AI-generated content is becoming harder to distinguish from real photos, videos and audio. On Instagram, YouTube and other platforms, users can combine platform labels, visual clues, source checks and reverse searches to assess whether a post may have been created or altered using AI.
Is AI-generated content detection still possible?
Identifying AI-generated content is an informational and evergreen topic, but the technology and platform policies are changing quickly. As of September 3, 2026, platforms are increasingly using disclosures and provenance technologies alongside automated detection, but there is no single visual trick that can reliably identify every AI-generated image or video.
This distinction matters. A piece of content can be completely AI-generated, partially altered using AI or simply edited using conventional software. Those are not necessarily the same thing.
YouTube, for example, requires creators to disclose realistic content that has been meaningfully altered or generated with AI. This includes situations where a real person appears to say or do something they never said or did, footage of a real event is altered, or a realistic scene that never happened is created.
For users, the safest approach is therefore not to ask only, “Does this look like AI?” Instead, ask whether the content has a trustworthy source, whether the platform provides an AI disclosure and whether other evidence supports what the post claims.
1. Check for AI labels on Instagram and YouTube
The first check should always be the platform’s own disclosure.
Meta has been working on identifying and labelling AI-generated content across its platforms. In July 2026, Meta said it was continuing work on AI-content identification and labelling and collaborating with industry initiatives such as the Coalition for Content Provenance and Authenticity, or C2PA.
On YouTube, viewers may see information under the “How this content was made” section. A “Made with AI” disclosure can indicate that content was meaningfully altered or synthetically generated. YouTube says such information can come from creator disclosures, YouTube’s own generative AI tools or valid C2PA Content Credentials.
Therefore, before analysing someone’s face, hands or voice, expand the video’s description and look for the platform’s disclosure.
However, the absence of an AI label does not automatically prove that a video is authentic. Platform detection systems are not perfect, and not every use of AI requires disclosure.
2. Look closely at faces, hands and small details
Visual inconsistencies can still provide useful clues, particularly in AI-generated images and videos.
Look at fingers, teeth, jewellery, glasses, hair, ears and small objects in the background. AI systems have improved significantly, but generated media can still contain inconsistencies in fine details.
Pay attention to text as well. AI-generated images may contain distorted letters, strangely shaped words, incorrect shop signs or logos that look almost right but not quite.
Background objects are another useful checkpoint. A photograph may appear convincing when viewed quickly, but closer inspection can reveal repeated objects, impossible architecture, unusual shadows or objects that merge into one another.
These clues should be treated as warning signs, not proof.
A genuine photograph can contain compression artefacts, motion blur, unusual lighting or editing mistakes. Conversely, modern AI-generated media can look extremely realistic. A single strange finger or blurry background should therefore never be enough to declare a post fake.
3. Check whether lighting and shadows make sense
Lighting is another useful way to examine suspicious images.
Look at the direction of shadows cast by people, buildings, vehicles and other objects. If several objects appear to be illuminated from inconsistent directions, the image deserves closer scrutiny.
Reflections can also reveal problems. Check mirrors, windows, sunglasses and shiny surfaces. A person’s reflection should correspond logically with their position, clothing and surroundings.
The same principle applies to video. If a person moves through a scene, their lighting and shadow should generally remain consistent with the environment.
Again, these checks are indicators rather than definitive AI tests. Professional photography can involve multiple light sources, artificial lighting, reflections and extensive editing. The purpose of this step is to identify inconsistencies that justify additional verification.
4. Listen for unnatural AI voices and lip movements
AI-generated audio has become increasingly convincing, making voice verification more difficult.
When watching a suspected AI video, listen for unnatural pauses, unusual pronunciation, inconsistent breathing or a voice that seems emotionally disconnected from the speaker’s facial expression.
Lip synchronisation can also be useful. Watch whether the mouth movements correspond naturally with the spoken words. Pay attention to fast speech, words containing difficult sounds and transitions between different expressions.
One particularly important warning sign is a video showing a public figure apparently making an extraordinary statement. Instead of deciding whether the voice “sounds real”, search for the same statement through credible news organisations or the person’s verified channels.
YouTube specifically recognises realistic synthetic or altered content involving people appearing to say or do things they did not actually say or do as material that may require disclosure.
5. Verify the original source before sharing
Source verification is often more valuable than trying to detect AI from appearance alone.
Suppose a video claims that a celebrity, politician, business leader or government official made a major announcement. Do not rely solely on the account that uploaded it.
Search for the claim using specific keywords. Check the person’s official social-media account, the organisation’s website and reports from established news outlets.
Also check when the account was created and whether it has a history of posting unrelated or suspicious material. A viral clip uploaded by an anonymous account deserves more scrutiny than the same footage published by a verified organisation with an identifiable source.
This approach is especially important for breaking news. AI-generated images can be created and distributed faster than traditional fact-checking can happen, while old genuine photographs can also be reposted with a completely false caption.
The question should therefore be both “Is this media authentic?” and “Is this media being presented in the correct context?”
6. Use reverse image searches for suspicious pictures
When an image appears unusual, a reverse image search can help establish where it appeared previously.
Upload the image to a reputable reverse-search service or use an image-search feature that supports visual searching. Look for older versions, original publications and credible reporting.
This can uncover cases where a genuine photograph is being used with a false caption. It can also reveal whether an image first appeared on an unrelated website years earlier.
However, reverse search results should also be interpreted carefully. The absence of an older result does not prove that an image was generated by AI. A newly created photograph may simply have limited online distribution.
For news verification, compare the image with other photographs from the same event. Consistent weather, clothing, locations and background details can help establish whether the scene is genuine.
7. Check Content Credentials when available
One of the more important developments in AI-content verification is provenance information.
C2PA’s Content Credentials framework is designed to record information about an asset’s origin and changes in a tamper-evident format. It can help communicate whether content was created, edited or processed using particular tools.
YouTube can use valid C2PA information to provide viewers with details about how content was made. The platform says its “How this content was made” disclosures can incorporate Content Credentials showing that an entire video was made with AI.
There is an important limitation, though. Content Credentials are not a universal “real or fake” certificate. C2PA explains that the system provides information about provenance and whether the credential is valid and associated with the asset. It does not itself make a value judgement that the content is true or false.
So provenance should be considered one part of verification, not the entire process.
Do not rely on AI detectors alone
Online AI-detection tools can be useful for additional analysis, but users should be cautious about treating their results as definitive.
AI-generated media changes quickly. Detection systems can produce false positives and false negatives, while compression and editing can make analysis more difficult.
The strongest verification method combines several signals: platform disclosures, provenance information, visual and audio examination, source verification and independent reporting.
This is particularly important for journalists, students and social-media users dealing with potentially damaging claims. A detector saying “AI-generated” should lead to more investigation, not automatically become the headline.
Likewise, a detector saying “human-made” should not be interpreted as proof that the underlying claim is true.
Key Takeaways
- Check Instagram, YouTube and other platforms for their own AI or altered-content disclosures before analysing the media.
- Look for inconsistencies in faces, hands, text, shadows, reflections, audio and lip movements, but treat them only as warning signs.
- Verify the original source and search for independent coverage before sharing unusual or sensational content.
- Use Content Credentials and reverse-image searches when available, but do not treat any single detection method as conclusive.
FAQs
1. Can AI-generated images always be identified by looking at them?
No. Modern AI-generated images can be highly realistic, and obvious visual errors are becoming less common. Visual inspection can identify warning signs, but source verification and platform disclosures are also important.
2. Does an AI label on YouTube mean the entire video is fake?
Not necessarily. YouTube’s disclosure system covers realistic content that has been meaningfully altered or generated with AI. A video may contain some AI-generated or AI-altered elements without the entire video being synthetic.
3. Can Content Credentials prove that a video is true?
No. C2PA Content Credentials provide provenance information about how content was created or modified. They are designed to establish and verify provenance, but they do not independently establish whether the claims made in the content are true.
4. What should I do if an AI-generated video appears to show a real person?
Do not share it as genuine until it is verified. Check for platform disclosures, search for the original source and look for confirmation from the person, organisation or reliable news outlets. YouTube also provides mechanisms for reporting realistic synthetic content involving a person’s likeness.