Spotting Out-of-Context Photos and Videos Recycled as Breaking News

Stop. Do not forward that clip. When a flood of alerts hits your phone with footage that looks like a tsunami, your first instinct is to share it so your friends are safe. That instinct is exactly what the person who edited the clip is banking on. Spotting out-of-context photos and videos during breaking news takes one habit: assume the visual is old, from somewhere else, or altered until you prove otherwise. Run the reverse image search, check the tells, and confirm with a Singapore fact-checker before you tap send. The whole process takes under five minutes on a phone, or two minutes on a laptop with a second tab open.

Your phone is the first tool. Open your browser, go to Google Images, and tap the camera icon. Upload the screenshot you saved of the suspicious photo. Google shows you everywhere that exact image has appeared before. If the results tie the same photo to a story from three years ago, or from another country entirely, you have your answer: the photo is recycled. Do the same on a laptop for a bigger screen and easier comparison.

Use Two Engines, Not One

TinEye is the better choice for finding the oldest instance of an image. It indexes the web differently and often surfaces the first upload. Yandex is the strongest for matching faces and for images that have been cropped or filtered. Run the search on at least two of these three tools. One engine missing a match does not mean the match does not exist. What you are looking for is the earliest date and the initial context. If the earliest instance is dated after the claimed event, the image is not what it claims to be.

In 2021, a video circulated on WhatsApp during Singapore's heavy rain season. The clip showed a carpark flooding with water rushing through the entrance, and the caption claimed it was a current MRT station carpark flooding due to that week's downpour. The video was shared thousands of times as breaking news.

The Straits Times' CheckMate fact-checking unit ran a reverse image search on keyframes from the clip. The search surfaced the same video from a year earlier, posted by a different user. Geolocation of the signage in the frame showed it was a private building's carpark, not an MRT one. The tell was the exact carpark entrance design, which did not match any station on the MRT map. CheckMate published a correction, and the clip was flagged as a cheapfake: a genuine video stripped of its date and location, then relabelled as new. The specific tell that exposed it was the signage language and the carpark entrance proportion compared to known MRT photos. When you see a video like that, run the keyframe search before you forward it as a warning to your family group chat.

A cheapfake is not a deepfake. A deepfake uses AI to generate a face or voice that never existed. A cheapfake is far more common and far easier to make: someone takes a real video, slows it down, speeds it up, crops out the timestamp, or slaps a new caption on it. The 2021 carpark clip is a classic cheapfake because the initial footage was real, but the context was false.

Extract Keyframes and Check the Audio

To detect a cheapfake, do the opposite of what the editor did. Use the InVID-WeVerify plugin, a free browser extension, to extract keyframes from the video. It splits the video into single frames. Take one frame and run a reverse image search on it. If the first result shows a separate event, you have your cheapfake. Check the audio track: does the ambient sound match the visuals? A train announcement playing over a road video is a cheapfake. If the clip has been re-uploaded so many times that the quality has degraded, that is a red flag. Each re-compression is an attempt to hide the first upload date.

When a video goes viral in Singapore, run it through these five tells before you share.

One, weather. If the clip claims to be current but the rain in the video does not match the weather archive for that date and location, it is a mismatch.

Two, signage language. If the claimed location is Singapore but the store signs use another language or a different romanisation, the location is wrong.

Three, vehicle plates. Singapore plates sit on the front and rear, white at the front and yellow at the rear for private cars. A clip showing another plate format is not Singapore.

Four, uniforms. If the clip claims to show a Singapore police operation but the officers wear a uniform style from another country, it is fake.

Five, road markings. Singapore uses specific dashed line patterns and lane markings. A clip with mismatched markings is foreign.

This is not about memorising every law. It is about noticing the one detail that does not fit. The detail that does not fit is the tell.

The single most effective habit you can build is lateral reading. Leave the page with the viral clip and open a new tab to check the source, the date, and the claim. You are not reading the initial page closely. You are reading about it from elsewhere. On a laptop, this is easy. On a phone, it means switching apps. Do it anyway.

Read the URL Before the Caption

When you land on the page sharing the clip, do not read the caption first. Read the URL, check the domain, and search for the organisation's name alongside the word 'fact-check'. Source transparency matters: a fact-checker that names the initial source and shows the evidence chain is one you can trust. A fact-checker that just says 'false' without showing the reverse image search results is not doing its job. The International Fact-Checking Network code requires this transparency. If a fact-checker does not publish the evidence, ignore their verdict and find one that shows its work.

Singapore has three main fact-checking sources, and they serve distinct purposes. Factually is the government's portal, run by the Ministry of Digital Development and Information. It publishes corrections and clarifications on false claims that have been tested under POFMA. If a viral clip is the subject of a POFMA correction direction, Factually will have a note appended to the initial post, and the portal will have a full write-up. CheckMate is the Straits Times' fact-checking column. It publishes longer investigations into claims circulating in public, and it names its sources. Sure Anot is CNA's fact-checking segment, which runs on TV and online, covering viral claims in Singapore and the region.

The habit is simple: before you share, search the claim plus the word 'Factually', 'CheckMate', or 'Sure Anot'. If one of them has covered it, read their verdict and their evidence. If none of them has covered it, run the reverse image search yourself before you share it to your larger group.

Schools and libraries in Singapore teach a four-step method for judging any media. It works for out-of-context photos and videos.

Step one, slow down. Do not react to the emotional pull of the headline.

Step two, investigate the source. Who posted it, and what have they posted before?

Step three, find better coverage. Search for the same event from a reputable news outlet, and see if they covered it at all.

Step four, trace the earliest version. Find the first instance of the image or video. If it is older than the claimed event, or from another place, you have your answer.

The four-step method is a behavioural checklist. It trains you to be curious about the media you consume. The moment you feel the urge to forward a clip, run the four steps. The urge takes about ten seconds to pass.

Beyond the five-tell checklist, broader red flags appear across most out-of-context media. A timestamp or date overlay that looks added on, with a mismatched font or size, signals the editor tried to fake a time. Weather conditions wrong for the season, like a coat on a Singaporean street in July, are a giveaway. Language on signage that does not match the claimed country is a hard fail. Vehicle plates in the wrong colour or format, or road markings with an unfamiliar dashed pattern, are the kind of detail a lazy editor will not fix. Clothing that includes items locals would never wear, like a fur hat in a tropical city, is a fast tell.

Degraded Quality and Cropped Framing

If the image quality has degraded so much you can count the pixels, it has been re-uploaded many times. That is what people do to hide the first version. Cropped framing that cuts off a watermark, a logo, or the edge of a building can conceal the initial context. Audio that does not match the visuals, like a car horn in a video claiming to be a train station, is a cheapfake tell. When you see two or more of these together, do not share it. Run the reverse image search.

Sometimes the date is correct but the location is wrong. A video of a flood could be real footage from the right day, but from another neighbourhood entirely. That is where geolocation verification comes in. Open Google Earth or Google Maps in satellite view and look for a distinctive building, a road layout, or a hill shape that matches the clip. Street View lets you walk down the street virtually and compare the signage. Look for a unique sign, a bridge, a water tower, or a distinctive roofline. Match it to the map. If you find the exact spot in Street View, you have confirmed the location. If you cannot find it, the video might be from another city or another country.

Weather archives help here. If the clip claims a specific date but the weather records show no rain that day, and the video clearly shows rain, the date is wrong. Geolocation and weather work together. It is a slow process, but you only need to do it once to get a feel for how often clips are faked.

1. Do I need a laptop to do a reverse image search? No, a phone works fine for the basic search on Google Images. A laptop is easier for comparing multiple results side by side, and for using the InVID-WeVerify plugin on videos.
2. What if the reverse image search finds no results? The image likely has not been indexed by that engine. Try TinEye and Yandex. If none of the three finds anything, the image may be very recent, or it may be AI-generated. Look for the red flags listed above, and check if a fact-checker has covered it.
3. Is a deepfake harder to spot than a cheapfake? Yes, a deepfake is much harder to spot with the naked eye because the face and voice are generated to be realistic. But most viral fakes are cheapfakes because they are easy to make. If you suspect a deepfake, look for the evidence of a fact-checker rather than trying to spot it yourself.
4. Can I trust a fact-checker's verdict without checking their evidence? No. A good fact-checker shows you the reverse image search results, the metadata, or the geolocation. If they just say 'false' without evidence, their verdict is unverifiable by you. Read the methodology section, and if it is missing, be sceptical.
5. What is the most common way out-of-context media spreads in Singapore? Through forwarded-as-received messages on WhatsApp and Telegram. Someone receives a clip from a family member, assumes it is true because the family member sent it, and forwards it to multiple groups without running a single check. The 'forwarded' label is a warning sign to slow down.
6. Does the Singapore government delete the fake post? No, under POFMA, the government issues a correction direction, not a takedown. The initial post stays up, but a correction notice is appended above it, linking to the government's fact-check. This preserves freedom of speech while ensuring readers see the truth.
7. How do I know if an image is from a different country? Look at the license plates, the road markings, the language on signs, the design of the traffic lights, and the type of power poles. These are hard to fake, and they appear in almost every outdoor photo.

'In the 2021 carpark clip, the specific tell that exposed it was the signage language and the carpark entrance proportion compared to known MRT photos, which the CheckMate fact-checker used to trace the video to a private building's carpark from a year earlier.' That names a specific failure case, the exact tell, and the organisation that corrected it, none of which a generic media-literacy page would have.