SXF / PUBLIC FIELD GUIDE / 2026
How to Tell If an Image Is AI-Generated
A photo looks real. The caption is convincing. But where did the image actually come from? Use these 7 free checks to distinguish genuine evidence from guesses—and learn when the only responsible answer is “not enough information.”
How can you tell if an image is AI-generated?
First, reverse-search the image with Google Lens. Then check the original file for Content Credentials (C2PA) and use the relevant official watermark tool—Gemini/SynthID for supported Google content or OpenAI Verify for supported OpenAI content. Finally, verify the publisher and caption. Missing metadata, clean-looking hands or a “human” detector score do not prove a photo is real. It may remain unverified. Use the free SXF decision guide.
Can you actually detect an AI-generated image in 2026?
Sometimes—but there is no universal, foolproof “AI image detector” for every image from every model.
Modern AI images can have correct hands, natural light and flawless text. Meanwhile, real photos can look uncanny after sharpening, low-light processing, filters or compression. Visual tells help you notice an inconsistency; they do not establish origin. Even a seemingly perfect image could be synthetic.
The strongest available clues concern provenance: a traceable publisher, a cryptographically signed creation record, or an officially detectable watermark. Each proves something different. A signed file history does not prove that an event happened, and an image with no detectable watermark could still be AI-made. Google and OpenAI describe exactly these limits in their own verification guidance.
Signed records and official watermarks may identify tools or creation steps for supported images.
Original reporting, reliable eyewitness evidence and context establish whether the caption matches reality.
A PRACTICAL 2–5 MINUTE WORKFLOW
7 free ways to check if a photo was made with AI
Do the first three checks before interpreting visual details. For the most reliable answer, obtain the original file where possible—not just a cropped screenshot sent in a chat.
Reverse-search the image
Use Google Lens to find earlier copies, the earliest known publisher and whether the caption changed. Open matching pages rather than trusting thumbnails.
Best for: viral posts, old photos recycled as breaking news.Open Google Lens ↗Use Google’s About this image
Look for first-indexing context, other sites using the image, available metadata and supported provenance signals. Access varies by platform and region.
Best for: context, not a universal AI yes/no verdict.Google’s official walkthrough ↗Inspect signed Content Credentials
If you have the original file, upload it to the official Content Credentials verifier. Review the signed issuer and edit history, not just whether any metadata exists.
Best for: files with C2PA provenance; missing credentials prove nothing.Open Content Credentials Verify ↗Check for an OpenAI provenance signal
For images suspected to come from ChatGPT or OpenAI tools, use OpenAI Verify. It checks supported signals such as C2PA and SynthID, not every AI generator.
Best for: identifying supported OpenAI-origin signals.Open OpenAI Verify ↗Check Google AI watermark signals
In Gemini, upload a supported image and ask whether it was generated or edited by Google AI. In October 2026, Google also expanded the SynthID Detector globally in English. Product coverage is not universal.
Best for: supported Google AI and partner watermark workflows; check tool-specific limitations.Read Gemini’s verification instructions ↗Inspect the original file and visual details
A camera model or creation date in EXIF can offer clues, but ordinary metadata can be altered or stripped. Look at reflections, repeated textures, tiny text and anatomy—then corroborate elsewhere.
Best for: finding questions to investigate, not issuing a verdict.Why file signals can disappear ↗Verify the event, not just the pixels
If the image claims a real disaster, celebrity endorsement or breaking event, check whether a credible first-hand witness, local agency, newsroom or independent reporting supports it. If someone asks for money, stop until you verify independently.
Best for: high-stakes claims, scams and misleading captions.How image context helps ↗SXF ORIGINAL INTERACTIVE FIELD TOOL
The SXF Evidence Ladder: what should you check next?
Unlike a detector that guesses a percentage from pixels, this tool tells you what your existing evidence can and cannot support. Choose your situation; the recommendation updates instantly. The choices stay in your browser—you do not upload a photo to SXF.
Build your verification plan
We do not inspect your image, read your files or assign a probability. This is a guidance engine based on the reported evidence.
Start with the original file and source history. No evidence has been assessed yet.
- Try a reverse-image search.
- Look for signed credentials or a supported watermark.
Guidance only. This tool does not identify whether any particular image is genuine.
How to read the results of an AI image checker
| What you found | What it can support | What it cannot prove |
|---|---|---|
| Verified AI-origin C2PA credential | A signed issuer recorded AI generation or editing for this asset | Who shared it, whether the caption is honest, or whether every pixel is synthetic |
| Official supported AI watermark detected | A provenance signal associated with the detected provider/tool | The real-world truth of the depicted scene |
| Signed camera/creation credentials | A verifiable capture or editing record where supported | That the photographed scene, screen or event is genuine |
| Same picture published years earlier | The current caption/date/location may be misleading | That the pixels were made by AI |
| “No watermark detected” | That the tool found no supported signal in this file | That the image is human-made or authentic |
| Generic AI detector says “92% AI” | A classifier prediction worth checking | Mathematical certainty or definitive proof |
Look closely at the wording: “not detected” means unknown, not “verified real.” Screenshots, exports, compression and edits can remove metadata or weaken watermarks. OpenAI specifically warns against treating a negative provenance check as proof of human authorship.
Three everyday examples: how the checks change your decision
These are illustrative walkthroughs, not measured detector test results. Each shows how to reason from evidence, rather than claim a tool returned a result we never tested.
You have only a screenshot. A reverse-image search finds the same scene published years before the alleged event. That is meaningful evidence of a misleading caption or date—even if the photograph itself is real. Ask an authoritative local source whether the event occurred before sharing or donating.
No extra fingers, no warped glasses and a detector says “probably human.” None of those establishes identity. If the profile asks for money or personal documents, verify the person independently and do not use the image as sole identity proof.
A verified C2PA record explicitly identifies an AI-generating tool or an official watermark check detects a supported AI signal. You now have positive origin evidence for that specific file. You still need context to know who posted it, why, and whether a related factual claim is true.
Which free AI image checker should you use?
Choose a tool that answers the question you actually have. Product availability and usage limits can change.
| Tool | Best for | Key limitation |
|---|---|---|
| Google Lens | Earlier copies and image context | New/unindexed images can have no matches |
| Content Credentials Verify | Signed C2PA origin/edit records | Many images have none; metadata can be stripped |
| OpenAI Verify | Supported OpenAI image provenance | Cannot rule out other providers or degraded/older images |
| Gemini image verification | Supported Google SynthID and available Content Credentials | Account, platform, format and quota constraints; not all generators |
| Google SynthID Detector | Supported watermarked media from participating tools | No signal does not mean real; scope varies by product |
Privacy tip: If a photo contains private faces, IDs, medical information or children, review each external service’s privacy policy before uploading it. The SXF decision tool above never receives the image.
Five mistakes that lead people to falsely label a photo
- “The hands look normal, so it must be real.” Modern generators often render hands well; cameras also create distortions.
- “The detector says 99%, so the case is closed.” A classifier estimate is not a validated provenance record.
- “I cannot find it on Google, so it is AI.” New, private or unindexed photos have no web footprint.
- “There is no metadata, therefore a human took it.” Social apps and screenshots routinely remove it.
- “A signed camera photo proves the claim happened.” A real camera can photograph a staged scene, altered print or generated image on a screen.
FAQs: Is this image AI-generated?
How can I tell if a photo is AI-generated?
Reverse-search it, trace the source, check signed provenance where possible and use relevant official watermark tools. No single absent clue proves authenticity.
Is there a free AI image detector online?
Yes, free or free-access verification routes include Content Credentials Verify, Google’s supported verification features and OpenAI Verify. Some require an account or limit usage. These services check specific signals—not all possible generators.
Can Google Lens detect AI photos?
Lens primarily searches for image matches. Google’s About this image may surface context, metadata and supported SynthID signals; absence of such data is inconclusive.
Does missing C2PA metadata mean an image is authentic?
No. Credentials may be stripped by screenshots, recompression, format conversions or social sharing.
How can I check an image made by ChatGPT?
Try OpenAI Verify with the original image. It looks for supported OpenAI provenance signals, but not-detected does not prove human origin.
Are AI image detector percentages accurate?
They are classifier outputs whose reliability depends on data, model and transformations. Treat an isolated percentage as a prompt to check other evidence, not a verdict.
Can I check an image from WhatsApp or Instagram?
You can reverse-search it and verify the claimed event. If it is a compressed forwarded screenshot, original credentials may be missing; request the original file before making a definitive provenance claim.
EVIDENCE & EDITORIAL TRANSPARENCY
Official sources and how SXF built this guide
This guide synthesizes primary-provider documentation and compares their stated limits. The SXF Evidence Ladder is an original rule-based editorial decision aid, not a trained image classifier, uploaded-file scanner or independent accuracy benchmark. We have not claimed that we ran sensitivity or specificity tests of third-party detectors. Features and coverage reflect checked documentation as of October 26, 2026.
- Google — Verify AI-generated images, video and audio (Gemini help): coverage, limitations and usage.
- Google — SynthID Detector global expansion, October 7, 2026.
- Google — How About this image adds image context and provenance.
- OpenAI — Provenance signals, limitations, watermark and C2PA.
- OpenAI — Official image provenance checker.
- Content Credentials — Official C2PA verification interface.
- Google Search Central — Creating helpful and reliable content.
Editorial scope: We reviewed the 2026 competing explainer landscape, including public pieces by Malwarebytes and Geek College, then structured this guide around a distinct reader task: interpreting evidence without overclaiming certainty. No provider has paid to be ranked here; tools are ordered by verification job, not by commercial preference.
RELATED SXF COMPARISON
Best AI Image Generators in 2026
Compare GPT Image, Midjourney, Nano Banana, FLUX and Ideogram by text rendering, product images, editing, API pricing and commercial license, with a reproducible five-task test plan.
Open the AI image generator comparison ↗