C2PA content credentials
Signed metadata in an image can provide provenance details such as the issuer, generation model, and creation time. It offers richer context, but may be removed during sharing, conversion, or editing.
For PNG · JPG · WEBPUse this OpenAI image and audio checker to verify C2PA Content Credentials and SynthID watermarks from supported tools including ChatGPT, the OpenAI API, and Codex.
Upload a file
Verify one file at a timeor
PNG · JPG · WEBP
AudioMP3 · WAV · AAC · FLAC · OPUS · PCM
Signed metadata in an image can provide provenance details such as the issuer, generation model, and creation time. It offers richer context, but may be removed during sharing, conversion, or editing.
For PNG · JPG · WEBPAn invisible watermark embedded in the media itself. It may survive some compression or editing better than regular metadata, although major changes can still weaken the signal.
For images and audioA single file result is easier to interpret and matches the API verification flow.
Screenshots, crops, compression, metadata removal, and format conversion can weaken or remove provenance signals.
Upload a clear excerpt and keep the total decoded duration within 60 seconds.
Metadata may be removed during sharing or conversion, watermarks may be damaged by major edits, and older models may not include these signals. This tool only checks supported OpenAI provenance signals. It is not a universal AI detector and does not judge whether content is true.
Content provenance gives editors, platforms, creators, researchers, and the public more context about where media came from, helping people decide what to trust, share, or investigate further.
Verify a file ↑FREQUENTLY ASKED
08Answers are based on the OpenAI Verify and Content Provenance API documentation.It checks an uploaded file for supported OpenAI content provenance signals and reports C2PA content credential and SynthID digital watermark results separately. It does not guess the source from visual style.
It is designed for content generated or modified by supported OpenAI tools such as ChatGPT, the OpenAI API, and Codex. It currently supports PNG, JPEG, and WebP images, plus MP3, Opus, AAC, FLAC, WAV, and PCM audio.
It means the file very likely came from a supported OpenAI tool. A detected signal is strong provenance evidence, but it is not a complete file history and does not explain how the file was shared, used, or modified later.
Yes. When a user asks an OpenAI tool to modify a file and the model creates a new file reflecting that change, the new file may contain detectable provenance signals. This tool does not reconstruct every editing step.
No. Provenance signals only indicate whether supported evidence of OpenAI generation is present. They do not judge factual accuracy or completeness of context and should not be used alone for high risk decisions.
It only means no supported signal was found in this check. It does not prove the content was human made. Metadata may have been removed or altered, a watermark may be damaged, the file may come from an older model, or another company’s AI system may have created it.
Upload the original file whenever possible and verify one file at a time. Avoid screenshots, crops, and format conversion for images. For audio, use a clear 10 to 60 second excerpt and keep decoded duration within 60 seconds.
This site does not write your file to local disk. It sends the file directly to the OpenAI API for verification. Zero Data Retention does not apply to this API. See the OpenAI API data controls policy for details.