The fastest way to find out whether an image was made or edited by AI is to check two things attached to the file: its Content Credentials and its SynthID watermark. As of 2026 you can do this in a few seconds, from your phone or a free web tool, and a growing share of AI images now carry these markers by default.
There is a catch, and it matters more than the how-to. A clean result, where no AI marker turns up, does not mean an image is real. It usually means the tool that made it left no marker, or the marker was stripped along the way. This guide covers how to run the check on the main platforms, and what the answer can and cannot tell you.
What you're actually checking
Two separate signals are in play, and the tools often report both.
The first is C2PA Content Credentials, a small block of signed metadata that travels with a file and records where it came from and what happened to it. A camera, an editing app, or an AI generator can write one. A verifier reads it back and confirms who signed it and whether the file has changed since.
The second is SynthID, an invisible watermark that Google embeds in the pixels of images its models generate. Because it is woven into the image itself rather than the metadata, it can survive some edits and screenshots that strip a credential. Through a 2026 partnership, OpenAI now applies SynthID to ChatGPT and DALL-E images too, so one detector covers both. We wrote about that shift and its limits separately.
Content Credentials tell you a provenance story. SynthID answers the narrower question of whether a supported AI model made the picture. A full check looks for both.
Is it from ChatGPT or Sora? Use OpenAI's Verify
To find out whether a picture came out of ChatGPT, Sora, or DALL-E, the most direct route is OpenAI's own Verify tool. Upload the image and it checks for both an OpenAI C2PA credential and a SynthID watermark, then tells you whether one of OpenAI's models produced it.
This is reliable for a yes. If the tool finds the credential or the watermark, the image really did come from an OpenAI model. A no is softer. A screenshot of a ChatGPT image, or a copy that has been re-encoded a few times, may have lost the credential even though the picture is synthetic.
On Google: ask "Is this made with AI?"
Google reads the same signals through its own products, which is convenient when the image is already on your screen.
In the Gemini app, upload the image and ask whether it was made or edited with AI. Gemini checks for SynthID and reports back. In Google Search, Lens, and Circle to Search on Android, you can point at an image and ask the same question, and Google surfaces whether it carries a SynthID signal or Content Credentials. Google has said this provenance check is expanding into Chrome, so before long you will be able to run it on an image anywhere on the web without leaving the page.
The same limitation applies here. Google can confirm that an image is AI when it finds a signal. It cannot clear an image when it finds nothing, because plenty of AI tools leave no SynthID at all.
On a Pixel or in Google Photos: the info panel
Some cameras now write Content Credentials at the moment of capture, and the newest phones do it automatically. A Google Pixel 10 signs every photo it takes, at the highest assurance level the C2PA program currently defines, and the Pixel 8, 9, and 10 add credentials to video.
To read them, open the photo in Google Photos and tap the information icon. If the file carries Content Credentials, a section appears showing the provenance details in plain language, including the device and whether the image has been edited since capture. A broken or missing signature on a photo that claims to be a straight Pixel capture is a sign that something changed after the shutter.
For any image: Adobe's verifier or Lumethic's inspector
For an image from any source, two free verifiers read Content Credentials without an account.
Adobe's Content Credentials Verify page at contentcredentials.org lets you upload a file and see a human-readable summary of its manifest, including the issuer and the edit history. Lumethic's own Content Credentials inspector does the same and shows exactly what a manifest asserts, which is often less than a green badge implies.
Either one turns the abstract idea of provenance into something you can read. If a credential is present and valid, you learn where the file says it came from. If it is absent, you learn only that.
What a "no credential" result does not mean
This is the part most guides skip. Finding no AI marker is not the same as confirming a real photograph, for two reasons.
First, coverage is partial. SynthID flags images from Google's models and now OpenAI's, but an image from Midjourney, Stable Diffusion, or other tools carries no SynthID at all. A "no watermark found" result says nothing about those. Content Credentials have the same limit in reverse, because a generator that chooses not to write one leaves nothing to read.
Second, the markers are fragile in ordinary use. A screenshot, a re-upload, or a platform that recompresses images on the way in can strip a Content Credential cleanly. The picture looks unchanged, but its provenance record is gone. So an image with no credential might be a fake that lost its watermark, or a real photograph that never had one, and the check cannot tell those apart.
That is the gap. Checking credentials is good at confirming that AI made something. It is poor at confirming that a camera did. For that harder question, which is the one a photographer or an editor actually needs answered, you need positive evidence of a real capture rather than the absence of an AI marker. That is what a RAW-to-image verification provides, and it is the problem Lumethic is built to solve.
Frequently Asked Questions
How do I check if an image was made by ChatGPT?
Upload it to OpenAI's Verify tool at openai.com/research/verify, which looks for OpenAI's C2PA credential and a SynthID watermark. If either is present, an OpenAI model made the image. If neither is found, the image may still be synthetic, because screenshots and re-encoding can remove the markers.
Can Google tell me if a photo is AI-generated?
Yes, when the photo carries a signal Google can read. In the Gemini app, or through Search, Lens, and Circle to Search, you can ask whether an image was made or edited with AI, and Google checks for SynthID and Content Credentials. It cannot confirm that an image is real, only that it did or did not find an AI marker.
How do I view Content Credentials on a Pixel photo?
Open the photo in Google Photos and tap the information icon. If the photo carries C2PA Content Credentials, which every Pixel 10 photo does by default, a Content Credentials section shows the device, the capture time, and any edits since.
What is the difference between C2PA and SynthID?
C2PA Content Credentials are signed metadata that record a file's origin and edit history and can be added by cameras, editors, or AI tools. SynthID is an invisible watermark woven into the pixels of images made by Google's and OpenAI's models. A thorough check looks for both.
If an image has no Content Credentials, is it real?
No. A missing credential only means none was found. Many AI tools add no marker, and ordinary actions like screenshots or re-uploads strip credentials from genuine and synthetic images alike. Proving an image is a real photograph takes positive evidence of a camera capture, such as a RAW-to-image forensic check, not the absence of an AI marker.
Checking Content Credentials is a five-second habit worth building, and it will catch a growing share of AI images cleanly. Just read the result for what it is. A found marker is proof of AI. A missing one is proof of nothing. When you need to show that an image is a genuine capture, Lumethic verifies photos against their RAW originals, with the first checks free and no account required.