How to label AI-generated content on social media
By William, Partner, technology · · 6 min read
Label as AI-made anything that looks real but is not: people, places or events that never existed. Use each platform’s own setting where there is one and add a plain sentence in the caption. What follows summarizes the official policies as of November 2026, which may change.
Do you have to label AI-generated content?
In practice, yes, whenever the content could be mistaken for something real, and several platforms now ask for it explicitly. A fantasy illustration of a unicorn fools nobody. A realistic video of a person saying something they never said does.
The useful rule is this: if a reasonable viewer could believe that what they are seeing really happened, label it. If it is obviously an illustration, a cartoon or a graphic element, the risk of confusion is minimal, although adding a note does you no harm.
This article is not legal advice. Rules vary by country and platforms update their policies often. The details below were checked against the official pages linked, as of November 2026.
What do the platforms ask for?
Each network has its own mechanism and criteria, but they all revolve around one idea: realistic content created or altered with AI.
| Platform | What its official policy says | How you flag it |
|---|---|---|
| YouTube | Creators must disclose realistic content that is AI-generated or meaningfully altered (for example, making a real person appear to say or do something they did not, or generating a realistic scene that never happened) | The "AI use" setting in YouTube Studio at upload |
| Instagram and Facebook (Meta) | Users are expected to disclose AI-generated content they upload; Meta also labels automatically when it detects industry-standard signals | The "AI info" label |
| Adopts the C2PA standard: files carrying signed content credentials show an icon | A C2PA icon generated from the file's metadata | |
| TikTok | Has its own label for AI-generated content | A control in the upload flow; check its help center |
| X | I could not verify its policy from an official source | Check its rules before posting |
Verified details:
- YouTube. According to its help page on altered or synthetic content, you do not need to disclose clearly unrealistic effects, production help (scripts, titles, thumbnails) or minor edits such as color filters or background blur. Disclosing does not limit a video's reach or monetization, and repeated failure to disclose can lead to penalties.
- Meta. In its explanation of how it labels AI content, Meta says it adds an "AI info" label when it detects industry-standard AI indicators or when the user discloses the content, and that it generally keeps labeled content up rather than removing it, unless it breaks other rules.
- LinkedIn. Its content credentials help page explains that image and video files signed with C2PA show an icon. Whether the icon appears depends on the tool that created the file writing that metadata.
- TikTok and X. TikTok has a labeling control, but I could not read its official terms page, so I will not give details I have not checked. The same goes for X. Look at their help centers before you publish.
What does European regulation say?
European AI transparency rules apply, according to the European Commission. Article 50 of the AI Act requires providers of generative systems to mark content so it can be detected as artificial, and requires those who publish it to label deepfakes and certain AI-generated text published to inform the public on matters of public interest. Text that has gone through human review under editorial responsibility is excluded.
That does not mean every company post needs a legal notice. It does show the direction: transparency about what is synthetic. If your case is sensitive (political advertising, health, finance, minors), talk to a lawyer in your country.
Which content should you always label?
Label everything that presents something unreal as real, and treat the notice as optional for what is obviously fictional. A practical list:
- Realistic people who do not exist, or real people in situations they never lived.
- Real places that were altered, or events that did not happen.
- Cloned or synthetic voices that could be taken for a real person.
- Products shown in a use or result that does not match reality. Here, besides the label, misleading-advertising rules apply.
- Any piece where the AI is the main thing you see or hear.
You can skip the label, without risk of misleading anyone, for what YouTube itself classes as minor: color corrections, quality upscaling, help with scripts or headlines.
How do you label it properly?
Use the platform's native option and reinforce it with a sentence in the caption, because not every viewer sees the automatic label. Three ways, from least to most explicit:
- The platform label. Turn it on at upload. It is what the network expects, and it stops the network adding one itself in a less flattering way.
- A sentence in the caption. "Image created with AI" or "Video generated with AI from a text description." Short and undramatic.
- A note on the image itself. Only when context calls for it, for example on a piece that may be reshared without its caption.
Example with a brand, "Brisa Coffee": a realistic AI-generated image of its café full of customers should say "Illustrative image created with AI, not our actual shop." If the same image is posted to advertise the real shop, it misleads. A label does not fix a message that is already deceptive.
Common mistakes
The most common mistake is labeling late, or not at all because "it's obvious." What looks obvious to you is not obvious to someone scrolling at speed.
Other frequent slips:
- Stripping or cropping metadata and assuming it then "doesn't count." Some platforms detect AI anyway and label it themselves.
- Using AI to fake testimonials, reviews or customers. That is deception, labeled or not.
- Labeling in a fearful tone ("Warning: AI!") when the content is harmless. One neutral sentence is enough.
- Assuming what was fine last year is fine today. Review the policies every few months.
Good practice for your team
Write one rule that everyone follows: what gets labeled, how, and who checks before publishing. A short checklist is enough:
- Does it look real when it is not? Label it.
- Is a real person involved? Check that they are not shown doing something they did not do.
- Is the platform's native option switched on?
- Is there a clear sentence in the caption?
- Have you checked the rights to any source material?
In igroui, generated images and videos are marked as AI-generated and the app recommends telling your audience; responsibility for what is generated and published stays with whoever publishes it. Because every piece goes through human review and approval before it goes out, that is the natural moment to apply the checklist above. You can see how it works in content approval and AI image generation.
Honesty with your audience
Labeling is not a concession: it is a habit that protects trust. People who find out on their own that a "real" image was not rarely forgive it. People who see it flagged from the start lose nothing, and you gain credibility.
To see how this fits with what AI video can and cannot do, also read AI video: what it can and cannot do.