Advertisement
Artificial Intelligence

How can you tell if an image is AI-generated? 5 checks to run

No tool says for certain whether an image comes from AI, but several signals can be checked: Content Credentials, the SynthID watermark, Google's image history and platform labels. Here are five checks to combine, from signals in the file to the image's context.

Updated 7 min read

Smartphone, robot and magnifying glass illustrating how to check where an image comes from
Table of contents
  1. Common symptoms and causes
  2. Before you start
  3. Step 1: Look for Content Credentials (C2PA)
  4. Step 2: Find the image with Google reverse image search
  5. Step 3: Open “About this image” in Google
  6. Step 4: Ask Gemini whether the image comes from Google’s AI
  7. Step 5: Check platform labels and trace the source
  8. If nothing works
  9. Frequently asked questions
    1. How do you check whether an image is AI-generated?
    2. Are AI image detectors reliable?
    3. Can Google tell whether an image was created by AI?

Updated on November 18, 2026

To tell if an image is AI-generated, start with a signal left inside the file: Content Credentials (C2PA) or Google’s invisible SynthID watermark. No method is 100% reliable, and a missing signal proves neither that an image is real nor that it is fake. Google’s tools give context, not a verdict.

Advertisement

Common symptoms and causes

  • A photorealistic image is going around with no mention of AI. Meta admits that not all AI-generated content can be identified and that invisible watermarks can be removed.
  • The checking tool finds nothing. Without a C2PA manifest or a SynthID watermark in the file, it cannot conclude anything.
  • The credit information looks doubtful. Google states that this metadata can be altered with image creation or editing tools.
  • The image was copied as a screenshot or reposted. Third-party sources say Content Credentials can disappear this way; the official documents we read do not confirm it.

Before you start

  • If you can, get the original file of the image, not a screenshot: C2PA tools read the file itself.
  • To query SynthID, you need the Gemini app or access to Google Lens. Google does not say in which countries these checks are offered.
  • Combine several checks. One tool alone does not settle the question.

Warning: uploading a photo to an online checker, to Google or to Gemini sends it to a third party. Do not upload a private or sensitive image. No step in this guide erases data on your device.

Good to know: judging by eye (hands, text, reflections) remains a weak clue. No official source we read gives reliable visual signs, and Google admits there is no silver bullet for establishing where content comes from. According to press reports, recent generators fix flaws such as odd hands: an image with no visible flaw proves nothing.

Step 1: Look for Content Credentials (C2PA)

Content Credentials, or C2PA, are an industry standard: signed metadata stored in the file that records how it was created and edited, with or without AI. According to Google, the Pixel 10 is the first smartphone to add them to photos from its native camera app. Images made with Nano Banana Pro (Gemini 3 Pro Image) in the Gemini app, Vertex AI or Google Ads carry them too.

The Content Authenticity Initiative’s official Inspect tool reads this metadata. It accepts AVIF, DNG, GIF, HEIC, HEIF, JPEG, JXL, PNG, SVG, TIFF and WebP files, as well as video, audio and PDF files.

Warning: the pages we read do not say whether Inspect analyzes the file on its servers or in your browser. Do not drop a private image into it.

  1. Open inspect.cr, the Content Authenticity Initiative’s Inspect tool.
  2. Click Select a file from your device and choose the file, or drag it onto the page.
  3. Read the left panel: the signer appears under “Recorded by”.
  4. In the right panel, read the “Content details” section: the app or device used, the actions taken (“Actions”) and the source elements (“Ingredients”).

For an image that is online, enter https://inspect.cr?source= followed by the public address of the image. The page must not require a login, and the site hosting it must allow this kind of access (CORS, a server security rule).

Advertisement

Expected result: if a generative AI tool created the image, Inspect shows an “AI-generated” note. The message “No Content Credential” means the file holds no manifest, which says nothing about where it came from. A signer marked “Unrecognized” is not on the C2PA trust list.

refresh-199720-1-1024x576 How can you tell if an image is AI-generated? 5 checks to run
No C2PA manifest in the file: nothing is proven, either way.

Good to know: according to the third-party C2PA Viewer site, a valid signature proves the file has not been modified, not that its content is true.

Reverse image search starts from a picture instead of words: it shows where the image appears and in what context. The steps below are for a computer. Our guide on using Google Lens and visual search covers the tool in detail.

Warning: this step sends the image to Google. Do not upload a private photo.

  1. On google.com, click the search-by-image icon.
  2. Click Upload a file.
  3. Choose the image, then click Open or Choose.

You can also drag the file into the search box. For an image that is online, right-click it, choose Copy Image Address, paste it into Paste image link, then click Search. In Chrome, right-clicking the image and choosing Search with Google Lens shows the results in a side panel.

Expected result: Google lists pages that use the image or close matches. This context helps, but it does not prove on its own that an image is real or AI-generated.

Advertisement

Step 3: Open “About this image” in Google

This feature gathers an image’s history on the web. According to Google’s help page, it is offered only in some regions, and the page does not say whether yours is one of them.

  1. Search for the image in the Google app or on google.com.
  2. Open the image in the image viewer.
  3. Select the three-dot icon ⋮.
  4. Choose About.
  5. Select “Learn more about an image”.

Expected result: the page shows the date Google first detected a close version (not necessarily the creation date), the other pages that use it and, if there are any, much older “Early uses”. The “How the image was made or edited” section may point to a camera or to AI creation. “Info from” says who confirmed it.

Good to know: Google reads C2PA metadata here when the file contains it, when the signer follows C2PA version 2.1 or later, and when its certificate comes from an authority on the C2PA trust list. Announced in September 2024, this reading covers Google Images, Lens and Circle to Search.

Step 4: Ask Gemini whether the image comes from Google’s AI

SynthID is Google’s digital watermark: imperceptible signals built into content generated by its AI. In May 2026, Google said it had watermarked more than 100 billion images and videos this way. It presents this check for images created or edited with its own AI, and does not say that Gemini detects images from other tools through this watermark.

Warning: the next step sends the image to Google. Do not upload a private photo.

  1. Open the Gemini app.
  2. Upload the image you want to check.
  3. Ask a question like “Is this AI-generated?” or “Was this created with Google AI?”, the examples Google gives.

Expected result: Gemini looks for the SynthID watermark and answers with context. Since May 19, 2026, the app also reads C2PA metadata: it says whether the photo is an unmodified camera original, or whether it was modified and with which tools.

From the same date, Google is extending SynthID checking to search through Lens, AI Mode and Circle to Search: ask “Is this made with AI?”. It also announced that the check would come to Chrome through Gemini in Chrome, without naming the countries concerned.

refresh-199720-2-1024x576 How can you tell if an image is AI-generated? 5 checks to run
The SynthID watermark is only found on images that carry it.

Good to know: a checker based on SynthID only finds images that carry this watermark. A negative answer therefore does not prove that a person took the photo.

Step 5: Check platform labels and trace the source

  1. On Facebook, Instagram or Threads, look for an AI label on the post.
  2. On YouTube, look in the video player, then in the expanded description.
  3. In “About this image”, open the “Early uses” to find pages where the image appeared earlier.
  4. Cross-check the image’s credit information against other sources.

In February 2024, Meta announced an “Imagined with AI” label for photorealistic images created with Meta AI, and detection of C2PA and IPTC signals to label images from other tools. The current wording may have changed. YouTube shows a label for altered or synthetic content and can add it itself, notably when the content carries C2PA metadata.

Expected result: you know where the image first appeared and whether a platform flagged it. A missing label proves nothing. To go further with the data inside a photo, read our guide on locating a place from a photo.

If nothing works

With no label, no Content Credentials, no watermark and a murky context, treat the image as unverified. Look for the same image from reliable news outlets or fact-checking services. Do not share it as fact.

Online detectors do not settle the question: no official source we read evaluates their reliability. Keep the original file and the address of the page where you found it, so you can pass them to a news outlet or a fact-checking service. If the image comes with an offer or a request for money, also read our overview of online scams in 2026.

Frequently asked questions

How do you check whether an image is AI-generated?

Look first for a signal in the file: Content Credentials with the Inspect tool, or the SynthID watermark by asking Gemini. Then add reverse image search and Google’s “About this image”, which show where the image appears and since when. None of these methods is 100% reliable, and a missing signal proves nothing.

Are AI image detectors reliable?

No official source we read evaluates their reliability, and no verified success rate is available. Treat their result as a clue, never as proof. A tool based on SynthID only recognizes images that carry this watermark.

Can Google tell whether an image was created by AI?

Yes, but only in some cases. Gemini recognizes the SynthID watermark on images created or edited by Google’s AI, and “About this image” shows AI creation when the file contains recognized C2PA metadata. For an image made by another tool without these signals, Google gives context, not a verdict.

Share this content:

Leave a Reply

Your email address will not be published. Required fields are marked *