Realistic images, voices and video can now be generated or altered with AI tools, and platforms have started to ask creators to say when that has happened. The rules differ from one service to another, and they have changed more than once. This guide summarises what YouTube’s Help Centre says about disclosing generative AI content and how Meta has described its approach to labelling AI-generated content on Facebook, Instagram and Threads. It describes the published rules as they were read, so creators should check the current pages before relying on any detail. It is general information, not legal advice.
Why do platforms label AI content?
YouTube states the reason plainly. It says it encourages “creators’ innovative and responsible use of content editing or generation tools” but recognises that “viewers want to know if what they’re watching or listening to is real”. It therefore requires creators to disclose when they use AI to meaningfully alter or generate photorealistic content.
Meta gives a similar reason. In a February 2024 post, Nick Clegg, listed as President, Global Affairs, wrote that as the difference between human and synthetic content gets blurred, “people want to know where the boundary lies”, and that users had told the company they appreciate transparency around the technology. Our guide to how recommendation algorithms decide what you see describes related platform transparency steps in the context of ranking.
What must creators disclose on YouTube?
YouTube’s Help page lists four kinds of content that must be disclosed when AI is used to make or meaningfully alter it. Creators must disclose content that:
- makes a real person appear to say or do something they did not do;
- alters footage of a real event or place;
- generates a realistic scene that did not actually occur; or
- creates music that is the main focus of the video.
The page says this applies to content that is fully or partially altered or created using AI tools. It also says “realistic AI content and meaningful changes require disclosure, while non-realistic or minor edits don’t”.
What does not need a label on YouTube?
The page gives examples of things creators do not have to disclose. Non-realistic content made with AI is one category, for example someone riding a unicorn through a fantastical world, or green screen used to show someone floating in space. The other category is minor edits, defined as edits that are “primarily aesthetic” and do not alter the content in a way that could mislead the viewer about what actually happened. The examples listed include:
- applying beauty filters;
- colour adjustment or lighting filters;
- special effects such as background blur or vintage effects;
- production assistance such as using generative AI to create or improve an outline, script, thumbnail, title or infographic;
- caption creation;
- video sharpening, upscaling or repair, and voice or audio repair;
- idea generation;
- cloning one’s own voice to create voice-overs or dubs; and
- gameplay footage from video games.
The page adds that the list is not exhaustive.
What does YouTube say must be disclosed?
For contrast, the page lists examples of content that creators need to disclose. They include AI-generated music, AI-generated extra footage of a real place, AI-generated realistic videos of a match between two real professional tennis players and making it appear as if someone gave advice that they did not give. They also include realistic depictions of events that did not happen, such as a tornado moving toward a real city, making it appear that hospital workers turned away patients, depicting a public figure stealing something they did not steal and making it look like a real person has been arrested or imprisoned.
The pattern in these examples is realism and the potential to mislead. Fantasy and routine editing are outside the rule; realistic scenes that did not happen are inside it.
How does disclosure work on YouTube?
Creators make the disclosure in YouTube Studio during the upload process. In the Attributes section, under “AI use”, the creator selects Yes if the content meets the disclosure requirements or No if not. If the creator selects Yes, YouTube says it adds a label to the video.
Where the label appears depends on the kind of content. The page says that for photorealistic AI content a label in the video player may also appear, and that for non-photorealistic or animated AI content the label may appear in the expanded description. It also says disclosing AI content “won’t limit a video’s audience or impact its eligibility to earn money”.
Can YouTube apply a label automatically?
Yes. The page says YouTube may automatically apply an AI label on the video player or in the expanded description for:
- content made using YouTube’s own generative AI tools;
- content that contains C2PA metadata; and
- content that its internal systems detect is AI generated or altered.
A creator receives a notification in YouTube Studio if a label has been applied automatically. If YouTube’s systems make an error, the creator can in most cases change the disclosure by selecting No in the AI disclosure survey. The page says content made with YouTube’s AI tools, content containing C2PA metadata, or content labelled after manual review cannot be adjusted.
What happens if a creator does not disclose?
YouTube says creators who “consistently choose not to disclose” may be subject to manual application of a label, or to penalties from YouTube, including removal of content or suspension from the YouTube Partner Program. It adds that Community Guidelines apply to all content, including AI-generated or AI-altered content. Our explainer on how creators make money describes why Partner Program eligibility matters for platform income.
How does Meta approach AI labels?
Meta’s two posts describe an evolving approach. In February 2024, it said it would label images that users post to Facebook, Instagram and Threads when it can detect industry-standard indicators that they are AI-generated. It said it had labelled photorealistic images created with its own Meta AI feature as “Imagined with AI” since launch, using visible markers plus invisible watermarks and metadata embedded in image files.
Meta also said it was working with other companies to develop common standards, including through the Partnership on AI, and that it was building tools to identify the “AI generated” information in the C2PA and IPTC technical standards. For video and audio, it said industry signals were not yet being added at the same scale, so it was adding a feature for people to disclose when they share AI-generated video or audio. The post said Meta would require people to use that tool when they post organic content with a photorealistic video or realistic-sounding audio that was digitally created or altered, and might apply penalties if they failed to do so.
In April 2024, Meta said it would begin adding “AI info” labels to a wider range of video, audio and image content, when it detects industry-standard AI image indicators or when people disclose that they are uploading AI-generated content. The post says it would keep this content on its platforms, with labels and context, unless it violates other policies, and that a more prominent label may be used if content creates a particularly high risk of materially deceiving the public on a matter of importance. Meta said it planned to start labelling organic AI-generated content in May 2024 and to stop removing content solely under its manipulated video policy in July.
A July 2024 update to the same post explained a change to the label. Meta said that for content detected as only modified or edited by AI tools, it would move the “AI info” label to the post’s menu, while still displaying the label for content it detected was generated by an AI tool. It said the earlier label “Made with AI” had sometimes been applied to content that included only minor modifications, such as retouching tools, because those tools can add industry-standard indicators.
What are the limits of labelling?
The pages are open about limits:
- Meta says it is not yet possible to identify all AI-generated content, and that people can strip out invisible markers.
- Meta describes the approach as “the cutting edge of what’s technically possible right now” and says it is developing classifiers and other methods.
- YouTube’s list of examples is not exhaustive, so some content will fall in grey areas.
- Labels depend partly on creators choosing to disclose, which is why both platforms refer to penalties.
Labels also do not decide whether content is accurate or lawful. They are a signal, not a verdict. Advertising rules for sponsored content, covered in our guide to influencer disclosure rules, are separate from AI labels.
What should creators do in practice?
Based on the platforms’ own pages, a creator or channel team can:
- Ask whether the content is realistic and could mislead. That is the test running through YouTube’s examples.
- Separate routine editing from meaningful change. Filters, captions and script assistance are listed as minor; realistic fabricated scenes are not.
- Use each platform’s disclosure setting. On YouTube it is the “AI use” field in Attributes; Meta describes a disclosure and label tool.
- Check metadata. Content with C2PA metadata can be labelled automatically on YouTube, and Meta says it looks for industry-standard indicators.
- Keep a record of how content was made. If a label is disputed, notes on the tools used can help.
- Re-read the rules regularly. The two platforms’ wording has changed over time.
This article summarises published platform pages as read on 13 October 2026 and Meta posts dated 6 February 2024 and April 2024 (the latter updated on 1 July 2024). Platform rules change often; check the current pages. It is general information, not legal advice.
Sources: YouTube Help (Disclosing use of GenAI content); Meta (Labeling AI-Generated Images on Facebook, Instagram and Threads, 6 February 2024); Meta (Our Approach to Labeling AI-Generated Content and Manipulated Media, April 2024)
