the provenance initiative (OpenAI Blog, 2026-05-19). No additional software or third‑party plugins are required.

Steps

1. Locate the Content Credentials. OpenAI embeds a small metadata packet—called Content Credentials—directly into the media file. For images, this appears as an EXIF block; for audio and video, it lives in the container’s metadata section. The credentials include the model name, generation timestamp, and a cryptographic hash.
2. Extract the SynthID fingerprint. SynthID is a unique, invisible watermark that OpenAI adds at the pixel‑level (or waveform‑level for audio). The fingerprint can be read by the verification tool without degrading the user experience. If the file carries a SynthID, the tool will return a matching identifier that ties the asset back to the originating model.
3. Run the verification tool. Upload the file to the verification portal or call the API with the file’s binary data. The service parses the Content Credentials, reads the SynthID, and cross‑checks both against OpenAI’s provenance ledger. The response includes:
   • Model provenance (e.g., DALL·E 3, ChatGPT‑4o).
   • Generation date and time.
   • Verification status (authentic, altered, or unknown).
4. Interpret the result. A “authentic” flag means the media matches the original credentials and SynthID, confirming it was generated by the claimed OpenAI model. An “altered” flag indicates the file was edited after generation, which may affect trust. An “unknown” status suggests the file lacks proper credentials—proceed with caution.
5. Document the outcome. For journalists, researchers, or platform moderators, record the verification JSON output. This audit trail can be attached to the story or content moderation ticket, ensuring transparency for downstream audiences.

Pro Tips

• Leverage trusted news partnerships. OpenAI’s May 25 partnership with Brazil’s Grupo Folha and Grupo UOL integrates Content Credentials into their news feeds, guaranteeing that stories delivered via ChatGPT carry built‑in attribution (OpenAI Blog, 2026-05-25). When you encounter a news snippet from these outlets, the provenance data is already verified, saving you a step.
• Batch‑process large collections. Use the API version of the verification tool to script bulk checks. Loop through a folder of images, send each to the endpoint, and collect the JSON responses in a spreadsheet for quick review.
• Combine with human judgment. While the verification tool provides technical assurance, always cross‑reference the content with reputable sources. The partnership with Brazilian news groups demonstrates that provenance works best when paired with editorial standards.
• Stay updated on tool upgrades. OpenAI’s provenance ecosystem is evolving. Subscribe to the OpenAI blog for announcements about new model support or expanded metadata fields.

By following these steps, you can cut through the noise and confirm whether a piece of media truly originates from an AI model you trust. The result is a safer, more transparent digital environment for creators, publishers, and consumers alike.