AI generated images now look real enough to fool most people at a glance. The quick answer is this: no single clue proves an image is fake, but a short checklist of visual signs, file checks and reverse searches will catch most of them. Use several checks together, and never trust one sign on its own.
Social feeds, news stories and even product listings now carry images that no camera ever took. Some are harmless fun. Others are scams, fake endorsements or false evidence. Learning how to spot them is now a basic online skill, much like spotting a phishing email.
Why the old checklist no longer works
Old advice told you to count fingers and look for melted hands. That advice is out of date. Modern image models handle hands, faces and text far better than the models of 2022 and 2023 did. A clean pair of hands no longer proves a photo is real.
The same applies to skin. Early AI portraits had plastic skin and glassy eyes. Current models add pores, stray hairs and natural blemishes. If you rely on one old clue, you will miss most new fakes and wrongly doubt real photos too.
The smarter approach is to look for a pattern of small errors, then verify with tools. One odd detail means little. Three odd details in one image mean a lot. This is the same caution you need with AI text, because AI tools can invent convincing details in words as well as in pictures.
The 2026 visual checklist
Work through these signs slowly. Zoom in to 100 percent or more before you judge anything, because small errors hide at feed size.
1. Hands, ears and jewellery
Hands have improved, but they still slip. Look for extra fingers, fingers that merge, rings that melt into skin, and earrings that do not match each other. Ears are a quiet giveaway: one ear may sit higher than the other, or an earring may appear on only one side.
2. Text in the image
Read every word in the picture, including signs, labels, book spines and shirt logos. AI models still garble small text. Look for misspelled words, letters that repeat, and text that dissolves into shapes at the edges. A shop sign that reads almost correctly is a strong warning sign.
3. Skin, hair and teeth
Real skin has uneven texture. AI skin can look too smooth in patches, or oddly detailed in one area and blurry in another. Hair may blend into the background, or single strands may end in mid-air. Teeth are another classic trap: too many teeth, teeth that merge, or a smile that looks pasted on.
4. The background
AI models focus on the main subject and guess the rest. Check the background carefully. Look for warped railings, staircases that lead nowhere, tiles that change pattern, and crowds where faces blur into each other. Reflections are revealing too: mirrors, windows and puddles often show a different scene from the one in front of them.
5. Light and shadows
Light should behave the same way across the whole image. Check that shadows fall in one direction and match the light source. Watch for faces lit from a different angle than the background, glowing edges around a person, and shadows that are missing entirely. Inconsistent light is one of the most reliable signs in 2026.
6. Too perfect, too clean
Real photos have clutter, dust, odd angles and imperfect timing. AI images often look staged in a way real life rarely is. If every element sits exactly where it should, the lighting is flawless and nothing is out of place, treat the image with healthy suspicion. Perfection is not proof, but it is a reason to check further.
Check the file, not just the picture
Visual checks alone are not enough. File checks catch fakes that look perfect.
Reverse image search
Run the image through Google Lens or a reverse image search tool. If the same picture appears elsewhere with a different story, or only on low quality sites, that tells you something. If no match exists at all for a supposedly famous event, that is also worth noting. A real news photo usually has a trail.
Content Credentials and watermarks
Some AI tools now attach Content Credentials, a tamper evident record that says how an image was made. Google also uses SynthID, an invisible watermark embedded in images from its tools. You can check for Content Credentials with the free Verify tool from the Content Authenticity Initiative. The catch is that screenshots and re-uploads often strip this data, so a missing record proves nothing on its own.
Metadata
Real camera photos often carry EXIF data such as camera model, lens and date. AI images usually carry none, or carry generator metadata instead. Missing metadata is weak evidence by itself, because social platforms strip it from uploads, but it adds to the picture when combined with other signs.
Use AI detectors wisely
AI image detectors analyse patterns that humans cannot see, such as noise and compression artefacts. They can help, but they are not reliable enough to trust alone. Studies and real world tests keep showing false positives on real photos and false negatives on new AI models. Use a detector as one vote among several, never as the final verdict. Our guide to AI content detectors explains which tools do what and where their limits lie.
The same rule applies to every AI safety question. Detectors support your judgement, they do not replace it. If you want the wider safety picture, read our guide on whether ChatGPT is safe to use, which covers the same habit of checking before you trust.
What still fools people in 2026
Three kinds of image fool even careful viewers. The first is the real photo with an AI edit: a genuine picture where one element has been changed, such as a face swapped or a sign altered. The rest of the image checks out, so the fake hides inside a real photo.
The second is the AI video still. A single frame from an AI video can look like a photograph, and motion hides errors that a still image would reveal. If a clip looks too smooth or the subject moves oddly, treat its stills with the same doubt.
The third is the plausible fake. A picture of an ordinary scene, with no famous person and no dramatic claim, draws no suspicion at all. These quiet fakes spread furthest because nobody thinks to check them. The habit that protects you is checking images that support a claim you are about to act on, not just images that look dramatic.
A quick routine you can use every time
Before you share or act on a surprising image, run this routine. It takes under two minutes.
- Zoom in and check hands, text, background and shadows.
- Ask whether the image is too perfect or too clean to be real.
- Run a reverse image search and check for an original source.
- Check for Content Credentials or a visible AI label.
- Run a detector if you still doubt it, and treat its verdict as one vote only.
- If doubt remains, do not share it. Sharing spreads the fake even when you add a warning.
This routine will not catch every fake. No routine can, because generators improve every month. But it will catch most of them, and it builds a habit that protects you across every platform.
Frequently asked questions
Can AI generated images be detected reliably?
No method is fully reliable on its own. Visual checks, file checks, reverse search and detectors each catch some fakes and miss others. Combining several methods gives the best result, and some new models will still pass every check.
Does a watermark prove an image is AI generated?
A visible AI label or a SynthID watermark is strong evidence, but its absence proves nothing. Watermarks can be cropped out, stripped by screenshots, or lost when an image is re-uploaded. Always check other signs too.
What is SynthID?
SynthID is Google’s invisible watermark for AI generated content. It embeds a signal in images, audio, text and video from Google tools. Specialised detectors can read it, but ordinary viewers cannot see it, and it may not survive heavy editing.
Can I trust reverse image search?
Reverse image search is one of the most useful checks, because it finds the original source of a picture. It works best for images that have circulated before. A brand new AI image may have no matches, so a clean result is not a pass.
What should I do if I already shared a fake image?
Delete the post if you can, then post a short correction in the same place. Say plainly that the image was AI generated and that you checked it after sharing. A quick honest correction does far more good than quietly deleting it.
Conclusion
Spotting AI generated images in 2026 comes down to one habit: slow down and check. Zoom in on the details, verify the source, and use detectors as a helper rather than a judge. Generators will keep improving, so the checklist will keep changing, but the habit of verifying before you share will stay useful no matter how good the fakes get.
