Photos, videos and deepfakes
Most fake images are real photos in the wrong place or time. Learn reverse image search, what AI images give away, and why the source matters more than the pixels.
THE MOST COMMON FAKE
A real photo with a false caption
The most common visual misinformation isn't a clever fake. It's a real photo or video shared with the wrong place, date or story: an old flood photo posted as today's storm, a crowd from another country or another year. Fact-checkers call these miscaptioned or out-of-context media, sometimes "cheapfakes", because they cost nothing to make.
A picture of a flooded street, captioned "Our city right now, and the news is silent". The flood was real. It happened in another country, years earlier.
Check yourself
What is the most common kind of misleading image online?
- A photo built from scratch by an AI tool
- A real photo shared with a false place, date or story
- A photo edited pixel by pixel in professional software
- A picture from a film shown as if it were news
Show the answer
A real photo shared with a false place, date or story
Right. Relabelling a real photo is fast, cheap and hard to argue with, because the picture itself is genuine. Only the story around it is false.
Reverse image search in four steps
- Save or screenshot the image
For a video, pause on a clear, distinctive frame and screenshot that.
- Search with the picture, not with words
Use Google Lens, Bing Visual Search or TinEye. Upload the image or paste its link. On a phone, Google Lens is often built into the camera or photo app.
- Look for the oldest copy
Scroll the results, or on TinEye sort by oldest. If the picture was online before the event it claims to show, the caption is wrong.
- Read the original caption
The earliest copy, ideally from a news agency or photographer, usually tells you the real place, date and story.
Check yourself
Your aunt forwards a dramatic photo captioned "Flooding in the north TODAY". You suspect it's old. What's the best first step?
- Zoom in and look for editing marks around the water
- Check the photo's hidden date and camera details
- Run a reverse image search and look for earlier copies
- Ask in the group whether anyone else thinks it's fake
Show the answer
Run a reverse image search and look for earlier copies
Yes. If copies of the photo exist from before today, the caption is wrong, however real the flood looks. It takes about thirty seconds.
Clues inside the picture
- Weather and season: snow, bare trees or summer clothes that don't fit the claimed date.
- Signs and shop fronts: the language, the script, phone numbers with a foreign format.
- Cars and number plates: which side of the road they drive on, the plate style.
- Landmarks: a mountain, a tower or a coastline you can compare with maps or other photos.
- Metadata: most social apps strip the date and camera details, so don't count on them.
EDITED, NOT INVENTED
Small edits that change the story
Between "real" and "fake" there's a wide middle. A photo can be cropped to remove the context (the empty street next to the "huge crowd"). A video can be slowed down, which can make a speaker seem drunk or confused. Audio can be spliced so sentences are rearranged. And a screenshot of a news article or a message is one of the easiest things to edit.
A clip of a speaker slurring words goes viral. The original, at normal speed, shows her speaking clearly. Nothing was added; the clip was just played at 75% speed.
Check yourself
What's wrong with each image or clip?
- A 2015 earthquake photo shared as last night's quake
- A speech slowed down so the speaker sounds confused
- A football crowd from another country labelled as a local protest
- A screenshot of a news site with the headline retyped
- A photo cropped so the half-empty square looks packed
- A snowy street from January shared as "today" in July
Show the answer
Real, but wrong place or time: A 2015 earthquake photo shared as last night's quake, A football crowd from another country labelled as a local protest, A snowy street from January shared as "today" in July
Changed after it was taken: A speech slowed down so the speaker sounds confused, A screenshot of a news site with the headline retyped, A photo cropped so the half-empty square looks packed
MADE BY AI
What AI images give away, for now
AI-generated images sometimes show tell-tale flaws: garbled text on signs, odd hands or ears, mismatched earrings, impossible reflections or shadows, skin that is too smooth. These are worth a glance. But the tools improve quickly, and each new version fixes some of them, so a missing flaw doesn't prove a picture is real. The better test is provenance: where the image first appeared and who posted it.
A "photo" of a famous landmark covered in snow has perfect hands and clean text. A reverse search finds its first appearance on an AI art account, with the prompt in the caption.
Check yourself
Babak zooms in on a viral image and finds normal hands, readable text and sensible shadows. That proves it isn't AI-generated.
Show the answer
False
False. Flaws are a clue when you find them, but newer tools often leave none. The absence of flaws proves nothing. Checking where the image first appeared, and who posted it, is the stronger test.
DEEPFAKES
Words put in a real person's mouth
A deepfake is AI-made video or audio that shows a real person saying or doing something they never did. Voice clones can be made from short recordings, so a voice note that sounds exactly like someone is no longer proof. Check whether reputable news outlets or the person's own official channels carry it. Some cameras and platforms now attach Content Credentials (the C2PA standard) that record how an image was made; useful when present, but most images don't have them yet.
A voice note "from" a famous doctor recommends a product. Her official page and every major outlet say nothing about it. That silence is the clue.
Liar's dividend
A term from legal scholars Robert Chesney and Danielle Citron: once people know deepfakes exist, it becomes easier to dismiss real evidence by calling it fake. So "it could be AI" isn't a verdict in either direction. It's a reason to check the source.
A genuine recording surfaces of someone making an embarrassing remark. They reply, "That's obviously a deepfake." The claim needs checking just as much as the recording does.
Check yourself
A public figure is caught on a genuine recording and simply says, "It's a deepfake." The idea that fakes make real evidence easy to deny is called…
- The liar's dividend
- False balance
- A cheapfake
- Cherry-picking
Show the answer
The liar's dividend
Yes. The existence of deepfakes pays a "dividend" to people who want to wave away real evidence. The answer is the same as always: check the source, not the claim of fakery.
Two ways to check a striking image
Hunting pixels
Zooming in for ten minutes, looking for a strange finger or a blur. Finds some fakes, misses every real photo with a false caption, and gets less useful as AI tools improve.
Checking provenance
Thirty seconds of reverse search and a look at who posted it first. Catches old photos, wrong places and many AI images, and keeps working however good the fakes get.
Check yourself
Mina sees a stunning photo of lightning over a city, captioned as news from last night. It's beautiful and she wants to share it. What's the best move?
- Share it; it's too beautiful to be harmful
- Share it with "not sure if real" so others can decide
- Do a quick reverse search first, and share only if the caption holds up or with the correct context
- Never share photos of weather
Show the answer
Do a quick reverse search first, and share only if the caption holds up or with the correct context
Right. A beautiful picture shared as news is still a news claim. Thirty seconds tells her whether it's last night, years ago or made by AI, and then she can share it honestly.
Lesson recap
- Most misleading images are real photos with a false place, date or story.
- Reverse image search (Google Lens, Bing Visual Search, TinEye) finds earlier copies; the oldest usually tells the real story.
- Crops, slowed video, spliced audio and edited screenshots can change a story without inventing anything.
- AI images may show flaws, but a missing flaw proves nothing. Provenance beats pixels.
- For deepfakes and voice clones, check reputable outlets and the person's official channels.
- The liar's dividend: "it's a deepfake" can be used to deny real evidence, so check the source either way.