These are the questions that come up most often about image provenance, answered briefly. Each answer links to the page that covers the topic in depth, and the glossary defines the terms. The answers reflect the state of the field in September 2026. Platform behaviour and product availability in particular change often.
The basics
What is image provenance?
Image provenance is the documented history of an image: where it came from, who created it, and what has been done to it since. In digital media the term usually means provenance recorded in a tamper-evident, cryptographically signed form rather than in editable metadata. It is different from detection, which guesses an image's origin from its pixels. See provenance vs. authentication vs. detection.
What is C2PA?
C2PA is the Coalition for Content Provenance and Authenticity, founded in 2021, and also the name of the open technical specification it publishes. The specification defines how to attach a signed manifest to an image, video, audio, or document file, recording who produced it and how it was edited. It is the most widely adopted provenance standard, used by camera makers, AI generators, and editing software. See the C2PA standard: overview.
What are Content Credentials?
Content Credentials is the consumer-facing name for C2PA provenance data. It covers the manifest embedded in a file and the tools and interfaces that display it. In practice, "this image has Content Credentials" means "this file carries a C2PA manifest." See how Content Credentials work, end to end.
What does the CR icon mean?
The small "cr" pin is the official Content Credentials icon, introduced by C2PA in October 2023. It indicates that the image carries provenance information you can inspect, usually by hovering over or clicking the icon. It does not mean the image is authentic, trustworthy, or free of AI. It means only that a signed record exists and can be read.
Does my phone add Content Credentials?
Some phones do. Google's Pixel 10 attaches Content Credentials to photos taken with its camera app, and Google said in May 2026 that video support would follow on Pixel 8, 9, and 10. Many other phones and camera apps do not yet sign captures by default. The adoption status page tracks which devices ship signing.
Checking an image
How do I check whether an image has Content Credentials?
Upload the original file to Content Credentials Verify at contentcredentials.org/verify or to Adobe's Inspect tool. You can also run the open-source c2patool on the command line. These tools show whether a manifest is present, whether its signature validates, who signed it, and what edits it records. See verification tools, compared.
How can I tell if an image is AI-generated?
Check for Content Credentials, check for a SynthID watermark using Google's Gemini app, read the IPTC digital source type metadata, run a reverse image search, and then look for visual tells and use detectors as supporting evidence. A positive provenance or watermark result is strong evidence. A negative result from any single check is weak, and no single method is conclusive. See how to tell if an image is AI-generated.
Can AI detectors be trusted?
Not on their own. Detectors that score well on academic benchmarks often do much worse on real images circulating online. In the Deepfake-Eval-2024 benchmark, open-source image detectors lost about 45% of their AUC compared with earlier benchmarks. Treat a detector score as one weak-to-moderate signal, and never as the sole basis for a public claim. See AI image detection: methods and accuracy.
Trust and limits
Can Content Credentials be faked?
The signature cannot practically be forged without the signer's private key, and changing a signed file makes validation fail. However, anyone can create a valid manifest with their own certificate, so the question is who signed it, which is why validators check the C2PA trust list. A trusted signer can also make a false claim. See the C2PA trust list and threat models.
Do Content Credentials prove an image is real?
No. A valid credential proves that a specific signer made specific claims about the file and that the file has not changed since. It does not prove that the scene was not staged, that the caption is true, or that the camera was not pointed at a screen. See what provenance cannot prove.
Metadata and platforms
Does removing metadata remove Content Credentials?
Usually, yes. The manifest is stored in the file's metadata area, so tools that strip all metadata, such as ExifTool with -all=, delete it too. Editing any other metadata while leaving the manifest in place makes it fail validation. However, an invisible watermark or durable credential may still let the provenance be recovered. See removing image metadata.
Does a screenshot remove Content Credentials?
Yes. A screenshot is a new file made from screen pixels, so no embedded manifest carries over. Provenance can survive only through the pixels: an invisible watermark such as SynthID, or a durable Content Credential whose watermark points to a stored copy of the manifest. See durable Content Credentials.
Does Instagram keep C2PA?
Meta reads C2PA and IPTC indicators when an image is uploaded to Instagram, Facebook, or Threads and uses them to apply its "AI info" labels. Independent testing reported in 2026 found that the copies Instagram serves back generally no longer carry the manifest, so a downloaded image cannot be verified. Behaviour changes, so test with your own upload. See platform policies and metadata handling.
Which platforms preserve Content Credentials?
LinkedIn is the most consistent: it keeps C2PA data on supported images and shows a "cr" pin in the feed. Most other large social platforms re-encode uploads and remove the manifest from served copies. Email, cloud-storage links, and "send as file" options in messaging apps usually deliver the original bytes unchanged. See platform policies.
Do my photos reveal my location?
They can. If location services were on for the camera, the file's EXIF data usually includes GPS coordinates, and anyone who receives the original file can read them. iPhone lets you turn off Location in the share sheet's Options, and most camera apps can stop recording location entirely. See removing image metadata and provenance and privacy.
Watermarks
What is the difference between C2PA and a watermark?
A C2PA manifest is signed data stored alongside the pixels in the file's metadata area. It can hold a detailed, verifiable history, but it is lost when the metadata is stripped. An invisible watermark is hidden in the pixels themselves. It carries far less information but survives stripping and screenshots. The two are designed to work together. See hard bindings vs. soft bindings.
What is SynthID?
SynthID is Google DeepMind's invisible watermarking technology. It embeds a signal in the pixels of images, video, and audio produced by Google's AI tools. In May 2026 Google announced that OpenAI, Kakao, and ElevenLabs would also adopt it. Checks are available in the Gemini app, and Google has been rolling them out to Search and Chrome. See SynthID and AI-generator watermarks.
Is SynthID visible?
No. SynthID changes pixel values in ways designed to be imperceptible, so a watermarked image looks the same as an unmarked one. It can be read only with Google's detection tools, not by eye or with a metadata viewer. See invisible watermarking explained.
Can AI watermarks be removed?
Invisible watermarks survive ordinary compression, resizing, and screenshots far better than metadata, but they are not permanent. Heavy cropping, aggressive editing, regeneration through another model, and targeted attacks can weaken or remove them, and Google describes SynthID as not foolproof against extreme manipulation. See attacks on watermarks.
Law
Is labelling AI-generated images legally required?
In the EU, Article 50 of the AI Act has applied since 2 August 2026. It requires providers of generative AI systems to mark output in a machine-readable, detectable way, with a grace period to 2 December 2026 for systems already on the market, and deployers must disclose deepfakes. In the US, requirements come from state laws such as California's AI Transparency Act. See the EU AI Act and provenance and US laws affecting image provenance.