The ;tldr of this article:
Does TikTok require AI disclosure on ads? Skip the AIGC toggle and TikTok can auto-label your content as AI-generated, on its terms, not yours.
Every Compliance Blog Is Quoting the Same Stat Right Now. It's From the Wrong Year.
Ever notice how every "TikTok AI policy 2026" article you find has the exact same number in it (or if your AIGC blog content even cites it..)?
🤨
51,618 synthetic media videos removed, a "340% year-over-year increase," always framed as the freshest, scariest proof that enforcement is exploding within TikTok.
I went and pulled the actual TikTok Transparency Report that number comes from. It's real - 51,618 is a genuine figure straight from TikTok's own newsroom. But it's not from H2 2025. It's from H2 2024 📊 51,618 synthetic media videos removed - TikTok's Fifth Transparency Report, H2 2024. A year-old figure, being recycled by article after article as if it dropped last week. And the "340%" bit? I couldn't find it anywhere in TikTok's actual reporting - it doesn't appear in the source these articles cite it to.
Hmm. So - real talk, that number's been laundered through enough blogs that nobody's checking it against the primary source anymore. Including, until I checked this week, after reviewing what my precious LLM model sourced for me too..
Here's the part that should actually worry you more: TikTok's most recent Transparency Report (their sixth, covering H1 2025) tells a completely different story. Policy-violating AIGC removals fell 53%, down to under 25,000. But creator-labeled AI videos rose 36% to over 8.7 million, and content TikTok's own systems auto-labeled as AI-generated rose 81% to roughly 5.5 million
📊 Removals down 53%, auto-labeled AIGC up 81% to ~5.5M - TikTok's Sixth Transparency Report, H1 2025.
Read that combination again.
Removals dropping while detection and labeling volume is climbing hard isn't TikTok going soft on AI content - it's TikTok's automod getting good enough that undisclosed content gets caught and labeled before it needs a takedown. That's not a less scary trend for advertisers. That's a scarier one. The old "outrun the enforcement wave" mental model doesn't apply anymore, because there's less of a wave to slip past - there's a filter (and TikTok’s automod) that's just quietly gotten a lot better at its job.
What the Automod Actually Flags (It's Not About What You Meant)
This is the mistake I see constantly, and it's the one that costs advertisers the most: treating "AI-generated" as a question about your intent. It isn't. It's a detection-threshold question.
You didn't set out to make "synthetic media." You upscaled some old b-roll, cleaned up a product shot, ran a voiceover through a text-to-speech tool because your presenter had a cold. None of that feels like deepfake territory to you. TikTok's automod doesn't care what it feels like to you. It only cares whether a viewer could reasonably believe they're watching something real when it is not. It is also taught to be extra vigilant.
TikTok's own ad policy spells out what actually triggers disclosure.
📊 Official AIGC disclosure triggers - TikTok Ads Policy: Misleading and False Content
📌 Four things advertisers routinely don't count, and should actually are :
- AI voiceover / text-to-speech narration - even a "boring," non-cloned AI voice reading your script counts.
- AI b-roll where a person or product does something that didn't happen on camera.
- AI-retouched product imagery - once the retouching goes past colour and lighting into materially changing what's shown.
- AI-upscaled legacy footage - especially if it's paired with an AI-cleaned or reconstructed voice track.
None of these scream "deepfake" in your head while you're editing tbh.
All four can trip the same disclosure requirement as one. (This is exactly the gap I spent years watching advertisers fall into during the review work that trained the systems catching it now 😉)
Auto-Labeling Overrides Your Choice and It's Not Theoretical
TikTok isn't only relying on advertisers to self-report anymore.
TikTok's own newsroom confirmed it's testing an "AI-generated" label it plans to eventually apply automatically to content it detects was made or edited with AI.
What that means in plain terms 👆
If you don't disclose and the automod catches it, TikTok doesn't just reject the ad and leave it there. The platform can slap its own label on your content, on its terms, not yours. You lose the ability to control how the disclosure reads, and depending on how obviously the content was AI-made, you're now the advertiser whose ad got publicly auto-flagged instead of the one who disclosed cleanly upfront.
That's a worse outcome than a straightforward rejection, and it's avoidable with one toggle.
The Public-Figure Line Is a Hard Reject, No Exceptions
One category where TikTok doesn't leave room for a disclosure workaround: using a real person's likeness, audio or visual, without their permission - especially a public figure (and a current political figure is just a no-no), especially in a misleading context.
📊 Public figure likeness misuse prohibited - TikTok Ads Policy: Misleading and False Content.
An AI voice that sounds like a recognizable celebrity endorsing your product, even labeled, even as a joke? Rejected 🙅.
A fabricated "interview" clip putting words in a real person's mouth? Rejected 🙅.
This isn't a gray area you disclose your way out of - it's prohibited outright, and private individuals' likenesses get pulled on discovery too. So don’t get flagged!
How to Disclose Without Tanking Your CTR
So - you actually have options here, and they don't all cost you the same amount of conversion.
The AIGC toggle ("This ad contains AI-generated content" in Ads Manager) is the cleanest technical route.
TikTok auto-generates the disclaimer badge for you - no editing your creative required 📊 AIGC toggle mechanics - TikTok Ads Help: Add Disclaimers to Ads. One catch worth flagging in bold: once you submit with it on, you can't unselect it unless you change the video or image - and it resets every time you duplicate a campaign, so you have to re-enable it on every duplicate that still contains the AI content. I've seen this trip up teams running the same creative across a dozen ad groups more than once - make sure this becomes second nature and not something you have to fix once its done.
In-creative disclosure - a watermark, caption, or sticker - is the alternative TikTok explicitly accepts. It needs to be visible (not buried for half a second in a corner), and the language has to actually communicate AI generation, not soften it into vague phrasing like "digitally enhanced."
I can't promise you which option protects CTR better for your specific creative - that genuinely depends on your audience and format, and anyone who tells you it's always one or the other is guessing. But I can tell you the toggle is the lower-effort, lower-risk default and it's the one to reach for first before you hit publish.
Make this an important SOP for your team if necessary!
The Pre-Upload AIGC Checklist
Before your next AI-touched ad goes live, run it against this quick checklist (go ahead and save it if you want!)
◻ Does the video, audio, or image involve full AI generation, or a subject doing/saying something they didn't?
◻ Have you enabled the AIGC toggle, or added a visible in-creative disclosure?
◻ If duplicating a campaign, did you re-enable the toggle (it resets)?
◻ Does any voice or face resemble a real, identifiable person, public figure or not?
◻ If yes to the above - do you have documented, explicit consent?
◻ Would a viewer reasonably believe this is real, unaltered footage if you didn't tell them otherwise?
Frequently Asked Questions
Does using an AI voice for narration count as AI-generated content, even without cloning anyone?
Yes. TikTok's disclosure requirement isn't limited to voice cloning - a generic text-to-speech narration still counts as AI-generated audio and needs disclosure.
What actually happens if I don't disclose?
Per TikTok's own policy language, an undisclosed ad "will be rejected or restricted." Separately, TikTok has confirmed it's testing automatic AI-content labeling for detected, undisclosed content - meaning the platform may label it for you if you don't.
Can I use an AI avatar as a spokesperson?
Generally yes... And make sure it is disclosed! The avatar absolutely cannot claim it has personal experience when in reality.. it doesn’t? For example making the claim "Oh I've been using this for years". Plus it must not resemble a real identifiable person without consent.
Does colour grading or background cleanup and editing need some form of disclosure?
Nope! TikTok's policy exempts minor technical edits like lighting, colour correction and background removal. A slight contrast to that image to highlight those product features? GO CRAZY!
Popular post
