Find the highest-quality version
Use the original file when possible. Avoid judging from thumbnails, screenshots, cropped previews or messaging-app downloads.
Checklist
A practical checklist for spotting AI-generated images.
Use this checklist when an image looks suspicious, when a detector score is unclear, or when you need a repeatable review process before publishing, buying, citing or reporting.
First pass
Start with the file itself.
The file you review matters. A detector is more useful on an original upload than on a compressed screenshot copied from a social feed.
Note where you found it
Save the page URL, account name, post date, caption and any claim attached to the image.
Check file type and size
Very small files, repeated recompression or unusual dimensions can make detector results less reliable.
Visual inspection
Zoom in before you trust the scene.
AI images often look convincing at a glance and weaker under close inspection.
Hands, teeth and eyes
Look for extra fingers, fused teeth, mismatched pupils, asymmetrical earrings, strange eyeglasses or skin texture that changes suddenly.
Text, logos and labels
Zoom into signs, receipts, packaging, shirt text and interface labels. AI-generated text often looks plausible but unreadable.
Light and shadows
Check whether shadows agree with light sources, whether reflections match the scene, and whether objects cast plausible shadows.
Object boundaries
Watch for melted edges, impossible object intersections, hair blending into backgrounds or jewelry that changes shape.
Repeated patterns
Background crowds, leaves, bricks, fabric, shelves and food textures can reveal copied or statistically repeated details.
Physical plausibility
Ask whether the scene could exist as shown: camera angle, gravity, reflections, anatomy, perspective and object scale should agree.
Evidence check
Verify the image outside the file.
The strongest evidence often comes from context, not pixels.
Reverse image search
Look for earlier copies, stock listings, generator galleries, news usage or pages that contradict the current claim.
Review metadata
Check EXIF or content credentials if available. Treat missing metadata as inconclusive.
Run detector tools
Use two or more AI image detectors. Record tool names, dates and scores if the decision matters.
Write a decision note
Summarize what you found: original source, detector results, visual clues, unresolved doubts and final confidence.
Decision language
Use careful wording.
When you are not certain, avoid overclaiming. Better language protects you and the people affected by the image.
| Weak wording | Better wording | Why |
|---|---|---|
| This is fake. | This image shows several AI-generation signals. | It names evidence without claiming certainty. |
| The detector proved it. | The detector returned a high AI-likelihood score. | Detector scores are not proof. |
| It is real because metadata exists. | The metadata supports a camera-origin claim, but needs context. | Metadata can be missing, stripped or edited. |
| It passed one detector. | One detector did not flag it; other checks are still needed. | False negatives happen. |