Short answer

Visual tells such as distorted hands, odd shadows and unnatural movement can reveal some AI-generated media, but they are unreliable and fading. Provenance labels such as C2PA Content Credentials can confirm origin when present, though a missing label proves nothing. Checking the source, context and a second channel is more dependable.

Key takeaways

  • Visual clues catch some fakes and miss many, and good fakes have none.
  • Context checks (who posted it, where else it appears) beat pixel inspection.
  • C2PA Content Credentials record where a file came from, but most media carries none.
  • A missing label does not mean fake, and a present label does not mean true.
  • For money or safety decisions, verify through a separate channel regardless of how real it looks.

Start with the honest answer

You cannot reliably tell a good AI fake from a real photo or video by eye. Guides that list “count the fingers” were useful for early generators, but newer tools often get hands, text and lighting right. If you are told that a quick look will protect you, be sceptical.

That does not mean you are helpless. It means moving from “does this look fake?” to “should I trust where this came from?”

Visual clues that still sometimes help

The FBI lists several signs in its 2024 alert on generative AI fraud:

  • Distorted hands or feet, unrealistic teeth or eyes, and irregular faces
  • Shadows and lighting that do not agree
  • Unnatural movement, or lag between lips and sound
  • Backgrounds that warp, or text that is garbled

Use these as prompts to look harder, not as proof. A clean result proves nothing, and ordinary compression or a poor connection can cause some of the same glitches in real footage.

Checks that work better

Find the source. Who posted this first? An anonymous account with a new profile is weaker than a named news organisation or the person themselves.

Look for the same event elsewhere. A real major event is reported by several outlets. If only one unknown account has the clip, wait.

Reverse image search. Upload the image, or a screenshot of a video frame, to a reverse search tool. Often the picture is old, from another event, or a stock image.

Ask what you are being asked to do. Many fakes have a goal: send money, join a crypto platform, share a password. The FBI notes fabricated promotional videos used in investment fraud and fake videos used to make an online contact seem real. Our guide to AI voice clone scams covers the audio side.

What provenance labels can and cannot do

C2PA is an open standard behind Content Credentials. A supporting camera, editing tool or AI generator can attach a signed record of where a file came from and what was changed. When you see one, it is a helpful clue about origin.

The limits are important:

  • Most images and videos online carry no credentials, real or fake.
  • The C2PA FAQ says metadata can be removed, inadvertently or intentionally, and manifests can be separated from the file.
  • A label says how a file was made, not whether what it shows is true.
  • Anyone can create a genuine, labelled image of a staged scene.

NIST’s 2024 report on synthetic content (AI 100-4) surveys provenance, watermarking and detection as separate technical approaches to content transparency. No single one settles the question.

A rule for high-stakes moments

When money, safety or reputation is involved, treat the media as unverified and confirm another way. If a video “from your boss” asks for a payment, call them back; see deepfake video call fraud. If a celebrity seems to endorse an investment, go to their verified channels, not the link. If a message pushes you to act quickly, see phishing messages.

Slowing down is the most effective defence, because scams depend on speed.

What you can do

  1. 01Pause before sharing or acting on shocking images, video or audio.
  2. 02Search for the original source and check whether reputable outlets report the same event.
  3. 03Run a reverse image search on the picture or a video frame.
  4. 04Check for Content Credentials on the file, treating them as one clue and not a verdict.
  5. 05Never send money or details because of a video, however real; confirm through a number you already have.
  6. 06Report fake investment videos and impersonations to the platform and to your national fraud service.

Related reading: AI voice clone scams and the data breach tracker.

Frequently asked questions

Can I reliably tell if an image is AI-generated by looking?

No. Clues like distorted hands or teeth, inaccurate shadows and unnatural movement can help, as the FBI notes, but generators improve quickly and many fakes show none. Treat visual checks as a hint.

What are Content Credentials?

Content Credentials are the label built on the C2PA open standard. They record who or what created a file and how it was edited, like a nutrition label for media.

If a picture has no Content Credentials, is it fake?

No. Most real photos have none, and the C2PA FAQ says credentials can be removed or separated from a file. Absence tells you nothing either way.

Do AI detection tools work?

They can give a probability but make mistakes in both directions. NIST's report surveys detection as one approach among several, alongside provenance and watermarking. None is complete.

Sources

  1. Criminals Use Generative Artificial Intelligence to Facilitate Financial Fraud (PSA I-120324), FBI Internet Crime Complaint Center, 2024-12-03
  2. NIST AI 100-4: Reducing Risks Posed by Synthetic Content, National Institute of Standards and Technology, 2024-11-20
  3. C2PA frequently asked questions, C2PA, 2026-11-05

This article reflects public information as of 2026-11-05. AI tools and rules change quickly; we update articles when new facts are published. Spot an error? See our editorial and corrections policy.