TikTok Shop's AI Content Rules Are Not the Ads Rules: Product Videos, Creators and LIVE

Quick answer:a video that promotes a TikTok Shop product from a Seller or Creator account is governed by TikTok Shop's own AIGC policy in the Seller Center, not by the Ads Manager disclosure policy. That policy allows AI actors with disclosure but binds the product: "Product promotion videos must clearly and accurately display the physical product." AI may not alter "a product's size, color, material, quantity, bundle contents, features, compatibility, safety attributes, or performance." Disclosure goes in the post, as on-screen text or the "AI-generated content" toggle, because the Ads Manager disclaimer is only available for an ad that "Is not a Spark ad". On the US Shop a LIVE may not use "non-real-time verbal interaction such as AI-generated voices, audio recordings, or radio." Penalties run from a warning to "permanently disabling commission withdrawal privileges".

Two rulebooks, one video

The same 9:16 file can enter TikTok two ways, and each picks a different document. Uploaded into Ads Manager as a non-Spark ad, it answers to the advertising policy on misleading and false content: if TikTok identifies undisclosed AI-generated content, "your ad will be rejected or restricted." Posted from a Seller or Creator account with a product link, it answers to the Seller Center. The US page, AI-Generated Content Restrictions and Requirements, is dated 2 September 2026 and states its scope in one line: "Creators and sellers must comply with this document, our Content Policy, and our Community Guidelines."

The Ads Manager toggle does not reach across. TikTok's disclaimer help pagelists three eligibility conditions for it, and the second is "Is not a Spark ad". Of the organic post behind a Spark ad it says: "As Spark ads live organically on TikTok, they must follow the TikTok Community Guidelines policy for AI-generated content disclosure." A Shop product video is in the same position. The Spark mechanics, including what locks at authorization, are in Spark Ads with AI creative. The Ads Manager side is in TikTok's AI disclosure rules for advertisers; this post stays on the Shop side.

The product is the regulated object, not the actor

Almost every prohibition on the US page is about the product. The section "Misrepresentation in Product Promotion" reads, in full: "Product promotion videos must clearly and accurately display the physical product. AI must not be used to replace or fabricate product displays in a way that misleads users about the actual product." Its examples are specific enough to design against: do not use AI to "Turn flat (2D) products into fake 3D or rotating visuals", "Animate static products in unrealistic ways", or "Add effects like glowing, lighting, or special features that the real product does not have." The UK page, Artificial Intelligence Generated Content (AIGC), dated 28 August 2026, illustrates the rule with a toy truck whose imagery "exaggerates the truck's apparent size" against the listed dimensions.

This turns a workflow preference into a policy requirement. Generating the product from a text description is how a model invents a bottle of the wrong height, and the policy calls that "product not as described". Starting from the listing's own photograph is the compliant route, and what each model demands of that photograph is in the input photo requirements post. In our model registry, every video model we run except the two Sora models requires a starting image. Three accept more than one reference: Seedance 2.0 up to 9, Happy Horse 1.1 up to 9, and Kling 3.0 a start image plus references to 7 in total. More references of the real product means fewer invented attributes; product-in-hand UGC walks through the composite. Both Sora models leave our registry on 24 September 2026 in any case, when their Replicate listing shuts down.

What an AI actor may and may not be

The US page does not ban synthetic presenters. It lists them as the first case where disclosure matters: "The video or audio includes synthetic faces, voices, digital humans, or highly realistic virtual figures". It then draws two lines around who that digital human may appear to be.

The first is a real person: "Do not use AI to imitate or replicate another person's identity, likeness, or voice to create false endorsements or promotions." The UK page removes the label as a defence, prohibiting content "that portrays any celebrity, medical professional, political, or public figure (including their voices) which is being used to promote a product, even if content is properly labeled as AIGC".

The second is a fabricated authority. The US page says fabricating the professional identity of "experts, doctors, and scientific research institutions" to endorse a product, "especially in health-related fields like dietary supplements and functional foods, will be a key focus of governance." One example describes a format rather than a costume: "Using virtual digital humans or AI-generated 'professional' images, packaged in formats like news broadcasts or expert lectures, to endorse the efficacy of health-related products." The UK page names the roles, from "Doctors, nurses, dentists, dermatologists, pharmacists, nutritionists, psychologists, therapists, and traditional medicine practitioners" to "Hosts, presenters, interviewers/interviewees, podcast guests, or self-proclaimed experts presented as qualified medical authorities". Its example, an AI doctor in scrubs saying "I take one of these capsules every morning", "may result in removal of e-commerce permissions." Uniforms fail the same way: police, military, customs and judges are all listed. What the health rulebooks outside the Shop say about the same script is in supplement and skincare claims in AI UGC ads.

Between those lines sits the ordinary case: an invented, unnamed person who claims no credentials, holds the real product, and is labelled. That is what the policy permits.

Before-and-after and efficacy

AI product demonstrations are allowed "as long as the scene accurately reflects the product's real size, material, functions, supported surfaces, compatibility, installation requirements, and normal expected results." The next sentence is the one for anyone scripting a transformation: "Before-and-after visuals must not compress time or imply instant results unless the product can support that claim with reliable evidence." The UK page catches the visual even when the script is clean: "AI must not be used to demonstrate or imply that a product can cure, heal, eliminate, treat, prevent, eradicate, or alleviate any disease or medical condition, or health condition, even if no explicit text claim is made." On a generator a time-lapse is one word in a prompt. On the Shop it is a claim you must be able to evidence.

How disclosure works on a Shop post

Both pages give the same two routes. The first: "Clearly state that the content is AI-generated by using on-screen text, watermarks, stickers, or a note in the video description". The second is the toggle, after which "a label stating 'Disclosed by creator as AI-generated' will appear in the bottom-left corner of the content." The US page says the label alone will not cost reach: "Content will not be demoted or restricted solely because the AI-generated content setting is enabled, provided it complies with platform rules, including the TikTok Community Guidelines." That is an assertion, not a published test.

• False disclosure cuts both ways.The US page lists "Labeling content as AI-generated when AI was not used" as a violation alongside "Failing to disclose AI use when AI plays a significant role in content creation." Do not toggle it on a real creator's footage to be safe.
• The platform may label for you, and you cannot undo it."An 'AI-generated' label may be automatically applied if content contains specific technical metadata" and "Creators cannot remove a label applied by the platform." Whether your file still carries that metadata after editing is not a given; on our own export pipeline a C2PA manifest did not survive a stream copy. The page closes the loop: "If a label is not automatically applied, creators are still required to self-disclose when necessary."
• One note softens on the UK page, the rule does not.Both pages say creators "are required to clearly disclose the use of AI in their content". Where the US auto-label note says creators "are still required to self-disclose when necessary", the UK equivalent says that "if a label is not automatically applied by the platform, and the content meets the disclosure scenarios outlined above, creators are still encouraged to voluntarily disclose AI use". The UK page still calls failing to disclose "a violation which may result in removal of products from content" and opens its prohibited list with "Any AI-generated content without clear disclosure". Treat both as required.

LIVE: no AI voice, and no half-screen avatar

The rule that surprises sellers is not in the AIGC policy. It sits in Requirements for High-Quality Videos and LIVEs, dated 31 July 2026, applying to "Sellers and creators". Prohibited during a LIVE: "Do not use non-real-time verbal interaction such as AI-generated voices, audio recordings, or radio." And: "Do not use animated figures or content that covers more than 50% of the screen." Required: "Interact with viewers through real-time verbal or sign language communication." An AI avatar hosting a US Shop LIVE fails all three. The page gives no start date for the voice rule. Which synthetic voices need consent or a label elsewhere on TikTok is in AI voices in ads.

The same document asks posted videos to "feature your face and the physical product" and prohibits "still, looping, or scrolling images through the entire video without featuring your face and the physical product." The pages do not say how that sits with a disclosed digital human. What they agree on is the product: keep the real item on screen, in motion, from more than one angle.

The UK volume rule

If your testing method is twenty near-identical hook variants, the UK page has a section for you. Under "High-Volume, Low-Quality AI Content" it says "Creators must not use AI to mass-produce" either "Multiple identical or near-identical videos" or "Repetitive, low-effort promotional content". Its example is "Uploading multiple promotional videos that reuse the same script, visuals, or structure with minimal changes". The US page has no equivalent section as of 2 September 2026. For UK Shop posting, variants need to differ in script and visuals, not only in the hook.

What it costs to get wrong

The UK page adds one concrete number: enforcement "may include temporary posting limits (for example, restricting an account to 15 videos within a 30-day period), traffic restrictions, and in cases of repeated violations or significant negative impact, suspension of the account, removal of e-commerce permissions or withholding the creators' commission." An ad rejection costs you a creative. A repeated or high-impact Shop violation can reach the commission itself. What that commission is worth per render, from TikTok's own fee lines, is in TikTok Shop affiliate math.

A checklist for an AI product video on TikTok Shop

• Generate from the listing's own product photo and check the render against the listing for size, colour, material, quantity and bundle contents. Our TikTok Shop workflow starts from the product page for this reason.
• No glow, no invented rotation, no 2D-to-3D, no accessories that are not in the box.
• The presenter is nobody: no resemblance to a real person, no credentials, no uniform, no "doctor" or "expert" in the script or on screen.
• No time-compressed before-and-after without evidence, and nothing that visually implies a cure or treatment.
• Disclose in the post: burned-in text plus the AI-generated content toggle. Do not rely on an auto-label, and do not toggle it on real footage.
• Keep the physical product on screen and moving. Never route an AI voice or avatar host into a US Shop LIVE.
• If the same file also goes into Ads Manager as a non-Spark ad, set the disclaimer there too; TikTok says duplication resets it.

None of this is legal advice, and the UK page says of itself: "TikTok Shop reserves the right to enforce any content it considers to be inappropriate for any reason." Read as a system, the Shop rules are not hostile to AI video. They are hostile to a product that is not the product, a person who is not a person, and a claim that is not evidenced.

Sources

• TikTok Shop Academy (US), Policy Center, "AI-Generated Content Restrictions and Requirements" (dated 09/02/2026)
• TikTok Shop Academy (UK), Policy Center, "Artificial Intelligence Generated Content (AIGC)" (dated 28/08/2026)
• TikTok Shop Academy (US), Policy Center, "Requirements for High-Quality Videos and LIVEs" (dated 07/31/2026)
• TikTok Business Help Center, "How to add disclaimers to ads in TikTok Ads Manager" (last updated September 2025)
• TikTok Business Help Center, Advertising Policies, "Misleading and false content" (last updated April 2026)

All five were read directly on 4 September 2026. The US AIGC page carries a date two days earlier; check the dates at the top of each page before relying on a quotation here.

The Ads Manager half is in TikTok's AI disclosure rules for advertisers. If the worry is the video looking synthetic rather than being declared as such, what makes AI UGC look fake is the more useful read. UGC Vids AI generates the videos from your product photo, from $49/month, free for 3 days.