How to verify images and videos

Images, videos and audio clips frequently go viral on social media purporting to show something that later turns out to be false.

They might be real images but with the wrong context, they might be digitally manipulated, or they might be entirely generated using artificial intelligence (AI). They often relate to divisive or emotive topics such as elections and conflicts.

The following questions can help investigate whether an image or clip of video or audio is real and shows what it purports to show. However, the answers to these questions might not necessarily resolve a report’s truth or falsehood forever: uncertainty is an inherent part of knowledge and knowledge is improved and revised over time.

This article is part of the Commons Library’s good information toolkit, which reviews the strategies and criteria for spotting misinformation and working out what information can be relied upon.

Who is posting it?

If the identity of the person posting the image or video is unclear, it is harder to judge whether it’s plausible for them to have recorded it. It is also impossible to assess how credible they are as a source.

Most social media platforms do not require users to verify their identity, and it is therefore easy to set up fake ‘sockpuppet’ accounts or create bots (programmes controlling an account to make posts or otherwise use the platform).

Users usually cannot always tell whether an account belongs to a real person, as social media accounts are often anonymous or pseudonymous.

Knowing who posted something also makes it possible to assess their reputation and associations, to see whether they are trustworthy or whether they have any conflicts of interest. For more, see the good information toolkit article on How to evaluate sources.

Is it the original source?

If an image or video was posted on social media, it might be original (posted by the person who recorded it), or it might have been shared from another source.

Finding the original source is important because it might come with extra information that can either verify or falsify the image or video. For example, videos often appear on social media claiming to show an event, but if the original source posted the video before the event occurred, the claim cannot be true.

If the poster says where the image or video came from, or it contains information about its source such as a watermark, readers can check the source to find the original version of it (for example, by using a web search). The original source might provide more context about where the image or video was recorded than is present on social media.

If the poster doesn’t say anything about the source, readers can search the internet using keywords that describe the image or video to see where else it’s mentioned.

Reverse image searches

They can also use a reverse image search tool (such as TinEye, Bing, Google, InVID verification and others) to see whether the image or video has been published online before. This works by uploading the file to the tool and searching the internet for the same and similar files.

If the poster claims they are the source of an image or video, a reverse search can be used to refute this. If the file already exists somewhere online, readers can see whether the context of the other instances support or refute what the post said.

Metadata

Files sometimes have metadata: this is information about the file stored within it. An image taken by a camera or smartphone automatically contains information about where and when the image was taken.

However, most social media platforms remove this information when an image is uploaded, and it can be edited by any user who downloads the file, making it an unreliable method of determining a file’s source.

Is it the only source?

A single image or video clip purporting to show something mgiht be less likely to be real than a collection of images and videos from independent sources independently showing the same event from different angles and perspectives.

If there are more versions of photo evidence of something, it is possible to compare the different versions to look for inconsistencies or other evidence that the image might not be real.

Does it show what it says it shows?

Images and videos are often shared with text or a voiceover explaining what it is supposed to show. Even if the image or video is genuine, the explanation of what it shows might be wrong.

For example, a video posted in 2024, viewed more than 278,000 times, said it showed an explosion on the Frances Scott Key Bridge in Maryland, USA. The video actually showed the explosion in 2022 on Kerch Bridge, which links the Crimean Peninsula with Russia. The Frances Scott Key Bridge collapsed in March 2024 after a container ship collided with it, but there was no explosion.

Some events will be covered in detail by news organisations, so checking these can help corroborate other sources (for example, several news sources reported the Frances Scott Key Bridge collapse at the time it happened, such as the BBC, NPR and Institute for Civil Engineers).

Open-source intelligence

Even when news organisations aren’t reporting an event, cross-referencing details in the image and the context with other sources can help verify whether the context is accurate. This work is frequently done by open-source intelligence (OSINT) researchers.

For example, if someone claims an image shows a particular location, OSINT researchers might look for landmarks, banners, adverts and signs in the image that they expect to be specific to a location. They can then use open-source tools and software, such as interactive maps, satellite imagery, and other images and videos posted on social media, to look at the location and see if the details match up for a particular date.

This can be a long and difficult process for readers to do themselves, but there are also many media organisations that conduct similar investigations when they fact check images and videos online. For more information, see:

• Bellingcat, A Beginner's Guide to Social Media Verification, November 2021. Bellingcat is an international investigative research organisation that uses open-source research methods to investigate issues and claims of public interest. It publishes the results of its investigations as well as the methods it uses.
• FullFact, How to spot misleading images online, April 2023. FullFact is a UK charity that checks claims related to UK current affairs and campaigns for better information and accountability in politics and the media.
• BBC Verify. This is a part of BBC News that explains how the organisation verifies the information it reports. It posts articles and videos demonstrating how footage is verified, such as footage of air strikes in Lebanon, or refuted, such as AI-generated footage of earthquake damage in Myanmar.

Are there inconsistencies?

The backgrounds of an image or video can sometimes reveal details that are inconsistent with what it is supposed to show. There might also be inconsistencies between how someone behaves in a video and how they would be expected to behave based on available information. Check the plausibility of:

• Language, speech or writing. Is it in the language you would expect? (This may include checking any visible writing in an image.)
• Equipment, dress, and physical context/landscape. Do they appear appropriate to the claimed context and time?
• Signs of inappropriate image/video manipulation. These could include selective or narrow framing of an image (though cropping or blurring) or signs that a video is not a single sequence.
• Geo-locating an image/video. Features and landmarks can be used to determine the location.

For example, if a picture is supposed to show an event in the United States but there are banners in the background with text in Norwegian, it might indicate the image was not taken in the US.

Is it generated by AI?

Generative AI is becoming increasingly convincing, and it can be difficult to tell just by looking at an image or video whether it is real, enhanced by AI or totally generated by AI. However, the same principles apply in investigating AI-generated images as with any other kind of image.

Some AI-generated images and videos can be identified because of mistakes the programme has made, particularly in the small details. Examples of this include:

• people having the wrong number of fingers or teeth
• lip movements not consistent with the words someone is saying
• patterns being regular where you’d expect disorder, or distorted where you’d expect them to be regular
• inconsistency with the physical world, such as:
  • light, shadow and reflections that seems to defy physics
  • incorrect perspective (such as straight lines not converging to a single vanishing point)
• text that doesn’t correspond to any language
• background features (such as people’s faces) being distorted

AI-generated audio might pronounce words in a way inconsistent with someone’s known accent or contain unnatural pauses or odd background noise.

However, as AI programs improve, these problems are being solved, making it impossible to say definitively that an image is not AI-generated because, for example, someone’s fingers look correct.

Instead, use the same principles of verifying online images to check something suspect of being an AI fake:

• Locate the original source (for example, using a reverse image search). Images originally generated as a joke or to illustrate a point are sometimes reshared as being real, even if the original source acknowledges it as a fake.
• Compare what the image is supposed to show with other images of the same location at the same time.
• Look for inconsistencies within the image or video and in the behaviour of people in videos.