AI Image Verification Checklist: How to Check If a Photo Was Made or Edited by AI

How AI Image Verification Works—and What It Cannot Prove

An image can mislead you in several different ways. It may be entirely AI-generated, or it may begin as a real photograph and later be expanded, retouched, combined with generated elements, or attached to an unrelated story.

This is why a useful image check asks two separate questions:

A genuine photograph can still support a false story. An AI-generated illustration can also be used honestly when it is clearly labeled. The main risk is not simply that AI was involved—it is that the image is being used to deceive, impersonate, pressure, or mislead you.

Why Visual Guessing Is Not Enough

Older AI images often contained obvious mistakes such as extra fingers, broken text, mismatched eyes, or melted objects. Those clues can still appear, but modern image-generation and editing systems can produce far more consistent results.

Visual details remain useful, but they should be treated as supporting evidence rather than proof.

The Seven-Step AI Image Verification Checklist

Use the following process before trusting a suspicious image, forwarding it to someone else, publishing it, sending money, or providing personal information.

1. Check the Claim and the Account

   Start with the message around the image rather than zooming immediately into the pixels. Ask who posted it, what they want you to believe, and what they want you to do next.

   • Is the original account identified?
   • Does the account have a credible history?
   • Is the post creating urgency, anger, fear, excitement, or sympathy?
   • Is money, identity, safety, reputation, politics, or breaking news involved?
   • Is the sender asking you to click, pay, donate, reply, or share quickly?
   • Does the post identify a date, location, photographer, company, or organization?
   Ask yourself

   Would I still trust this claim if the image were removed from the post?

2. Find the Original or Earliest Credible Source

   Reposts often remove captions, dates, metadata, context, and attribution. Try to locate the earliest credible version rather than accepting the copy that reached you.

   • Search for the photographer, organization, seller, business, or publication.
   • Compare publication dates.
   • Look for a larger, uncropped, or higher-quality version.
   • Check whether the original page provides a date, location, caption, or explanation.
   • Be cautious when every version leads to another anonymous repost.

   Finding an older version may reveal that a supposed current event happened years earlier, in another country, or under completely different circumstances.

3. Check Available AI Provenance Signals

   Provenance tools can help identify whether compatible systems recorded information about how an image was created or edited.

   Google’s Gemini verification feature currently checks two main technologies:

   • SynthID: an invisible watermark used to identify content created or edited by supported Google AI models.
   • Content Credentials: signed provenance information that may describe a file’s origin, composition, editing history, and whether compatible AI tools were involved.

   A detected signal can provide meaningful evidence. A missing signal cannot certify that the image is genuine.

4. Reverse-Search the Image

   Reverse-image search looks for identical or visually similar copies online. It can reveal information that an AI detector cannot.

   A reverse search may uncover:

   • An older version with a different caption.
   • A cropped original.
   • A product image copied from another seller.
   • A real photograph reused in an unrelated scam.
   • A stock photo presented as a personal image.
   • Several accounts using the same supposedly unique photo.

   Search more than once when possible. Try the full image, a tight crop around the main subject, and a crop containing a distinctive object or background.

5. Compare Reliable Independent Sources

   One post repeating another post is not independent confirmation. Look for sources that gathered their information separately.

   • Check the official website or verified account of the organization involved.
   • Look for reputable reporting that identifies the photographer or source.
   • Confirm dates, locations, names, weather, landmarks, uniforms, and event details.
   • Contact the business or organization using contact information you found independently.
   • Do not rely on phone numbers, email addresses, or links supplied only by the suspicious post.

   For a product or rental listing, compare the images with official manufacturer pages, map views, other marketplaces, and earlier listings.

6. Inspect Metadata and Visual Details

   If you have the original file rather than a screenshot or social-media download, inspect any available metadata. It may contain a camera model, date, editing application, dimensions, or other file information.

   Metadata has limits. It can be removed, altered, copied, or lost during platform processing. Missing metadata is common and does not prove AI generation.

   Then inspect the visible image for supporting clues:

   • Garbled or inconsistent text.
   • Reflections that do not match the scene.
   • Conflicting directions of light and shadow.
   • Objects blending into each other.
   • Repeated people, textures, windows, trees, or background objects.
   • Unnatural perspective or scale.
   • Jewelry, clothing, or accessories changing shape.
   • Edges that look unusually smooth, smeared, or sharply cut.
   • Different levels of detail between the subject and the background.

   One strange detail is not proof. Look for several inconsistencies that support what you learned from the source and provenance checks.

7. Pause Before You Share, Pay, or Act

   When important doubts remain, treat the image as unverified. You do not have to prove exactly how it was made before deciding not to trust it.

   • Do not repost it as confirmed.
   • Do not send money through gift cards, cryptocurrency, wire transfers, or unfamiliar payment apps.
   • Do not provide passwords, identity documents, banking details, or personal information.
   • Ask the sender for the original file and source.
   • Save screenshots and links if you need to report impersonation or fraud.
   • Use neutral language such as “unverified image” when discussing it.
   Safe decision rule

   If the image is essential to the claim but you cannot confirm its origin, do not use the image as proof.

How to Use Gemini for AI Image Verification

Signed-in Gemini users can upload an image and ask whether it was created or edited using AI. The tool checks for a supported SynthID watermark and, when available, compatible Content Credentials.

Step-by-step Gemini image check

1. Open Gemini and sign in

   Use the Gemini web app or a supported mobile app.

2. Upload one image

   Use the original file when possible. Avoid collages, heavily compressed copies, or screenshots containing several separate images.

3. Ask for a precise verification explanation

   Request both the detected signal and the limitation of the result.

4. Read the exact wording

   Do not reduce the answer to a simple “real” or “fake” label. Look for whether SynthID was detected, whether Content Credentials were available, and what the result does not establish.

Was this image generated or edited using AI?

Please:
1. Check for a SynthID watermark
2. Check for any available Content Credentials
3. Explain exactly what each detected signal confirms
4. Explain what the result does not prove
5. Do not call the image real simply because no watermark or credential was found

Before uploading a personal, customer, confidential, or business image to any online service, review what the service can access and how the file may be handled. The AI Agent Safety Checklist can help you review permissions and limit unnecessary data exposure.

How to Interpret Gemini’s Verification Result

How to Check Content Credentials Directly

Compatible files can be inspected through the official Content Credentials verification tool. Upload or drag the file into the verifier to view any available provenance information and see how the file may have changed over time.

When credentials are available, review:

• Who signed the credential.
• Which application or device supplied the information.
• Whether AI use is recorded.
• Whether the media is a composite.
• Which edits or transformations are listed.
• Whether the credential is valid or reports an error.

You can inspect compatible media using the official Content Credentials verifier.

Content Credentials are still being adopted. Many genuine photographs and many AI-generated images will not contain them.

How to Reverse-Search a Suspicious Image

Reverse-image search can be especially valuable when the main problem is not generation but reuse, impersonation, false context, or stolen product photography.

1. Search the full image

   Start with the largest and clearest copy you have.

2. Search a tight crop

   Crop around a face, product, landmark, building, vehicle, pet, sign, or other distinctive subject.

3. Compare dates and captions

   Look for the earliest indexed version and note whether different websites describe it differently.

4. Open the original pages

   Do not rely only on thumbnail results. Visit the source pages and check their publication date, author, caption, and context.

5. Repeat with another crop or service

   Different search tools may find different matches. One unsuccessful search does not prove that the image is new or authentic.

What Image Metadata Can and Cannot Tell You

Original image files may contain EXIF or other metadata showing information such as the camera model, creation date, dimensions, location settings, or editing application.

Metadata can also be altered or copied. A listed camera model does not prove that every visible part of the image came from that camera.

When to Be Extra Cautious

The need for verification increases when an image is being used to influence a financial, identity, safety, legal, or reputational decision.

How to Verify a Product-Listing Photo

A polished product photo does not prove that the seller owns the item, that the item matches the picture, or that the listing is legitimate.

• Reverse-search the main listing photo.
• Compare it with official manufacturer images.
• Check whether several unrelated sellers use the exact same photograph.
• Look for inconsistent logos, labels, dimensions, materials, or accessories.
• Ask for a new image from a specific angle.
• Request a photo that includes the current date or a unique handwritten note.
• Review seller age, feedback, return terms, and contact information.
• Use a payment method with appropriate buyer protection.

For a wider purchasing review, use the free Designs24hr Decision Helper to compare the evidence, risks, seller credibility, and safer alternatives. It is a thinking aid, not an image-authentication service.

What to Do When the Image Remains Uncertain

Not every investigation will produce a clear result. Uncertainty is itself useful information—especially when someone is asking you to act quickly.

• Do not describe the image as confirmed.
• Ask for the original file rather than another screenshot.
• Ask who created the image and where it was first published.
• Seek independent evidence for the underlying claim.
• Contact the relevant company or organization through its official website.
• Avoid sending money or personal information.
• Use “unverified” or “not independently confirmed” when discussing it.
• Report fraud, impersonation, or harmful deception to the relevant platform or authority.

You do not have to prove that an image is fake before deciding that the evidence is too weak to trust.

Official Sources and Verification Tools

AI-verification tools and platform support can change. Use current official documentation when a high-stakes decision depends on the result.

Frequently Asked Questions About AI Image Verification

Fact-checked against official Google, Google DeepMind, Content Credentials, and FTC information on July 13, 2026. Verification features, limits, and availability may change.