How to Tell If an Image Is AI Generated Before You Share It

How to Check Whether a Photo Is AI Generated

Check provenance first, trace the source second, and look for visual tells last. A missing watermark or credential does not mean a photo is real, and detector scores are probabilities. The order below helps you spot AI generated images before you share or believe them.

Start with the file.

1. Look for Content Credentials with the Content Credentials Verify site, or open the photo in Google Photos and check "How this was made."
2. Run a watermark check. Upload the file to the Gemini app, or use the SynthID Detector at synthid.com.
3. If OpenAI may have made the image, use the provenance checker that OpenAI describes in its content provenance guide.
4. Run a reverse image search to trace the source of the image.
5. Only then compare visual tells, such as hands, text, skin, shadows and reflections.

For fakes of real people, see our guide to how to spot a deepfake. For clips, see the red flags of an AI-made video.

Content Credentials and Watermark Checks

C2PA stands for the Coalition for Content Provenance and Authenticity. It publishes an open technical standard for recording where content came from and how it was edited. C2PA describes Content Credentials as working "like a nutrition label for digital content."

Content Credentials are cryptographically signed and tamper-evident, according to the Content Authenticity Initiative. Inspection tools such as the Verify website display them when you upload a file.

Google says Content Credentials do not show whether a photo is real or fake. They are only available for files from compatible devices and apps, not for every photo. Google says the Pixel 10 lineup is "the first to have Content Credentials built in across every photo created by Pixel Camera."

In Google Photos on Android or iOS, open the photo, tap More, then About, and scroll to "How this was made." Up to four categories can appear, including "Media captured with a camera," "Edited with AI tools," "May have been edited by AI tools" and "Edited with non-AI tools." Files without C2PA support may show an "AI info" section instead. Content Credentials are not shown on Google Photos Web.

Google added image verification to the Gemini app on 20 November 2025. You upload an image and ask a question such as "Is this AI-generated?" Gemini then checks for a SynthID watermark, which Google says has been applied to over 20 billion pieces of AI content.

On a computer, go to gemini.google.com, sign in, and click Add files > Upload Files. Upload one image, video or audio file at a time, and keep it at 100 MB or less. Videos must run under 90 seconds, and audio under 1 hour.

If SynthID is detected, Gemini says all or part of the image "was created or edited by Google's AI models." If it is not detected, the content "wasn't created or edited by Google AI, but it could have been created by other AI systems." Unclear results include "Not enough details to watermark" and "Likely too small an edit."

Gemini can currently only recognize content created by Google AI tools, even though other companies have started adopting SynthID. Google says that on 7 October 2026 it opened the SynthID Detector to everyone, globally, in English. It checks whether a file was made with AI from Google or its partners, including OpenAI, NVIDIA and Kakao.

At Google I/O on 20 May 2026, Google said people can ask "Is this AI generated?" in Lens, AI Mode, Circle to Search or Gemini in Chrome. The same announcement said C2PA verification was rolling out in the Gemini app, with Search and Chrome to follow "in the coming months." Google also said OpenAI, Kakao and ElevenLabs are bringing SynthID to their AI-generated content.

OpenAI's provenance check looks for C2PA and SynthID signals, through a web tool at openai.com/verify and through an API. OpenAI says the check "isn't a general-purpose AI detector and doesn't identify content generated by every AI system." A "not_detected" result "is not proof that the content is human-created." Not all OpenAI-generated content contains a supported signal.

Meta changed its "Made with AI" label to "AI info" across its apps on 1 July 2024. Meta says those labels rely on industry-shared signals of AI images or on people self-disclosing. On 12 September 2024, Meta moved the label for content that was only modified with AI into the post's menu. The label stays visible on content detected as AI generated.

Metadata can be stripped, and a watermark can degrade. OpenAI says content may still be its own in those cases. Treat each result as one signal among several.

Reverse Image Search

A reverse image search helps you trace a picture to where it appears online. On google.com, click "Search by image," then "Upload a file," as Google's help page describes.

In Chrome, right-click an image and choose "Search with Google Lens" to see results in a sidebar. TinEye also exists as a reverse image search engine.

Check where the picture appears online before you rely on any detector score.

Visual Tells and Their Limits

A June 2024 arXiv paper by Northwestern University researchers sorts clues for telling AI images from photographs into five categories. The categories are anatomical, stylistic, functional, violations of physics, and sociocultural.

The paper lists hand artifacts such as missing fingers, extra fingers, merged fingers, nonexistent fingernails, and unlikely hand proportions. It describes skin that looks "waxy, shiny, cartoonish, or glossy." It also describes text that looks "glyph-like" or is written in a "nonexistent language."

Shadows can be "cast in inconsistent directions," and reflections may not match the scene. The authors say low resolution gives you less information to check. Full bodies and busy backgrounds leave more room for errors than simple portraits do.

The authors also stress that "Human-perceptible artifacts are not always present in AI-generated images." Use these tells after the provenance and source checks.

Do AI Image Detectors Work?

The only independent test in our sources is NewsGuard's May 2026 report. NewsGuard ran 15 authentic U.S.-Iran-war photos through five AI image detectors.

NewsGuard reports that the five tools together called real images AI-generated in 10 of 75 verdicts, or 13.33%. ScamAI flagged 6 of the 15 real photos (40%), ZeroGPT flagged 3 (20%), and AI or Not flagged 1 (6.67%). Hive and Sightengine flagged none.

NewsGuard also tested lightly AI-edited versions, with the lighting enhanced and the background blurred. Flag rates on those versions were 93% for ScamAI, 87% for AI or Not, 80% for ZeroGPT, and 27% for Hive and Sightengine. At least one tool disagreed with the rest on 35 of the 45 images.

NewsGuard did not test fully AI-generated images. Its results cover false positives on real and lightly edited photos.

Detector vendors publish their own accuracy claims, and those were not independently verified. For the tools themselves, see our AI art detector guide. Detector scores are probabilities, and they can fail on real photos too.

How Do I Know If a Picture Is Made by ChatGPT?

Run the picture through OpenAI's provenance check at openai.com/verify first. OpenAI says the check looks for C2PA and SynthID signals, and that it is not a general-purpose AI detector.

A "not_detected" result is not proof that the picture is human-created. OpenAI says not all of its generated content carries a supported signal, and metadata can be stripped. A negative result therefore leaves the question open.

Google said in May 2026 that OpenAI is bringing SynthID to its AI-generated content.

Run the OpenAI check, the Content Credentials check and a reverse image search before you share the picture.

Sources

• C2PA homepage
• Content Authenticity Initiative, "How it works"
• Content Credentials Verify site
• Gemini Apps Help, "Verify AI-generated images, videos, and audio"
• Google blog, image verification in the Gemini app (20 Nov 2025)
• Google blog, "100 things we announced at Google I/O 2026" (20 May 2026)
• Google blog, SynthID Detector expansion (7 Oct 2026)
• SynthID Detector
• OpenAI API docs, "Content provenance"
• Google Photos Help, "How Google Photos helps you identify how photos & videos were made"
• Google blog, Pixel and Android Content Credentials (10 Sep 2025)
• Meta Newsroom, "Our Approach to Labeling AI-Generated Content and Manipulated Media"
• Google Search Help, "Search with an image on Google"
• Google Lens
• Kamali, Nakamura, Chatzimparmpas, Hullman and Groh, "How to Distinguish AI-Generated Images from Authentic Photographs" (arXiv, June 2024)
• NewsGuard, "Leading AI Image Detection Tools Mislead Online Users" (8 May 2026)