Short answer: the AI label meaning is straightforward, a platform note saying media was made or changed with AI. On YouTube it appears as an “AI” tag under the title; TikTok shows “Creator labeled as AI-generated”; Instagram shows “AI info” or “AI content”. A label means AI was involved, but no label never proves a post is real. A label also says nothing about who owns the picture: our guide to AI image copyright explains whether AI pictures can be copyrighted. For Instagram specifically, our Instagram AI info label guide explains what trips the label and how to check a file before you post. Snapchat is the awkward case: its Ghost watermark and Context Card often do not survive outside the app, so our guide to how to tell if an image was made with Snapchat AI walks through which of its signals survive and which die on export.
AI labels are now part of everyday scrolling. A video of a realistic event that never happened, a photo of a person who does not exist, a voice clip of someone saying words they never said, each of these can now arrive with a small tag attached. The trouble is that the tag is not the same on every app, it does not always sit in the same place, and it is missing far more often than people assume. This guide explains the ai label meaning on each of the four big platforms, what each label does and does not prove, and the one check that still works when there is no label at all.
An AI label is a platform’s own disclosure marker on a post, telling viewers that the image, video or audio in it was generated or meaningfully altered with artificial intelligence. It is not a verdict that the content is fake, and it is not a stamp of authenticity on everything without one. It is simply the platform passing on what it knows (or what the creator admitted) about how that piece of media was made.
AI label meaning: what does the label actually mean?
A label answers one narrow question: was AI involved in making or changing this media in a way a viewer could mistake for real? Every major platform now draws the line in the same place, even though each words it differently. If the content is realistic (a real-looking person, place or event) and AI created it or meaningfully changed it, it should carry a label. If it is obviously not real, such as a cartoon, a fantasy scene or a clearly animated explainer, it generally does not need one. Minor help from AI, such as writing a script, adding captions, correcting colour or sharpening a photo, does not need a label either.
That “realistic” test is the key to reading every label correctly. YouTube’s own help page states the rule plainly: creators must disclose when they use AI to “meaningfully alter or generate photorealistic content”, and it lists four cases: making a real person appear to say or do something they did not say or do, altering footage of a real event or place, generating a realistic scene that did not actually occur, and creating music that is the main focus of the video. TikTok’s rule is the same idea in different words: creators must label AI-generated content that contains realistic images, audio or video. Meta’s rule, set out in its February 2024 announcement, requires people to use its disclosure tool when they post organic content with a photorealistic video or realistic-sounding audio that was digitally created or altered.
Three things can put a label on a post. The creator can disclose it themselves, using a toggle at upload time. The platform can detect it, by reading invisible markers in the file. Or the platform can apply it automatically because the content was made with its own AI tools. Which of the three happened matters, because each one can fail in a different way, and that is why a missing label proves nothing, as the section below explains.
The invisible markers deserve a short explanation, because they are behind most automatic labels. Content Credentials, the standard from the Coalition for Content Provenance and Authenticity (C2PA), is tamper-evident metadata baked into a file that records which tool made or edited it and whether AI was involved. Meta also reads IPTC metadata, a standard image field that records how the media was produced. When those markers survive to upload, a platform can label the post without the creator doing anything. When they do not survive (because the file was screenshotted, screen-recorded, re-encoded or downloaded and re-uploaded) the platform may see nothing at all. Our guide to checking Content Credentials on a photo explains how this file-level provenance works in detail.
What does the YouTube AI label say, and where do you find it?
YouTube’s disclosure is called “altered or synthetic content” in its policy and help pages, and YouTube launched it in March 2024. But that phrase is not what viewers actually see on the page. Its current help page calls the same rule “Disclosing use of GenAI content”. The setting creators answer lives in YouTube Studio, in the Attributes section of the upload flow, under “AI use”: they select Yes if the video contains realistic altered or synthetic content, or No if it does not.
We checked two labelled videos live on 10 November 2026, “The World’s Most Realistic AI Film”, from the verified channel Higgsfield AI (28 September 2026), and “AI interviews the people onboard the Titanic”, from Arabella vs Celebrity (12 June 2025), and both showed the label the same way. There was no label on the video player. Directly under the title, above the channel name, there is a small grey line showing only an info icon and the word “AI”; its full text is “AI: Content was made with AI”. The phrase “Altered or synthetic content” did not appear anywhere on either page. Opening the expanded description shows a section headed “How this was made”, which reads “Made with AI: Sounds or visuals were altered or fully generated”, with a “Learn more” link. YouTube’s help page does say a label may also appear on the player for photorealistic content, but neither video we checked showed one, so do not rely on the player. In practice the label is easy to miss: it is one small grey line, and the detail sits in a description many viewers never open.
YouTube can also label a video itself. Its help page says it may automatically apply an AI label on the video player or in the expanded description for content made with YouTube’s own AI tools, content carrying C2PA metadata, or content its systems detect as AI-generated. Creators are told in YouTube Studio when a label has been applied automatically, and in most cases they can change the disclosure, except for content made with YouTube’s AI tools, content containing C2PA metadata, and content labelled after manual review, which cannot be adjusted.
The consequences for not disclosing are real, and YouTube states them openly. Creators who consistently choose not to disclose may have a label applied manually, or face penalties including removal of the content or suspension from the YouTube Partner Programme. At the same time, YouTube says disclosing AI content will not limit a video’s audience or affect its eligibility to earn money, the label itself is not a monetisation penalty. YouTube’s own examples of what must be disclosed include a realistic tornado moving toward a real city that never happened, an AI-generated realistic video of a match between two real professional tennis players, and making it look like a real person was arrested or admitted to something they did not do. Examples it says do not need disclosure include beauty and colour filters, captions, sharpening and upscaling, gameplay footage, fully animated videos, and cloning your own voice for a voiceover. Full details are on YouTube’s help page on disclosing GenAI content.
For the full workflow for checking a viral clip, see how to check if a viral video clip is AI-generated.
What does TikTok’s “AI-generated” label mean?
TikTok’s label is the most direct of the four. When a creator switches on the AI-generated content setting before posting, the published video or photo carries a label telling viewers the creator disclosed it as AI-generated. A label tells you what a video is, but it does not change how often TikTok shows you such videos. For that, our guide to turning down AI on TikTok covers the Manage Topics slider and the fallback stack. TikTok’s help page describes two versions of the label: the creator label, which reads “Creator labeled as AI-generated”, and the auto label, which reads “AI-generated” and is applied by TikTok itself.
The auto label is where TikTok goes further than the other platforms. TikTok applies it when it identifies content as completely generated or significantly edited with AI, including content made with TikTok’s own AI effects, and uploads carrying Content Credentials. TikTok said in May 2024 that it would automatically label uploads carrying Content Credentials, and in November 2025 it said it would start adding invisible watermarks to AI content made with TikTok tools and to uploads carrying those credentials, so the marking survives even when the metadata is stripped. One detail matters for creators: once content is labelled as AI-generated with an auto label, the label cannot be removed from the post. Apple takes the same invisible approach for its own tools: from iOS 27.2, Image Playground creations and photos edited with Clean Up, Extend or Spatial Reframing carry a SynthID mark, and our guide to telling whether a photo was made or edited with Apple Intelligence explains how to check one.
TikTok’s rule covers more than fully generated clips. Its help page defines AI-generated content as images, video or audio generated or modified by AI, and says creators should label content that has been completely generated or significantly edited by AI, including synthetic images or video in which the primary subjects are portrayed doing something they did not do. Labelling is required for AI content containing realistic images, audio or video. Small edits such as colour correction or cropping do not need a label, and obviously artistic styles such as anime do not either. Misleadingly labelling unaltered content with the AI label is itself a violation of TikTok’s Terms of Service and can result in removal.
To add the label, a creator turns on the AI-generated content setting in the post screen before publishing: for a photo, after tapping the Add post button and capturing the photo; for a video, in the More options on the post screen. Content made using only TikTok’s own AI effects may not need the toggle at all, because TikTok labels that content automatically. On the TikTok website, the creator label appears as a grey pill in the right-hand info panel, below the caption and the music line, we saw exactly that on a real video checked on 10 November 2026, described in the worked test below. Note that TikTok’s website is heavily login-gated: video search would not load without signing in, so finding a labelled video took a hashtag page rather than a search. TikTok’s own help page on AI-generated content sets out the full rule and the steps.
Note the boundary of this article: the label discussed here is about the content of a TikTok post, not about finding where a TikTok photo originally came from. If you want to trace a TikTok photo or video back to its source, that is a different job. See our guide to TikTok reverse image search.
What does Instagram and Facebook’s “AI info” label mean?
Meta, which runs Instagram, Facebook and Threads, calls its label “AI info” in its own documentation. The name itself is a clue to its history. Meta originally called the label “Made with AI”, but renamed it to “AI info” in July 2024 after photographers complained that lightly retouched photos (images that had only been edited, not generated) were being tagged as AI-made. The rename was meant to make the label less of an accusation and more of a note: tap it for information about how the content was made. But the wording still varies by surface: a reel we checked on the Instagram website on 10 November 2026 showed “AI content” (a small grey button directly under the username in the post header) instead of “AI info”, so do not be surprised if the two wordings appear in different places.
A post gets the “AI info” label in two ways, and Meta has described both. The first is detection: Meta applies the label when it detects industry-standard AI image indicators in the file, principally C2PA Content Credentials and IPTC metadata. The second is self-disclosure: the person posting says the content is AI-generated, using the “Add AI label” toggle on the share screen (on Instagram, in Advanced settings if it is not visible). For photorealistic video or realistic-sounding audio that was digitally created or altered, Meta said in February 2024 that people must use this disclosure tool, and that it may apply penalties if they do not.
Since September 2024, Meta has also drawn a distinction that confuses many viewers. Content Meta detects was fully generated with AI gets the visible “AI info” label on the post. Content it detects was only modified or edited with AI tools gets the note tucked into the post’s menu instead, under Options, where almost nobody looks. So on Instagram and Facebook, a post can be AI-edited and still look unlabelled in the feed. That is not a bug; it is Meta’s stated policy, and it is one of the reasons a clean-looking feed proves nothing.
There is one more label to recognise on Meta’s apps. Content made with Meta’s own AI features carries visible markers and metadata and is labelled “Imagined with AI”, a different wording from the “AI info” label, and one that tells you the image came from Meta AI itself. To check a specific image, see our guide to how to tell if an image was made with Meta AI, which tests Meta’s Content Seal checker and explains why a blank result proves nothing. If you want to check whether a photo on Instagram is real or trace where it first appeared, our guide to Instagram reverse image search walks through the photo-first method.
Why does some AI content have no label at all?
This is the question that matters most, because it is where most people get misled. A missing label is not a clean bill of health. There are at least five ordinary reasons a piece of AI content can reach your feed with no label on it.
First, the creator may simply not have disclosed it. Every platform’s system leans heavily on self-disclosure, and a creator who does not tick the box, does not flip the toggle, or posts on a platform with no toggle at all will publish an unlabelled video. Platforms can catch some of these with automatic detection or manual review, but they do not catch all of them, and they say so themselves. When a post carries no label at all, a detector is the next check to try; our Pangram image detector guide walks through one, including how to read its heatmap. Meta has acknowledged that not every ad that qualifies for a label carries one yet, and its labelling method is still changing.
Second, the metadata may not have survived. Content Credentials and IPTC markers live inside the file, and they are fragile. A screenshot strips them. A screen recording strips them. Downloading a video and re-uploading it, or re-encoding it in an editing app, can strip them. A photo that has been passed through three group chats and two re-uploads may carry no trace of the tool that made it, even though the pixels are identical. This is the single most common reason a familiar AI image circulates unlabelled. Midjourney’s own marker is tougher than most metadata: our Midjourney Verify guide shows how its checker reads both the marker and the metadata, and why one can survive when the other is gone. FLUX sits at the other end of the scale: it has no checker and no label of its own, so our guide to telling if an image was made with FLUX explains the triage to use instead.
Third, the content may fall below the platform’s own threshold. All four platforms target realistic content, not any use of AI. A video with an AI-written script, AI-generated captions, an AI voiceover reading a list over stock footage, or a thumbnail polished with AI does not need a label on any of them. A cartoon or a fantasy scene does not either. So a post can be heavily AI-assisted and still correctly carry no label, the label was never meant to flag assistance, only realistic depiction.
Fourth, the platform may not have a label for that format. X used to be an example of that, but it added a creator-facing AI toggle in March 2026, so a post there can now carry a Made with AI label added by the poster or by X itself. Its synthetic and manipulated media policy is aimed at deceptive media that could cause harm, which it may label, restrict or remove, and in practice Community Notes often supply the context instead. Pinterest labels Pins it identifies as generated or altered with AI as “AI modified”, at the bottom-left of the Pin close-up, but that is Pinterest’s own system, not a creator toggle. A video cross-posted from a platform with a label to one without it loses the label in transit.
Fifth, the law is still catching up. The EU AI Act’s Article 50, in force since 2 August 2026, requires providers of AI systems to mark synthetic output in a machine-readable way, and requires anyone publishing a deepfake to disclose it visibly, but the provider marking duty for systems already on the market before August 2026 has a grace period running to December 2026, according to the European Commission. The publisher-side duties, and what they mean in practice, are a separate topic covered in our guide to the EU AI Act labelling rules. Until those duties bed in, viewers cannot rely on the law to have labelled everything they see.
The practical conclusion is simple: treat a label as useful information and a missing label as no information at all. A label tells you AI was involved. No label tells you only that no label was applied, not that no AI was involved, and not that the content is real.
Worked test: what real AI labels actually look like
Policies describe labels; screens show them. On 10 November 2026 we opened real labelled posts on YouTube, TikTok and Instagram without logging in anywhere, and recorded exactly what appeared, including one post that should have carried a label and did not. Facebook was not checked in this test, because its posts are login-gated; Meta’s documentation describes the same “AI info” label there.
YouTube: two labelled videos, same treatment. “The World’s Most Realistic AI Film”, from the verified channel Higgsfield AI (175,493 views, 28 September 2026), and “AI interviews the people onboard the Titanic”, from Arabella vs Celebrity (940,905 views, 12 June 2025), both showed the same small grey line directly under the title, above the channel name: an info icon and the word “AI”, reading in full “AI: Content was made with AI”. Neither video showed a label on the player, and the phrase “Altered or synthetic content” appeared nowhere on either page. Both expanded descriptions contained a section headed “How this was made” reading “Made with AI: Sounds or visuals were altered or fully generated”, with a “Learn more” link. As a control, an Associated Press video about the US midterms (4 November 2026) showed no “AI” line, no player label, and no “How this was made” section at all, its description had a Transcript section instead. The contrast is the whole lesson: on YouTube, the presence of that one small line, or of the “How this was made” section, is the signal, and it is small enough to miss if you never look below the title.
TikTok: the creator label, found via a hashtag. TikTok’s website would not run a video search without a login, so the labelled video was found through the #aigenerated hashtag page: a clip from Ineffable.ai (@ineffable.ai22) whose caption begins “Your brain thinks I’m real. That’s the scary part. I don’t exist. AI made this.” On the video page, the label appears as a grey pill in the right-hand info panel, below the caption and the music line, reading “Creator labeled as AI-generated”. There was no “AI-generated” overlay on the video player itself in the web view. Clicking the label opened TikTok’s own explainer, which confirms the two variants: the creator label appears on content a creator indicates was completely AI-generated or significantly edited with AI, while the automatic “AI-generated” label is one TikTok applies itself, and that automatic badge was not seen on TikTok web in this session. The explainer also notes that TikTok automatically labels realistic AI-made content from government agencies, politicians and political parties.
Instagram: the label’s wording is not fixed. Instagram’s home page is a login wall, but public profiles and direct post links load without signing in. A reel by Monica Muse (@monica.vogue.ai, posted 9 November 2026), a digital fashion model whose bio says “AI created based on my REAL face”, showed a small grey button directly under the username in the post header reading “AI content”, above the audio credit. That is the current wording on that post, not “AI info”. Clicking it opened a dialog that only said “Failed to load” for a logged-out visitor, so its explanation text could not be read. Then the counter-example: a post by Miquela (@lilmiquela, a virtual CGI influencer whose bio says “Robot”, 20 April 2026) showed no AI label at all, only the username and the location “Los Angeles, California”. A computer-generated character with a bio that says “Robot” carries no label, while a labelled reel sits one profile away. No label proves nothing, and here is the proof on a single screen.
What the test proves. First, the wording you see is not always the wording in the policy: YouTube’s help pages say “altered or synthetic content” while the page shows a small “AI” tag, and Meta’s documentation says “AI info” while a live reel shows “AI content”. Second, placement is everything: the YouTube tag sits under the title, the TikTok label sits in the info panel, and the Instagram label sits under the username, all small, all grey, all easy to scroll past. Third, absence is not evidence: Miquela’s unlabelled post is a CGI character. If you take one habit from this article, make it this: before you believe or share a realistic post, look for the label in the exact spot that platform puts it, and if there is none, run the reverse-search check in the next section anyway.
The one check that still works when there is no label
When a post carries no label, one check still works, because it does not depend on the creator, the platform or the file’s metadata: reverse image search. If a photo or a video frame has appeared anywhere before (on another account, on a stock site, in a news article, in an earlier post) a reverse search can find it, and the earliest copy usually tells you what the image really is.
The method takes under a minute. Save the image, or screenshot a single clear frame of the video and crop out the player controls, captions and interface. Then run it through more than one engine, because each has a different index:
- Start with Google Lens. Open Google Images, click the camera icon, and upload the photo or frame. Lens shows visually similar images and the pages that use them.
- Cross-check in TinEye or Yandex. TinEye is strong on first-appearance dates; Yandex is strong on faces and on images that Google has not indexed. If the two disagree, trust the earlier date, not the bigger number.
- Read the earliest result, not the most popular one. The page that published the image first (a photographer’s portfolio, a stock library, a news article from months ago) is usually the truth about the image. A dozen recent reposts are just copies.
- Search any visible text in quotation marks. A shop sign, a watermark, a caption or a name in the frame can lead straight to the original context that the repost stripped away.
- Check the account, not just the image. A real photo reposted by a new account with no history is still a repost. Compare the account’s age, its other posts and whether the same image appears under a different name elsewhere.
Full step-by-step instructions for every device are in our guide to how to reverse image search. And if the image itself looks AI-made (too-smooth skin, hands with the wrong number of fingers, text that dissolves into nonsense letters) our guide to how to tell if an image is AI-generated covers the visual tells that no label can replace.
Reverse search has limits, and they should be stated plainly. It cannot identify a brand-new AI image that has never been published before, because there is nothing earlier to match it against. In that case, the absence of matches is itself a weak signal (a real news photo of a real event almost always has an earlier copy somewhere) but it is not proof. Combine the search with the visual tells, the account check, and the question of whether the claim in the post is independently verifiable. No single check settles a realistic image on its own; the label, the search and the source check together are what make a judgement reliable.
Does an AI label mean the content is fake, and does labelling hurt reach? No, on both counts, and the distinction matters. A label means AI was involved in making or changing the media. It does not mean the claim in the post is false. A labelled video can be an honest illustration, a disclosed dramatisation, or a clearly marked piece of entertainment. Equally, an unlabelled video can be a complete fabrication. The label describes the production, not the truth of the content, so a label should prompt a check, not a conclusion.
As for reach, no platform has said that an AI label reduces distribution, and none of the published policies lists the label as a ranking signal. YouTube says disclosing AI content will not limit a video’s audience or affect its eligibility to earn money. The reach risk runs the other way: undisclosed realistic AI content can be removed outright, and repeat offenders face account-level penalties, and a removed video reaches nobody. TikTok’s newer “see less AI” viewer setting is a personalisation control, not a penalty: it lets individual viewers dial down how much AI content their own feed shows, which is different from the platform suppressing a labelled video.
The sensible rule for viewers is short. If you see a label, read it as “AI was involved, check the claim before you believe it or share it.” If you see no label, read it as “nothing has been disclosed, check anyway.” And if you are the one posting, the rule is shorter still: when in doubt, label it. No platform penalises over-disclosure, and a label on a borderline video costs nothing compared with a removal.
| Platform | Label wording | Where it appears | What triggers it | Can the creator remove it? | The check that still works with no label |
|---|---|---|---|---|---|
| YouTube | Small “AI” tag under the title (“AI: Content was made with AI”); “Made with AI” in the expanded description’s “How this was made” section | Directly under the title, above the channel name; details in the expanded description. No label seen on the player in our live check; “Altered or synthetic content” is the policy name, not the on-screen text | Creator answers Yes to the “AI use” question in Studio; or YouTube applies it automatically (own AI tools, C2PA metadata, detection) | Usually yes, except content made with YouTube’s AI tools, content with C2PA metadata, or content labelled after manual review | Reverse-search a video frame; check for the “AI” line under the title and the “How this was made” section |
| TikTok | “Creator labeled as AI-generated” (creator label) or “AI-generated” (auto label) | Grey pill in the right-hand info panel on the web, below the caption and music line; no overlay on the video player in the web view | Creator switches on the AI-generated content setting; or TikTok auto-labels (own AI effects, Content Credentials, invisible watermarks since Nov 2025) | Creator label can be left off at posting; an auto label cannot be removed once applied | Reverse-search a frame; check whether the same clip appears elsewhere with a label |
| “AI info” in Meta’s documentation; “AI content” observed on a live reel on the web on 10 Nov 2026 (content made with Meta AI: “Imagined with AI”) | Small button directly under the username in the post header for fully AI-generated content; tucked into the post menu under Options for content only edited with AI | Detection of C2PA or IPTC markers, or self-disclosure via the “Add AI label” toggle; disclosure required for photorealistic video or realistic audio | Self-disclosed label is the creator’s choice at posting; Meta-applied labels are not removable by the creator | Reverse-search the photo; check the post header and the post menu, not just the feed view | |
| “AI info” per Meta’s documentation (content made with Meta AI: “Imagined with AI”), not checked live in this test | Same as Instagram: on the post for generated content, in the post menu for edited content | Same as Instagram: C2PA/IPTC detection or self-disclosure; disclosure required for photorealistic video or realistic audio | Same as Instagram | Reverse-search the photo; open the post menu and check for the tucked-away note | |
| “AI modified” | Bottom-left of the Pin close-up | Pinterest detection (IPTC metadata, its own classifiers) or self-report by the content owner | Applied by Pinterest; not a creator toggle | Reverse-search the Pin image; check the Pin’s source link | |
| X | No standard AI label | , | No creator toggle; deceptive synthetic media may be labelled, restricted or removed under the synthetic media policy; Community Notes often add context | , | Reverse-search the image; read the Community Notes and replies before trusting the post |
FAQs
What does an AI label mean on a video?
It means the video was made or changed with AI. On YouTube it is a small “AI” tag under the title; on TikTok, “Creator labeled as AI-generated”; on Instagram, “AI info” or “AI content”. It describes how the media was produced, not whether the claim is true, treat it as a prompt to check, not a verdict.
Does no AI label mean a photo or video is real?
No. A label only appears when the creator discloses the AI use, the platform detects it, or the file still carries provenance metadata such as Content Credentials. Creators can skip disclosure, and screenshots, screen recordings and re-uploads strip the metadata that triggers automatic labels. A missing label means nothing was disclosed, not that nothing was generated.
Why did my own real photo get an AI label?
This happens most often on Instagram and Facebook, where Meta reads markers in the file. If any editing tool wrote C2PA or IPTC data saying AI was used (even an AI edit such as generative fill) Meta can label the post. That is why Meta renamed the label from “Made with AI” to “AI info” in July 2024.
Can a creator remove an AI label after posting?
It depends on the platform and how the label got there. On TikTok, an automatic “AI-generated” label cannot be removed once applied. On YouTube, creators can usually change their disclosure, except for content made with YouTube’s AI tools, content carrying C2PA metadata, or content labelled after manual review. On Instagram and Facebook, Meta-applied labels are not removable by the creator.







