TikTok Shop is becoming increasingly easy to fill with videos created or transformed using AI: avatars that present products, synthetic voices, generated demonstrations, automatically edited variants and affiliate content produced with almost no filming. The promise is appealing to a seller: more videos, more tests and more chances to appear alongside a product. For a buyer, the problem arises when a piece of content looks like a personal recommendation even though nobody has tested the product or can support the claims being made.
Reports in the business press and creator commerce industry show that some brands are already limiting affiliate content generated entirely with AI. This is not a universal TikTok Shop ban or an identical rule for every program. It is a brand-control decision: the company may accept AI for ideas, editing or localization, but may stop paying commission on videos in which an avatar simulates a real experience with the product.
The distinction is crucial. AI can be an efficient production tool. It becomes risky when it replaces the very evidence the consumer believes to be authentic: a person who has tried the product, can show how it works and takes responsibility for what they say. In health, beauty, education, automotive or premium products, the cost of an artificial recommendation can far exceed the saving achieved through a few rapidly created videos.
Why AI videos are multiplying in commerce
Affiliate content rewards speed. A creator can attach a product to a video and receive commission for attributed sales. When an avatar, voice and edit can be generated from a script, the barrier to producing dozens of variants falls dramatically. The same product can immediately receive a new hook, different voice, a demonstration in another setting or a version localized for a different market.
This volume is not automatically a problem. For a simple demonstration with accurate information and clear disclosure of how it was created, AI can help a small team test ideas and adapt materials. The problem begins when speed removes verification: the product is shown doing something it cannot do, an avatar claims to have used it, or the emotion of a testimonial is constructed without a genuine experience behind it.
In a very fast feed, the public may not immediately notice the difference between a filmed demonstration and a synthetic one. But comments, returns, complaints and screenshots remain. A brand is judged not only by the video it publishes directly, but also by the affiliate advertisement carrying its product, name or purchase link.
Three types of AI content that need to be separated
AI as a production assistant
This includes script ideas, audio correction, subtitles, adaptation of a video into several languages, removal of pauses or organization of edit variants. The underlying material can remain a real demonstration performed by a person who held the product in their hands. This is usually the lowest-risk scenario, provided the information remains accurate and the platform's transparency rules are followed.
AI that reconstructs or extends real material
A real shot can be supplemented with backgrounds, product images, voice-over or generated visual elements. Proportion becomes important here: is the public still seeing a faithful demonstration, or a scene that may suggest properties that do not exist? When results, comparisons or before-and-after transformations appear, the review needs to be much more rigorous.
An avatar that claims an experience
This is the sensitive case. A synthetic character can say “I tried it,” hold the product and react to its taste, effect or result despite having no experience of its own. For ordinary consumer products, the public may perceive the video as entertainment. For supplements, care products, courses, cars or expensive products, the same technique can appear to be false evidence and can damage the customer relationship.
Why some brands are limiting AI affiliate content
An affiliate program is an extension of the brand. Even if the video comes from an independent creator, the product link, commission and sometimes paid distribution connect it directly to the seller. If the public discovers that the demonstration was not real or the promise was exaggerated, it will not always separate the affiliate from the company.
Restrictions announced by some US programs generally share the same objective: to prevent videos that imitate personal reviews, use avatars to claim consumption of the product or place unverified claims in the mouth of a synthetic character. Not every brand chooses a total ban. Some may accept AI for scripts, editing, subtitles or ideas, but require real filming to present the product. This distinction is more mature than a vague “AI allowed” or “AI prohibited” rule.
Brands also have a problem of control. In an open affiliate model, content can appear faster than it can be reviewed. A good policy should therefore state what happens after a non-compliant video is identified: whether commission is stopped, the right to promote is withdrawn, a modification is requested, the content is excluded from paid distribution or an incident history is retained.
An AI label helps, but cannot preserve trust by itself
TikTok requires realistic AI content to be labelled and continues to expand identification technologies, including Content Credentials, creator-applied labels and invisible watermarks. A label gives the public important context: the material was generated or significantly modified. It does not, however, turn an inaccurate claim into an accurate one or replace evidence of a real experience.
A simple rule for brands is this: transparency answers the question “how was the video made?”, while verification answers the question “is what it claims true?” Both are necessary. A labelled avatar may be perfectly suitable for explaining an offer or animating a visual story. It is not suitable for playing a customer who achieved a medical, cosmetic or financial result that nobody validated.
Where the credibility risk is greater
Health, supplements and products promising results
Every claim about effects, symptoms, safety or transformation needs very careful verification here. An avatar that appears to consume a product or describe a physical change may be interpreted as a testimonial. Production speed does not justify unsupported promises or replace the advertising restrictions applicable to the category.
Beauty and personal care
Synthetic content can make the skin, texture, colour or result of a product look better than it does in reality. For this category, filming the real product, disclosing edits and keeping results plausible matter more than a perfectly polished feed. The buyer needs to understand what they are purchasing, not merely react to a flawless image.
Education, services and premium products
A person chooses a course, service or expensive product partly because of the expertise of the person recommending it. An avatar can explain features or support a visual presentation, but should not imitate qualifications, customer experiences or reviews it does not possess. In the premium space, authenticity is often part of the product.
Automotive and technical products
Demonstrations need to be precise. A simulation can illustrate an idea, but should not suggest performance, range, safety or compatibility that has not been tested. For technical products, real materials, measurable explanations and accurate usage scenarios are more valuable than a spectacular AI presentation.
An affiliate policy that keeps AI useful
Not every brand needs to choose between “no AI” and “anything is permitted.” A practical policy can distinguish three levels:
- Permitted: brainstorming, scripts, editing, subtitling, localization, audio cleanup and graphic elements that do not change the facts about the product.
- Permitted with approval: labelled avatars, synthetic voice-over, generated images and partially reconstructed demonstrations when the message, assets and context are reviewed before publication.
- Not permitted: a synthetic testimonial presented as real, invented results, imitation of a person without agreement, a falsely demonstrated product, unsupported health or performance claims and unauthorized use of a creator's likeness.
Operational steps then need to be established: who approves exceptions, how AI is labelled, what the creator documents, how problematic material is reported and what consequences follow a breach. For a brand working with many affiliates, a simple approval form and a library of accepted examples can reduce risk more effectively than a general two-line rule.
What to measure beyond volume
A batch of AI videos can rapidly produce impressions and even short-term sales. To understand whether it genuinely helps the brand, also monitor indicators showing the quality of the relationship with the buyer: return rate, complaints, repetitive questions, comments about authenticity, conversion rate after the first order, repeat purchases and the sentiment of conversations.
In TikTok Ads campaigns, the AI-generated variant can be tested alongside a real demonstration rather than in its place. The comparison should include not only cost per view, but also traffic quality, conversion, return costs and the effect on trust. Sometimes real content costs more to produce, but saves far greater costs after purchase.
A mature social media promotion plan uses AI to accelerate research and creative variants, but retains real people for expertise, demonstrations and the relationship with the community. The public may readily accept a new tool; what it does not easily accept is the feeling of having been persuaded through an invented experience.
Frequently asked questions
Is AI content prohibited in TikTok Shop?
TikTok has not presented a universal ban on every form of AI content. The platform requires realistic AI-generated content to be labelled and has rules against misleading material. In addition, every brand or affiliate program can impose its own terms for commissions and promotion.
Can I use an AI avatar to promote a product?
It may be possible if the platform and brand rules permit it, and the video is labelled correctly and does not mislead the public. Avoid having the avatar claim it tested the product, play the role of a real customer or make unverified promises.
Is activating the AI label enough?
No. The label provides transparency about the method of creation, but claims about the product must remain accurate, and rights in images, voices and materials need to be verified separately.
What is the safer option for a brand?
Use AI for acceleration, not to invent experiences. Retain real filming for demonstrations and testimonials, approve sensitive materials before publication and measure effects through conversion, returns and trust, not only by the number of videos.
Sources
- TikTok Newsroom Romania – updates on labels, Content Credentials, invisible watermarks and AI spam detection.
- TikTok Newsroom – the labelling rule for realistic AI content.
- Affiverse – reporting on AI videos in TikTok Shop and the review of controls in affiliate programs.
- The Wall Street Journal – context on the trend of some brands publicly distinguishing materials created without AI.
