Algorithm updates

When a TikTok clip needs the AI-generated content label

TikTok requires a label on AI or heavily edited content that shows realistic people or scenes. Generic text-to-speech and cropping are exempt. Rules for clips.

Updated October 4, 20269 min read8 sources

In short

It depends on realism. Under TikTok's Community Guidelines in force since September 24, 2026, you must label AI-generated or significantly edited content that shows realistic-looking people or scenes, for example AI audio that mimics a real person's voice. Generic text-to-speech that is not a recognizable voice, color correction, reframing and cropping need no label. TikTok also labels content itself: it has read C2PA Content Credentials since May 9, 2024, and by July 10, 2026 it had labeled over 3 billion videos as AI-generated.

On this page

The rule: realistic AI or heavy edits need a label
Clip edits that need a label, and edits that do not
A normal clip edit can count as "significantly edited content"
What TikTok does not allow, with or without a label
TikTok can label your clip without you: Content Credentials and watermarks
Does an AI label reduce reach?
What to do now, step by step
Questions
Sources

The rule: realistic AI or heavy edits need a label

TikTok's Community Guidelines (version "2026 August", released August 25, 2026, effective September 24, 2026) say TikTok requires creators to label AI-generated or significantly edited content that shows "realistic-looking scenes or people." Unlabeled content "may be removed, restricted, or labeled by our team," depending on the harm it could cause.

You can disclose in two ways. You can add your own clear caption, sticker or watermark. For AI-generated content, you can also use TikTok's AIGC label. AIGC means AI-generated content.

Two definitions in the guidelines decide most cases for clippers:

AI-generated content (AIGC): "Any image, video, or audio made or changed by AI." This includes realistic scenes and artistic styles such as anime, cartoons or paintings.
Significantly edited content: media that makes it seem like someone did or said something they didn't, or changes their appearance until they are unrecognizable. TikTok lists three examples: cropping or cutting phrases to change meaning, rearranging or combining clips, and changing speed or adding or removing audio or video parts.

The second definition matters even if you never touch an AI tool. It describes ordinary editing used to change what a person appears to say.

Clip edits that need a label, and edits that do not

The table applies TikTok's "Disclosure is needed" and "Disclosure isn't needed" lists to edits that clippers often make. Rows marked "our reading" are our application of the text; TikTok has not ruled on those cases by name.

Edit in your clip	Label needed?	Basis in TikTok's text
Reframing a landscape podcast into 9:16, cropping, color correction	No	Listed as small edits that need no disclosure
Generic text-to-speech narration	No	Exempt when the TTS "isn't a recognizable voice of a known individual"
AI voice that sounds like the podcast host or a celebrity	Yes	"AI-generated audio mimics the voice of a real person"
AI lip-sync that makes a guest say new words	Yes	AI makes it look like someone said something they didn't
AI dubbing of the guest into another language in the guest's own voice	Yes (our reading)	The audio mimics a real person and puts words in their voice that they did not say
Face swap, or adding or removing a person in a misleading way	Yes	Both are listed cases
Realistic AI B-roll, such as a fake crowd or street scene	Yes	AI content that shows realistic-looking scenes
Anime or cartoon-style AI B-roll	No	Artistic styles are listed as exempt
Auto-generated captions, AI moment selection, AI reframing (including TikTok's own Smart Split)	No (our reading)	These tools do not depict anything that did not happen
Cutting two answers together so a guest seems to say one new sentence	Yes, if allowed at all	"Rearranging or combining clips" to change meaning is significant editing

For TikTok's own AI clipping tool and how it compares with third-party tools, see Smart Split.

A normal clip edit can count as "significantly edited content"

Most clips are cuts. You remove the setup, the tangent and the dead air. That is not significant editing, because the speaker still says what they said. TikTok's definition turns on meaning: media that makes it seem like someone "did or said something they didn't."

The line is crossed when your cut changes the meaning. Two hypothetical examples:

Allowed with a label, our reading: a guest says "Everyone thinks I hate pineapple on pizza. I love it." You cut the clip to "I hate pineapple on pizza" for a hook. The clip now says the opposite of what the guest meant. The change is harmless but confusing, and TikTok's rule for that case is to disclose it. Better still, do not make the cut.
Not allowed even with a label: the same splice makes a public figure appear to support a product or take a political stance. TikTok does not allow AIGC or significantly edited content showing a public figure "supporting products" or taking political stances they never took. A label does not fix it.

The safe practice is simple: keep each sentence's meaning intact. When you join two moments, make the jump visible with a cut, a caption or a title card, so the viewer does not hear one invented sentence.

What TikTok does not allow, with or without a label

The guidelines say that "Even with labels," some AI or edited content is still harmful. These are listed as not allowed:

Using the likeness of private figures without consent. Private figures are everyone under 18 and adults who are not public figures. Podcast callers and stream chatters are often private figures.
Sexualized, fetishized or victimizing depictions, and AI-created likenesses made to bully or harass.
AIGC or significantly edited content that misleads about a matter of public importance, such as content made to look like it comes from a real news source, a fake crisis event, or a public figure linked to crimes or comments they never made.
Any content that breaks other rules, such as impersonation, misinformation or hate speech, "even if it's AI-generated".

This wording dates from the guidelines that took effect on September 13, 2025. TechCrunch reported at the time that TikTok replaced narrower rules on specific deepfakes with broader language about content "misleading about matters of public importance or harmful to individuals."

One further rule limits reach rather than removing content. Realistic content that is not yet confirmed to be AI or edited, but presents matters of public importance in a way that could mislead, is ineligible for the For You feed (FYF), TikTok's main recommendation feed. Clips of news events with unclear origin can fall into this zone.

TikTok can label your clip without you: Content Credentials and watermarks

TikTok does not rely only on creators. It added automatic labeling in two steps.

May 9, 2024: Content Credentials. TikTok began reading C2PA Content Credentials, which is metadata that some tools attach to a file to record how it was made. TikTok said it could use that metadata to "instantly recognize and label AIGC" made on some other platforms. It also said it labels AIGC made with TikTok AI effects, and that it would attach Content Credentials to TikTok content, which "will remain on content when downloaded."

November 19, 2025: invisible watermarks. TikTok said it would add invisible watermarks to AI content made with TikTok tools such as AI Editor Pro, and to content uploaded with C2PA Content Credentials. It said the watermarks "will help us label content more reliably."

The scale has grown fast. TikTok reported over 1.3 billion labeled videos on November 19, 2025, and "over 3 billion" on July 10, 2026.

Two practical results follow. This is our inference; TikTok has not described these cases. First, if your AI B-roll comes from a tool that writes Content Credentials, TikTok may label the clip even if you forget. Second, if you download a TikTok video made with TikTok AI tools and reuse it, the credentials may travel with the file. Reposting other people's TikToks also fails TikTok's originality rules, so avoid it for both reasons.

Does an AI label reduce reach?

TikTok has not said that a label lowers ranking. In the 2026 guidelines, the AI section lists only one kind of FYF-ineligible content: unconfirmed realistic content about matters of public importance. Labeled AI content is not on that list. Fastlane, a third-party blog that compared the two versions, reported that the AI section is "identical in the 2025 and 2026 versions."

Reach can still fall for three other reasons:

Viewer choice. TikTok's AI-generated content control lets viewers choose how much AIGC they see. TikTok called it a test through July 10, 2026, and "a new feature" on October 1, 2026. Our reading: viewers who turn AI content down will see labeled clips less often. See viewer controls.
AI spam detection. On July 10, 2026, TikTok said it would test better detection of "accounts dedicated to posting AI-generated spam" on politics and current events, financial advice and medical content. It said AI can be misused to mass-produce spam that "crowds out authentic creators." TikTok did not say what happens to accounts it detects.
Originality. Creator Rewards treats content with "minimal original input or edits" as unoriginal. An AI voice reading someone else's clip does not add original input by itself. See the low-quality and originality list.

Our reading: the label costs a little reach with viewers who chose less AI content. Skipping a required label risks removal or restriction. Label when the rule requires it.

What to do now, step by step

List the AI in each clip. Note every AI tool you used: voice, dubbing, lip-sync, B-roll, upscaling, captions.
Apply the realism test. Label when AI or editing shows realistic people or scenes, or when audio mimics a real person's voice. Do not label for captions, cropping, reframing, color or generic text-to-speech.
Use the AIGC label or a clear on-screen note. TikTok accepts its own label, or your own clear caption, sticker or watermark.
Protect meaning in every cut. Never join two sentences into one new claim. Never make a public figure appear to endorse a product or a political position.
Get consent before you change private people. Do not put AI voices or faces on callers, chatters or guests who are not public figures. Rights to the footage are a separate question; see copyright and fair use.
Keep AI voice accounts out of high-risk niches at volume. Politics, finance and medical clip accounts that post AI narration at scale fit the profile TikTok said it is testing detection for.
Check the campaign rules. Some campaigns ban AI voices or AI footage. A campaign rule does not override TikTok's label rule, and TikTok's rule does not override the campaign's. You must meet both.

Questions

Do AI-generated captions or subtitles need the AI label?

TikTok's guidelines do not list captions as a disclosure case. The rule targets realistic depictions of people or scenes. Accurate captions of what the speaker said do not depict anything new, so in our reading they need no label.

I translated a podcast clip with an AI voice that sounds like the guest. Do I label it?

Yes, in our reading. TikTok lists AI audio that "mimics the voice of a real person" as a disclosure case, and the dub puts words in that voice that the guest never spoke. A generic narrator voice that does not imitate anyone would be exempt.

Does TikTok's own AI tool content get labeled automatically?

TikTok says it labels AIGC made with TikTok AI effects, and adds invisible watermarks to content made with tools such as AI Editor Pro. You should still check the label before you post.

Is labeled AI content still eligible for the For You feed?

TikTok's 2026 AI rules list only unconfirmed realistic content about public matters as FYF-ineligible. Labeled AI content is not on that list. Viewers can still choose to see less AI content.

What happens if I do not label realistic AI content?

TikTok says unlabeled content may be removed, restricted, or labeled by its team, depending on the harm it could cause. If the content misleads about a matter of public importance, it is not allowed even with a label.

Do anime or cartoon AI visuals need a label?

No. TikTok lists artistic styles such as anime as exempt from disclosure. They still count as AIGC under TikTok's definition. Our reading: if such a clip carries a label, viewers who chose less AI content may see it less often.

Sources

We checked every claim on this page against these sources on the date above. Platform rules and rates change often. Found an error? Tell us.