AI Deepfakes — How to Tell What Is Real and What Is Fake.

Introduction: When Seeing Is No Longer Believing

Imagine watching a video of a politician making a shocking announcement, hearing a celebrity confess to something unexpected, or receiving a voice message that appears to come from someone you know. The content looks convincing, the voice sounds familiar, and the recording seems genuine.

But what if it never happened?

Artificial intelligence is making it increasingly possible to create realistic images, videos and audio recordings that imitate real people or depict events that never took place. These creations are often called deepfakes.

Deepfake technology can be used for entertainment, creative projects, education and visual effects. However, it can also be misused to spread false information, impersonate people, commit fraud, damage reputations or create intimate images without consent.

As synthetic media becomes more sophisticated, learning how to assess what we see and hear online is an important part of digital safety.

The key is not to assume that every video is fake or that every image is genuine. Instead, we need to question the source, check the context and look for reliable evidence before deciding what to believe or share.

What Is an AI Deepfake?

A deepfake is a piece of synthetic or manipulated media that uses artificial intelligence to imitate a person, alter their appearance or voice, or create a realistic representation of something that did not happen.

The term is commonly associated with AI-generated or AI-manipulated videos, but similar techniques can also be used to produce images and audio recordings.

For example, a deepfake might:

• Make it appear that a person is saying words they never said.
• Change a person's facial expressions or movements in a video.
• Generate a realistic image of a person in a situation that never occurred.
• Imitate someone's voice to create a fabricated recording.
• Make a fictional person appear to be a real individual.
• Alter a genuine recording so that it communicates a misleading message.

Not every form of manipulated media is a deepfake. Some misleading content is produced through ordinary editing, selective cropping, misleading captions or the reuse of old footage. AI is only one of the ways digital content can be made deceptive.

Understanding this distinction matters because misinformation can spread even when no sophisticated AI technology has been used.

How Are Deepfakes Created?

AI systems can learn patterns from examples of images, voices, facial movements or other data. Depending on the method, they can then generate new material or transform existing content.

Some systems create images or videos from written instructions. Others can imitate a voice from audio samples or modify the appearance and movements of a person in existing footage.

Modern generative AI can produce convincing results, particularly when the person viewing the content has little reason to question it. The quality of a result depends on the technology, the available material, the instructions and the circumstances in which the content is produced.

Importantly, a realistic appearance does not prove that a recording is authentic. Equally, an unusual visual detail does not automatically prove that a recording is fake.

Different Types of Deepfakes

1. Deepfake videos

Deepfake videos can alter a person's face, facial expressions, lip movements or apparent actions. They may be used to create fabricated speeches, misleading public statements or fictional scenes that look like real footage.

A video may contain a mixture of genuine and synthetic material. For example, the background and setting might be real while the person's face or spoken words have been manipulated.

2. AI-generated images

AI image generators can create realistic-looking photographs of people, places and events that never existed. They can also alter existing photographs or place a real person into a fabricated situation.

Such images may be used in harmless artwork, but they can also be presented deceptively as evidence of an event or as a photograph of a real person.

3. AI voice cloning

Voice cloning attempts to imitate the sound of a person's voice. A fabricated recording may be used to make someone appear to say something they never said or to impersonate a relative, colleague, public figure or business contact.

Voice cloning is particularly relevant to fraud because people may trust a familiar voice, especially when a message appears urgent. A familiar voice alone is not proof of identity.

4. Real-time face and voice manipulation

Some systems can modify a person's apparent face or voice during a live call or broadcast. This may be used for entertainment, but it can also be abused to impersonate another person during an online meeting or video call.

When a request involves money, confidential information or account access, verification should take place through a separate, trusted channel rather than relying solely on what appears on the call.

Why Are Deepfakes a Concern?

Deepfakes can create confusion about what is real, who said something and whether an event actually happened. The potential harm depends on the content, how it is used and who is affected.

Misinformation and public trust

A fabricated video of a public figure may be circulated as genuine footage. If people believe it without checking the source, it can distort public discussion and influence how events are understood.

False content can also be used to impersonate journalists, businesses, public services or organisations. During a developing news story, misleading material may spread before reliable information becomes available.

Fraud and impersonation

Criminals may use cloned voices, fabricated videos or impersonated identities to make requests for money, sensitive information or urgent action.

A message may appear to come from a manager, family member or trusted professional. The safest response is to verify the request independently, especially when it involves payments, passwords, personal details or unusual instructions.

Harassment and non-consensual imagery

Deepfake technology can be misused to create humiliating, sexualised or otherwise harmful images of real people without their consent. Such material can cause serious distress and reputational damage even when the content is entirely fabricated.

Sharing or reposting this material can increase the harm. Do not forward intimate or humiliating synthetic images simply because they are circulating online. Respect the person's privacy and use appropriate reporting channels.

Damage to reputations and relationships

A convincing fake recording can lead people to believe that someone has behaved dishonestly, made offensive comments or taken part in an event when they have not.

Before accusing someone or making a public claim based on a recording, seek reliable confirmation. Publishing a correction after a false allegation has spread may not fully undo the damage.

Undermining genuine evidence

There is another risk: people may dismiss authentic photographs, recordings or videos by claiming they are AI-generated.

This creates a difficult situation in which fabricated material may be believed while genuine evidence is rejected. The appropriate response is to assess evidence carefully rather than automatically accepting or dismissing it.

How Can You Tell Whether an Image or Video Is Fake?

There is no single visual trick that reliably identifies every deepfake. AI-generated media is improving, and genuine recordings can also contain unusual visual or audio details because of lighting, compression, camera quality or editing.

Instead of relying on appearance alone, use several checks together.

1. Check the original source

Ask where the content first appeared. Was it published by a recognised news organisation, an official account or the person it supposedly features? Is there a clear source for the original recording?

A post with no source, no date and no explanation deserves additional scrutiny. However, an account that appears official is not automatically trustworthy, because accounts can be impersonated or compromised.

2. Look for independent confirmation

If a video claims that a major event has happened, check whether credible and independent sources are reporting the same thing.

For statements attributed to a public figure, look for a complete recording, an official transcript or a reliable report that provides context. Several websites repeating the same unverified post do not necessarily count as independent confirmation.

3. Check the date and context

Real footage can be used misleadingly when it is old, taken from a different location or presented as evidence of an unrelated event.

Check when the content was first published, where it was recorded and whether the accompanying description matches the original material. A genuine video can still support a false claim if its context has been changed.

4. Look carefully at the details

When reviewing an image or video, you can look for inconsistencies such as unusual reflections, distorted objects, inconsistent lighting, strange text or movements that do not appear to match the scene.

In audio or video, you might notice unnatural transitions, mismatched lip movements or changes in sound quality. These can be reasons to investigate further, but they are not conclusive proof of manipulation.

Some genuine recordings contain similar imperfections, and some convincing deepfakes have few obvious visual clues. Do not rely on a checklist of supposed AI mistakes as your only test.

5. Search for the original image or footage

Reverse image search and other visual search tools can help locate earlier appearances of an image. They may reveal that a photograph is old, that it has been cropped or that it was originally published with a different explanation.

For video, searching for distinctive frames or phrases from the alleged speech may lead to a longer or original recording.

These tools can help establish context, but they do not always find the original source. Failure to find an earlier version does not prove that content is genuine or fake.

6. Check available provenance information

Some digital content includes provenance information that records details about its origin or editing history. Standards and systems such as Content Credentials can help communicate how a piece of media was created or changed.

When this information is present and its authenticity can be verified, it may provide useful evidence. However, provenance information is not available for every file, can be lost when content is copied or transformed, and does not automatically establish whether the content's claims are true.

The absence of a content credential is not proof that a file is fake.

Can AI Deepfake Detectors Identify Fake Content?

Some tools attempt to identify AI-generated or manipulated images, videos and audio. They may analyse patterns or characteristics that are associated with particular types of synthetic media.

These tools can sometimes provide useful additional evidence, but they should not be treated as a definitive test.

A detector may incorrectly flag genuine material or fail to recognise a sophisticated fake. Results can vary depending on the type of media, the generation method, compression, editing and whether the detector was designed for that particular content.

A score or label from an automated tool should therefore be interpreted cautiously. It is not the same as independently establishing the origin of a recording.

For important decisions, combine technical checks with source verification, contextual research and confirmation from reliable sources. Do not make serious accusations about a person based only on a detector result.

What Should You Do If You Receive a Suspicious Video or Voice Message?

If a recording appears to show someone making a shocking statement or asking for urgent help, pause before reacting.

• Do not act immediately: Be cautious about urgent demands for money, confidential information or unusual action.
• Verify the person's identity: Contact them through a phone number or communication channel you already trust.
• Use a separate channel: Do not rely solely on the same call, message or account that delivered the suspicious request.
• Check the claim: Look for the original source and independent confirmation.
• Protect personal information: Do not provide passwords, verification codes, financial details or sensitive information just because a voice or face seems familiar.
• Avoid forwarding it: Do not spread the content as fact while its authenticity is uncertain.
• Report suspected abuse: Use the platform's reporting tools and seek appropriate help if the content involves threats, fraud, harassment or non-consensual intimate imagery.

If a supposed family emergency is involved, use a previously known number to call the family member directly. A family verification phrase may also help in some situations, but it should not replace other checks if the phrase could have been discovered or shared.

If money has already been transferred because of a suspected impersonation scam, contact your bank promptly using its official contact details and report the incident to the relevant authorities.

How to Protect Your Own Identity and Voice

It may not be possible to prevent every attempt to imitate your face or voice, particularly when recordings of you are already publicly available. However, sensible precautions can reduce some risks.

• Review what personal information and recordings you make publicly available.
• Be thoughtful about sharing sensitive information, identity documents or private recordings online.
• Use strong, unique passwords and enable multifactor authentication on important accounts.
• Verify unexpected requests for money or confidential information, even when they appear to come from someone you know.
• Keep important accounts and devices updated and secure.
• Review privacy settings on social media and other services.
• Report impersonation accounts and harmful fabricated content through the relevant platforms.
• Keep evidence of suspected abuse where it is safe and appropriate to do so.

Publicly available material can sometimes be used to imitate a person, but there is no simple rule that every online voice recording or photograph can be turned into a convincing deepfake. The risks depend on the available material and the technology being used.

Deepfakes, Consent and Digital Responsibility

Just because technology makes something possible does not mean it is ethical to do it.

Using someone's likeness or voice in a misleading way can violate their privacy, damage their reputation or cause distress. Creating or distributing sexualised synthetic imagery of a real person without consent is particularly harmful.

Creators should consider whether they have permission to use someone's likeness, whether audiences could be misled and whether synthetic material could cause harm.

Where AI-generated or materially altered content could reasonably be mistaken for genuine footage, clearly labelling it can help audiences understand what they are seeing. Labels should be accurate and visible enough to provide meaningful context.

Consent, transparency and respect should be part of responsible media production. People should not have to discover that their identity has been used deceptively after the content has already spread.

Deepfakes in News and Journalism

News publishers and content creators have an important responsibility when handling potentially manipulated media.

Before publishing a suspicious recording, journalists should establish where it came from, consider whether it has been edited, seek independent verification and distinguish confirmed facts from unverified claims.

It is also important to avoid amplifying harmful content unnecessarily. In some cases, explaining the claim and providing reliable evidence may be more responsible than repeatedly displaying a deceptive image or video.

When a mistake is made, a clear correction helps audiences understand what was wrong and what the evidence now shows.

For readers, the same principles apply: examine the source, question the context and avoid treating a viral post as verified news simply because it has attracted attention.

Can Deepfake Technology Be Used Positively?

Although deepfakes can be misused, AI-generated and AI-manipulated media also have legitimate creative and practical applications.

Depending on the circumstances, similar technologies may support visual effects, fictional storytelling, dubbing, creative experimentation, educational demonstrations and the reconstruction of material for authorised projects.

Responsible use depends on the purpose, the permissions involved, the treatment of the people represented and whether the audience is being misled.

Clearly fictional content, authorised visual effects and transparent creative projects are different from deceptive impersonation designed to exploit trust.

The goal should not be to reject every use of synthetic media. It should be to encourage innovation while protecting consent, privacy, safety and public trust.

Conclusion: Think Critically Before You Believe or Share

AI deepfakes are changing how images, videos and audio can be created and manipulated. They can provide new creative possibilities, but they can also be used for fraud, harassment, misinformation and impersonation.

There is no perfect visual test that can identify every fake, and no automated detector can guarantee that every result is correct. A convincing recording is not proof that an event happened, just as an unusual image is not automatic proof of deception.

The most reliable approach is to examine the source, verify the context, seek independent confirmation and treat technical detection tools as supporting evidence rather than final answers.

Before sharing a shocking video, forwarding a suspicious voice message or accusing someone based on a recording, take a moment to check the facts.

In a world where digital content can be generated and altered with increasing ease, critical thinking, consent and responsible sharing are essential parts of digital safety.

Seeing should encourage us to investigate, not stop us from asking questions.

Further Information and Resources

• Coalition for Content Provenance and Authenticity (C2PA) — information about technical standards for communicating the origin and editing history of digital content.
• National Institute of Standards and Technology (NIST) — research and resources relating to AI, digital identity and trustworthy technology.
• UK National Cyber Security Centre — cybersecurity guidance for individuals and organisations.
• Action Fraud — information on reporting fraud and cybercrime in England, Wales and Northern Ireland.
• Ofcom — UK information about online safety and communications regulation.

AM-MediaSpace encourages responsible AI use, digital safeguarding and informed media consumption. When content raises questions, check the evidence before you believe it or share it.