How to Tell If an Image or Video Is AI-Generated

AI-generated images and videos have become convincing enough that deciding whether something is real can no longer depend on spotting strange fingers or distorted faces.

Those clues can still appear, but AI systems continue to improve. Real photographs can also contain blur, compression artifacts, unusual lighting and editing mistakes that resemble supposed signs of AI generation.

The better approach is verification.

Instead of asking only, “Does this look fake?” ask: Where did this media come from, and what evidence supports its origin?

Start With the Source

Before examining individual pixels, find the account or website that originally published the material.

Ask:

• Who uploaded it?
• Is that person claiming to have captured it?
• Is there an original version?
• Does the account have a history connected to the location or event?
• Have credible organizations confirmed the footage?
• Is the uploader simply reposting someone else?

An impressive video appearing on an anonymous account without a source deserves more caution than media published with identifiable provenance.

Search for Earlier Versions

One of the most useful verification techniques is finding where an image appeared previously.

Visual-search tools such as Google Lens can help locate similar images across the web.

Google’s “About this image” feature can also provide information about when Google may have first encountered the image or visually similar copies.

This can expose a common form of misinformation: genuine media reused with a completely different story.

For example, a photograph might really show a flood—but not the flood mentioned in the viral caption.

Check the Context

Look beyond the central subject.

Background details can sometimes provide valuable evidence:

• street signs;
• store names;
• vehicle models;
• weather conditions;
• shadows;
• buildings;
• uniforms;
• languages;
• clocks;
• or landmarks.

Compare these details with what the post claims.

If a video supposedly happened in one location but visible evidence points elsewhere, further investigation is warranted whether AI was involved or not.

Look for Provenance Information

A growing approach to digital-media verification involves Content Credentials.

The Coalition for Content Provenance and Authenticity, or C2PA, has developed an open standard designed to provide information about a digital asset’s origin and editing history.

Content Credentials may contain information about how media was created and what changes were made to it.

Think of them as one potential source of provenance information rather than a universal truth detector.

Content provenance can help establish a media file’s history. It does not determine whether every statement made about that media is accurate.

Likewise, the absence of Content Credentials does not automatically prove that an image is fake.

Examine Visual Problems Carefully

AI-generated media can still contain inconsistencies.

Possible warning signs may include:

• unreadable or changing text;
• physically inconsistent reflections;
• objects appearing or disappearing between video frames;
• strange interactions between hands and objects;
• inconsistent architecture;
• unnatural movement;
• or details that change unexpectedly.

But none of these should be treated as proof by themselves.

Compression, stabilization, poor lighting, editing and low-quality cameras can also create unusual artifacts.

Don’t Trust an AI Detector as the Final Answer

Automated detection systems can be useful as one signal, but they should not replace verification.

Media can be resized, compressed, edited, screenshot, filtered or re-uploaded repeatedly.

Those changes can affect the information available to detection tools.

A responsible conclusion should therefore rely on several pieces of evidence rather than a single percentage produced by a detector.

Check Whether the Event Exists Outside the Post

Suppose a viral video claims to show an extraordinary explosion in a major city.

Search for evidence that such an event actually occurred.

Check:

• emergency agencies;
• local authorities;
• credible local journalists;
• established news organizations;
• eyewitness material;
• or other independent footage.

A massive public event would normally create evidence beyond one anonymous social-media upload.

This does not mean a story is automatically false when major outlets have not covered it. It means the absence of supporting evidence should affect how confidently the claim is presented.

Separate Synthetic Media From Misleading Media

Not all misinformation is AI-generated.

A real image can be just as misleading when paired with:

• the wrong location;
• the wrong date;
• an invented explanation;
• a fabricated quote;
• or an unrelated event.

Sometimes the important question is not whether the media itself is authentic.

The important question is whether the story attached to it is accurate.

A Better Verification Checklist

Before treating questionable media as authentic or synthetic, check:

Source: Who originally published it?

History: Can you find earlier versions?

Context: Do the location, date and surrounding details match the claim?

Provenance: Is trustworthy information available about how the media was created or edited?

Independent evidence: Has anyone else verified the event?

Original material: Can you locate a higher-quality or unedited version?

Only after those steps should visual anomalies or automated AI-detection results be used as additional evidence.

What you can responsibly conclude

The era of deciding whether an image is real simply by looking at it is ending.

Verification increasingly depends on provenance, sourcing, context and independent confirmation.

AI makes that process more important—but the principles themselves are not new.

Find the original source. Trace the media’s history. Check the context. Look for independent evidence.

And when the available evidence cannot establish what happened, the most accurate conclusion may simply be: unverified.

A designer’s evidence ladder

For an image used in a poster, thumbnail or social card, judge three different things: the source file’s history, the changes made during production, and the claim in the final caption. A legitimate design edit does not make the caption accurate, and an authentic photograph does not make an invented quotation authentic.

1. Ask for the source asset. An original public file or a documented export offers more to inspect than a screenshot. Record the supplied filename, source URL, creator attribution and the date you received it. A filename such as “original.jpg” is a label, not proof.
2. Record the production steps. Note cropping, compositing, retouching and any generative additions. Keep a source version and a final version so a later reviewer can understand what changed. Do not call a composite an untouched photograph.
3. Check the editorial claim. Verify names, quotations, event date and location independently of the image. If the source only says “illustration,” keep that qualification in the caption.
4. Make the uncertainty visible. A useful label is “AI-generated illustration; does not depict a documented event.” If the origin is unresolved, say so rather than assigning a detector’s score as a verdict.

Worked example: a fictional event poster

Imagine a poster showing a crowded town square beneath the words “Yesterday’s protest drew 20,000 people.” This is a fictional exercise, not a real protest or an analysis of an actual photograph.

The designer supplies a real stock photograph and says the crowd was enlarged with a generative fill tool. There are now three separate findings: the base photo exists, the final crowd is altered, and the attendance number has no supporting source. The appropriate response is to remove the unsupported attendance claim and label the image as an illustration if it is used at all. Knowing the base photograph is real would not justify calling the finished poster documentary evidence.

If instead the image were untouched but taken several years earlier in another town, the problem would be context. Searching only for AI artifacts would miss the misleading date and place. This is why the source-and-caption check comes before pixel inspection.

Copy this media review record

• Asset URL and public uploader:
• Claim in the accompanying caption:
• Earliest matching version found, with link:
• Creation or editing information supplied:
• What the file history establishes:
• What it cannot establish:
• Evidence for the event, place and date:
• Conclusion and proposed publication label:
• Date checked and remaining questions:

Keep the record with the published asset. Do not upload private client files or images of vulnerable people to a third-party detection service without an appropriate reason and permission. A verification workflow should not create an unnecessary disclosure.

Use the companion viral-story worksheet to investigate the caption. To report an error or supply missing evidence, use Contact Us.

Sources

• C2PA: Content Credentials and provenance
• Google Search Help: learn more about an image
• Google Search Help: search with an image

Updated 8 October 2026: added an original design-production review record and a clearly fictional poster example, plus direct references. This guide does not claim to have tested detection products or verified a specific real-world image.