Spotting fake pictures, voices and videos

A video reaches your family group at night. A famous face promises that one app will double your money. It looks real, the voice sounds right, and your thumb is already on the forward button.

The old advice has stopped working

You have heard the checklist. Count the fingers. Look for a twisted hand, a strange ear, or nonsense writing on a signboard. That advice made sense a few years ago. It is not dependable now.

The tools that make these pictures improved faster than the checklists did. Hands come out normal. Signboards read correctly. Skin, hair and shadows look right on a small phone screen.

Some fakes still carry a mistake. Many carry none. So “it looks fine to me” tells you nothing at all. Your eyes are no longer a test you can rely on.

In July 2026 India’s cyber crime coordination centre gave a warning about fake investment adverts on Facebook and Instagram. The adverts used AI-made videos of the Prime Minister, union ministers and well known businessmen. They pushed fake trading apps. One rule comes out of that. A famous face plus a promise of sure money is a fake.

No app can tell you whether a picture is AI

You will want to upload the picture to an AI app and ask if it is real. Do not rely on that answer. The app will reply, it will sound sure, and it is guessing.

One AI company says this plainly in its own manual for developers. The model cannot work out whether an image was made by AI, and it may be wrong if you ask. That is the maker of the tool speaking, not a critic. It guesses here in the same way it guesses facts, which why AI makes things up explains.

Detector websites have the same weakness. Look at what happened with tools that claim to spot AI writing. A study published in 2023 tested seven popular ones on essays by people whose first language is not English. More than half of those essays were wrongly marked as AI written. One American university switched its detector off that same year.

Think about who that hurts. Simple, correct English gets flagged. A detector can call your own honest work fake, and let a real fake pass. Both mistakes are common. There is no magic checker, so stop hunting for one.

A label proves one thing only

Many AI tools now hide a marker inside the pictures, sound and video they make. Some phones and apps also attach a record of where a file came from. India has brought in rules for large platforms too. AI-made pictures, audio and video are meant to carry a clear label, and that label is not meant to be removed.

You can sometimes read the marker yourself. In the Google app on an Android phone, look for the option that searches using a picture instead of words. With the picture on screen, ask whether it was made with AI. This started reaching phones in May 2026, so it may not work on your handset yet.

Now the honest limit, and it is the half that matters.

A label proves the file is AI-made. If a file carries no label, that tells you nothing at all.

Markers do not survive normal life. A screenshot removes them. WhatsApp squeezes pictures to save data, and that removes them too. Many tools never add one. The labelling rules also cover pictures, sound and video, not written words. A whole message can be written by an AI with nothing on it to say so.

The four questions that still work

Stop asking whether the picture looks real. Ask about the claim instead. Four questions, in this order.

1. Where did this come from? Ask the sender who sent it to them.
2. Who published it first? A named news organisation, an office, or a person with a name.
3. Does any established source say the same thing? A newspaper you know, or the department’s own site.
4. Does it appear anywhere other than a forward? Search and see.

The fourth question settles most cases. A real event leaves a trail. If a bridge fell in your district, a district paper carries it. If a scheme opened, the department’s own site carries it.

Pick the five plainest words in the claim and search for them.

Claim in the video: a bridge collapsed in our district yesterday.
Type into a search box: bridge collapse (district name) news

A good result looks like a district newspaper page, a state news site, or a note from a government office. Read the end of the web address on any official looking page. Real Indian government sites end in gov.in or nic.in, as checking sources and links shows.

You can also search using the picture itself. Press and hold the image, then look for the option to search with it. The same picture may turn up in an older story about a different event. That is your answer.

The government runs a fact check unit under the Press Information Bureau. Search for its official page, and check the address ends in gov.in before you trust it. It publishes checks on viral scheme and job forwards.

An AI app is still useful here. Just do not make it the judge. Ask it to help you search.

A forward says a new central scheme gives free laptops to students.
Do not tell me whether it is true.
Give me five short search phrases to find the official notice.

A good answer comes back as plain search phrases, like these.

laptop scheme official notice gov.in
education ministry laptop scheme press note
laptop scheme fact check

Then you do the searching yourself. The judgement stays with you, which is where it belongs.

The feeling that tells you to slow down

Here is the test that catches most fakes, and it costs nothing.

Content made to be forwarded is made to make you angry or frightened. That is the design, not an accident. A calm message travels slowly. An angry one travels in minutes.

So treat your own feeling as a warning light. The stronger the pull to press forward, the harder you should check. Wait ten minutes. Nothing true stops being true in ten minutes.

Voices work the same way. An industry survey covered adults in seven countries, including India. Nearly half of the Indian adults had met an AI voice scam, or knew someone who had. The same report said about three seconds of recorded speech is enough to copy a voice. A familiar voice on a distress call is not proof of anything.

Do this instead. Cut the call and ring that person back on the number saved in your phone. Agree one family safe word today, and use it.

Ask on any call about money: what is our family safe word?
Then cut the call and ring back on the saved number.

AI scams and how they work goes through these calls in full.

Do not forward what you have not opened

One rule covers all of this. If you have not opened it and checked it yourself, it stops with you.

A communal or political forward is the dangerous kind. A message that names a religion or a caste, with a picture of violence, can bring people onto the road within an hour. In a village everybody knows everybody. One fake picture can start a fight between families who have lived side by side for years.

Deleting it later does not undo it. By then it has left your group many times over. Screenshots exist. People remember the message and not your correction. Inside that group you were the source, and lost trust takes years to rebuild.

There is a legal side as well. Some things are barred outright on Indian platforms. Fake news shown as real is one of them. A fake nude picture of anyone is another. So is pretending to be a real person. So is content that stirs hatred on religious or caste lines. Your account and your phone number can be traced. AI does not make anybody anonymous.

When a fake targets someone you know

One day this will hit a friend, a sister or a cousin. Move fast, and keep proof.

Do not forward it, not even to warn people. Every forward spreads it further.

Save the evidence first. Take screenshots. Save the web address. Note the date and the time.

Report the post on the platform itself, using its own report option. Indian rules now put platforms on a clock once such content is reported or ordered removed.

Then report at cybercrime.gov.in the same day. Complaints about crimes against women and children can be filed there without giving your name. If money has gone, call 1930 straight away.

A fake nude picture counts, even though the picture is not real. India’s rules for these images cover morphed and AI-made pictures. Platforms are required to remove a reported one within 24 hours. Never pay anyone who threatens you, and never delete the proof. Arrests do happen in India, so reporting is not a waste of your time. Safety for girls and young women takes this much further.

Pictures, voices and videos are one kind of fake. Whole apps and websites are another. Fake AI apps and copied websites comes next, and it shows how a copy of a real tool quietly takes your money. Keep every check you now have on one short page in your notes app.