ChatGPT

At a glance

Images

Content CredentialsA signed record (C2PA) of how the file was made, sealed so any change shows.

OpenAI's provenance help article, updated October 12, 2026, says images generated with ChatGPT include Content Credentials (C2PA) and a SynthID watermark, though coverage may vary by product, model, export path, file type and when the content was created. OpenAI began adding Content Credentials to DALL-E 3 images in 2024, and in March 2025 said all GPT-4o images carried C2PA metadata.

OpenAI does not name the signer for ChatGPT images. Its verification page says Content Credentials rely on metadata and can be removed.

Provenance signals in OpenAI-generated contentOpenAI (OpenAI Help Center)

Quoted passage Which provenance signals does OpenAI use? (table row: Images; note beneath the table)

Images Content Credentials and SynthID watermarks Images generated with ChatGPT, Codex, or the OpenAI API include both signals. […] Coverage may vary by product, model, export path, file type, and when the content was created. API outputs are covered except for models planned for imminent deprecation, including gpt-image-1, gpt-image-1.5, and sora-2.

Advancing content provenance for a safer, more transparent AI ecosystemOpenAI

Quoted passage Building the trust ecosystem through C2PA conformance (para 1)

OpenAI has been engaged in the development and adoption of provenance standards since 2024, when we began adding Content Credentials to images generated by DALL·E 3 […] We also joined the Steering Committee of the Coalition for Content Provenance and Authenticity (C2PA), the cross-industry group behind the open technical standard for content provenance.

Introducing 4o Image GenerationOpenAI

Quoted passage Provenance via C2PA and internal reversible search

All generated images come with C2PA […] metadata, which will identify an image as coming from GPT‑4o, to provide transparency. […] We’ve also built an internal search tool that uses technical attributes of generations to help verify if content came from our model.

Verify OpenAI-generated contentOpenAI

Quoted passage What supporting evidence does the tool use? (Content Credentials paragraph)

For supported OpenAI-generated content, they can indicate that a file was created using tools such as ChatGPT, Codex, or the API. They are a strong signal when present, but because they rely on metadata, they can be removed.

Invisible watermarkA signal woven into the pixels, frames, audio or wording, meant to survive copying.

OpenAI says images generated with ChatGPT include a SynthID watermark from Google DeepMind. It announced SynthID on May 19, 2026, starting with images, to complement C2PA metadata.

OpenAI's pages differ on robustness. Its verification page says SynthID is designed to stand up to cropping, filters and lossy compression; its help article says heavy cropping or file compression can make a watermark undetectable.

Provenance signals in OpenAI-generated contentOpenAI (OpenAI Help Center)

Quoted passage Which provenance signals does OpenAI use? (table row: Images; note beneath the table)

Images Content Credentials and SynthID watermarks Images generated with ChatGPT, Codex, or the OpenAI API include both signals. […] Coverage may vary by product, model, export path, file type, and when the content was created. API outputs are covered except for models planned for imminent deprecation, including gpt-image-1, gpt-image-1.5, and sora-2.

Advancing content provenance for a safer, more transparent AI ecosystemOpenAI

Quoted passage A multi-layered approach to provenance with Google SynthID for images

To make provenance more resilient, we are taking a multi-layered approach and incorporating watermarking through Google DeepMind’s SynthID […] SynthID embeds an invisible watermarking layer that complements C2PA metadata-based approaches.

Verify OpenAI-generated contentOpenAI

Quoted passage What supporting evidence does the tool use? (SynthID paragraph)

Unlike metadata, the signal is part of the content itself and is designed to stand up to modifications like cropping, adding filters, or lossy compression.

Provenance signals in OpenAI-generated contentOpenAI (OpenAI Help Center)

Quoted passage How robust are provenance signals against content modifications? (para 1)

SynthID watermarks for images and audio, and textGrain watermarks for text, are designed to withstand common changes. Depending on the content type, these can include light edits, copying and pasting, cropping, or screenshots. More extensive changes—such as heavy image cropping, file compression, substantial paraphrasing, or translation—can make a watermark undetectable. After extensive changes, describing the result as “OpenAI-generated” may also be less meaningful.

AI label in metadataA plain metadata field (IPTC Digital Source Type) saying AI made the content. Anyone can add or remove it.

Not stated. None of OpenAI's pages listed under Sources read says so.

Visible markA mark you can see, such as a logo or a sparkle in a corner.

Not stated. None of OpenAI's pages listed under Sources read says so.

OpenAI's Images in ChatGPT help page and provenance article describe no visible mark. OpenAI says provenance signals do not replace visible labels or other notices that may be required.

Tool named in metadataThe tool or model written into ordinary file fields, such as Software or Creator.

In 2024 OpenAI said that when a user edits a DALL-E 3 image in ChatGPT, the Content Credential records the edit, including the app and tool used (OpenAI's DALL-E) and the actions taken.

OpenAI does not publish a current list of fields.

Understanding the source of what we see and hear onlineOpenAI

Quoted passage Expanding our image detection tools as we build new features (August 4, 2024 update)

As users can now edit DALL-E 3 generated images in ChatGPT, we wanted to ensure that provenance information continues to be demonstrated along with those edits. If a user edits an image, we’ve built in a means for our C2PA credential to show that the image was edited and how. In the example image, you can see the original AI-generated picture of a caterpillar, then edited to have the caterpillar wearing a Santa hat. When a user edits this image, the C2PA credentials update to reflect these changes, ensuring transparency. The C2PA credential includes details like the app and tool used (OpenAI’s DALL-E), the actions taken (like format conversion and edits), and other modifications.

Audio

Content CredentialsA signed record (C2PA) of how the file was made, sealed so any change shows.

Not stated. None of OpenAI's pages listed under Sources read says so.

OpenAI's provenance article lists only SynthID for audio, and its API checks audio for SynthID only.

Invisible watermarkA signal woven into the pixels, frames, audio or wording, meant to survive copying.

OpenAI says audio generated by ChatGPT includes an inaudible SynthID watermark embedded in the audio itself. Its update of July 31, 2026 says supported audio from its tools now includes SynthID.

Provenance signals in OpenAI-generated contentOpenAI (OpenAI Help Center)

Quoted passage Which provenance signals does OpenAI use? (table row: Audio)

Audio SynthID watermarks Audio generated by ChatGPT or the OpenAI API includes an inaudible watermark embedded in the audio itself.

Advancing content provenance for a safer, more transparent AI ecosystemOpenAI

Quoted passage Update July 31, 2026 (page header)

Update July 31, 2026: We’re expanding this work beyond images. Supported audio generated with OpenAI tools, including through ChatGPT and the OpenAI API, now includes SynthID watermarking. Our public verification tool will now allow for verification of supported audio files in addition to images.

AI label in metadataA plain metadata field (IPTC Digital Source Type) saying AI made the content. Anyone can add or remove it.

Not stated. None of OpenAI's pages listed under Sources read says so.

Visible markA mark you can see, such as a logo or a sparkle in a corner.

Not stated. None of OpenAI's pages listed under Sources read says so.

Tool named in metadataThe tool or model written into ordinary file fields, such as Software or Creator.

Not stated. None of OpenAI's pages listed under Sources read says so.

Text

Content CredentialsA signed record (C2PA) of how the file was made, sealed so any change shows.

Not stated. None of OpenAI's pages listed under Sources read says so.

Invisible watermarkA signal woven into the pixels, frames, audio or wording, meant to survive copying.

OpenAI's provenance article, dated October 12, 2026, says text generated by ChatGPT in the EU carries a textGrain watermark, an invisible change to the statistical pattern of word choices that inserts no hidden characters or watermark-only tokens. On October 5, 2026 OpenAI said it would add the watermark to eligible ChatGPT and Codex text in the EU over the coming weeks.

Provenance signals in OpenAI-generated contentOpenAI (OpenAI Help Center)

Quoted passage Which provenance signals does OpenAI use? (table row: Text)

Text textGrain watermarks (EU only) In the EU, ChatGPT-generated text includes an invisible change to the randomness used in the model’s word choices to comply with the EU AI Act. API customers globally can turn on text watermarks to text outputs, from their project or organization settings.

Provenance signals in OpenAI-generated contentOpenAI (OpenAI Help Center)

Quoted passage Does text watermarking add hidden characters, extra tokens, or visible changes?

Text watermarking changes the statistical pattern of word choices; it does not insert hidden characters or add watermark-only tokens. The watermark is not visible to readers, and copying and pasting the text does not introduce hidden material.

Our approach to EU text provenance rulesOpenAI

Quoted passage Introduction (phased approach list)

Starting today, API customers globally will be able to opt in to text watermarking for select models. Text watermarking will remain off by default in the API. Over the coming weeks, we will add an invisible watermark to eligible ChatGPT and Codex text output in the European Union.

AI label in metadataA plain metadata field (IPTC Digital Source Type) saying AI made the content. Anyone can add or remove it.

Not stated. None of OpenAI's pages listed under Sources read says so.

Visible markA mark you can see, such as a logo or a sparkle in a corner.

Not stated. None of OpenAI's pages listed under Sources read says so.

Tool named in metadataThe tool or model written into ordinary file fields, such as Software or Creator.

Not stated. None of OpenAI's pages listed under Sources read says so.

Documents and code

Content CredentialsA signed record (C2PA) of how the file was made, sealed so any change shows.

Not stated. None of OpenAI's pages listed under Sources read says so.

OpenAI's ChatGPT Work help page and provenance article say nothing about Content Credentials for documents, spreadsheets or presentations.

Invisible watermarkA signal woven into the pixels, frames, audio or wording, meant to survive copying.

Not stated. None of OpenAI's pages listed under Sources read says so.

OpenAI does not say whether files ChatGPT creates carry a watermark. It says code is harder to watermark and that short passages are hard to detect.

AI label in metadataA plain metadata field (IPTC Digital Source Type) saying AI made the content. Anyone can add or remove it.

Not stated. None of OpenAI's pages listed under Sources read says so.

Visible markA mark you can see, such as a logo or a sparkle in a corner.

Not stated. None of OpenAI's pages listed under Sources read says so.

Tool named in metadataThe tool or model written into ordinary file fields, such as Software or Creator.

Not stated. None of OpenAI's pages listed under Sources read says so.

Limits and detection

Checking tool from the makerA way the maker offers to check a file for its traces.

OpenAI's verification page at openai.com/verify checks supported image and audio files for a SynthID watermark or a trusted C2PA manifest, and the Content Provenance API runs the same checks. Text detection is available by application to qualifying organizations, such as academic and research organizations, and checking student work is outside its scope.

OpenAI says a detected signal means content likely came from its tools and that false positives are rare, but content may still be OpenAI-generated when no signal is found.

Provenance signals in OpenAI-generated contentOpenAI (OpenAI Help Center)

Quoted passage OpenAI's verification tools: How can I check whether content was generated with OpenAI tools?

You can visit openai.com/verify to check whether a supported image or audio file, contains provenance signals associated with content generated or exported by OpenAI tools. After you upload a supported file, the tool will look for supported OpenAI provenance signals, such as a SynthID watermark or a trusted C2PA manifest associated with OpenAI. If the tool finds a supported OpenAI provenance signal, it indicates that the content was likely generated by an OpenAI model. Developers and organizations can also use OpenAI’s Content Provenance API to check images and audio for OpenAI provenance signals and integrate verification into their own applications or workflows.

Provenance signals in OpenAI-generated contentOpenAI (OpenAI Help Center)

Quoted passage OpenAI's verification tools (text detector access); Who qualifies for text detector access

For text, consistent with our obligations under the EU AI Act, organizations with a qualifying use case, such as academic and research organizations studying text provenance, detection reliability, or how people understand provenance results, are able to apply for text detector access. We will review applications from qualified organizations on a case by case basis […] Academic integrity applications, such as checking student work for AI use (e.g. in writing essays or theses), are also currently outside the program scope.

Verify OpenAI-generated contentOpenAI

Quoted passage What does this tool do?; What content can it detect?

This tool checks whether an uploaded file contains provenance signals associated with OpenAI tools. It looks for supported signals, including C2PA metadata and SynthID watermarks, and reports whether they are detected. […] The tool is designed to detect content generated with ChatGPT, the OpenAI API, or Codex. It currently supports images and audio files.

Verify OpenAI-generated contentOpenAI

Quoted passage What does a detected signal mean?; What if no signal is found?

A detected signal indicates that the content likely originated from OpenAI tools. Detected signals are reliable, and false positives are rare. […] The content could still have been generated by OpenAI if its metadata was stripped or tampered with, its watermark was degraded, it came from a legacy generation model, or it was created before provenance signals were available.

What the maker says about limitsIn the maker's own words.

OpenAI says its SynthID and textGrain watermarks are designed to withstand common changes, but that heavy cropping, compression, substantial paraphrasing or translation can make a watermark undetectable.

Provenance signals in OpenAI-generated contentOpenAI (OpenAI Help Center)

Quoted passage How robust are provenance signals against content modifications? (para 1)

SynthID watermarks for images and audio, and textGrain watermarks for text, are designed to withstand common changes. Depending on the content type, these can include light edits, copying and pasting, cropping, or screenshots. More extensive changes—such as heavy image cropping, file compression, substantial paraphrasing, or translation—can make a watermark undetectable. After extensive changes, describing the result as “OpenAI-generated” may also be less meaningful.

What the maker says about limitsIn the maker's own words.

OpenAI says metadata may be removed by platforms, editing tools and file conversions, and that Content Credentials can be lost through screenshots, uploads and downloads, resizing or format changes.

Provenance signals in OpenAI-generated contentOpenAI (OpenAI Help Center)

Quoted passage Why do some content types use multiple signals?; How robust are provenance signals (para 2)

Metadata can carry detailed information, but platforms, editing tools, and file conversions may remove it. […] Content Credentials, developed by the Coalition for Content Provenance and Authenticity , can be lost through everyday actions, such as taking a screenshot or converting a file.

Advancing content provenance for a safer, more transparent AI ecosystemOpenAI

Quoted passage A multi-layered approach to provenance with Google SynthID for images (para 1)

But metadata is not foolproof. It can be stripped, lost through uploads and downloads, or broken by transformations like file format changes, resizing, or screenshots.

Verify OpenAI-generated contentOpenAI

Quoted passage What supporting evidence does the tool use? (Content Credentials paragraph)

For supported OpenAI-generated content, they can indicate that a file was created using tools such as ChatGPT, Codex, or the API. They are a strong signal when present, but because they rely on metadata, they can be removed.

What the maker says about limitsIn the maker's own words.

OpenAI says the absence of a detected text watermark does not prove human authorship, and that short passages usually lack enough text for reliable detection. In its own evaluation at a 1% target false positive rate, its detector found watermarks in about 80% of 200-token and about 95% of 400-token passages for content such as psychology, with lower rates for content such as mathematics. In a separate evaluation of 400-token passages, replacing 10% or 25% of words with synonyms cut detection from about 92% to 66% or 17%.

Our approach to EU text provenance rulesOpenAI

Quoted passage What a text watermark doesn't tell you

The absence of a detected watermark does not prove human authorship. Text generated with OpenAI tools may be too short, edited, or translated for detection to work reliably. It may also come from an unsupported model, predate watermarking, or have been generated by another company’s tools.

Our approach to EU text provenance rulesOpenAI

Quoted passage How our watermarking works and performs

Shorter or more constrained text is harder to detect. At a target false positive rate of 1%, our detector identified watermarks in about 80% of 200-token passages, compared with about 95% of 400-token passages, for content such as psychology. Detection rates were substantially lower for content such as mathematics, where there is less flexibility in word choice. […] Editing can weaken the watermark. In an evaluation of 400-token passages, replacing 10% of words with synonyms reduced detection from about 92% to 66%. Replacing 25% of words reduced it to 17%.

Provenance signals in OpenAI-generated contentOpenAI (OpenAI Help Center)

Quoted passage Do text watermarks apply to short answers and code?

Short passages usually do not contain enough text for reliable watermark detection. Code is also harder to watermark because there are fewer plausible choices for what comes next than in ordinary prose.

What the maker says about limitsIn the maker's own words.

OpenAI says watermarks and C2PA metadata do not replace visible labels, banners or other notices that may be required.

Provenance signals in OpenAI-generated contentOpenAI (OpenAI Help Center)

Quoted passage Do provenance signals replace visible AI labels or other disclosures?

No. Watermarks and C2PA metadata provide information that software can read. They do not replace visible labels, banners, or other notices that may be required.

What the maker says about limitsIn the maker's own words.

OpenAI says its provenance signals include no details about the user, organization or prompt.

Provenance signals in OpenAI-generated contentOpenAI (OpenAI Help Center)

Quoted passage Do results show my prompt or personal information?

Provenance signals do not include details about the user, organization, or prompt.