A global brand receives feedback that its AI ad feels artificial. The next cut replaces the synthetic presenter, changes the voice and removes the AI notice. People prefer it. The team concludes that the notice was the problem—although its comparison cannot isolate that explanation.
For brands commissioning AI production near their Chinese supply chain, the useful question is where the next production dollar should go. Better product movement, a different performance and clearer speech are different purchases. A stronger finished ad can be commercially useful without proving which change caused the improvement.
A new TikTok guide points toward specific creative questions
On October 28, 2026, TikTok announced its Brand Safety Center and resources including AI Safety Essentials at its APAC forum. The guide separates disclosure from permission to run content and says fully AI-generated or significantly AI-modified ads require disclosure.
Its early user research in two major markets is directional, identifying issues such as unrealistic people, artificial voices, misleading product depictions and formulaic creative. That is not a universal acceptance benchmark. The guide and announcement were checked October 28. They do not establish how this brand’s audience will respond.
The production opportunity is to investigate a specific objection. This article proposes comparisons between AI samples with the same applicable disclosure requirements. It does not propose suppressing a required notice, or treat platform guidance as a complete legal assessment for every market.
Correct invalid candidates before a public comparison
If a generated connector is wrong, a performance claim is unconfirmed or asset rights are unresolved, address that issue first. Do not leave a false product depiction in a public campaign merely to see whether it attracts more attention.
Internally, retain the marked problem excerpt and the information needed to repair it. That is a diagnostic record, not approval to advertise it. Actual platform review, brand approval and market-specific requirements remain separate project work.
ONCE’s proposed sample brief separates constants from changes
This is a production method, not a platform feature or an ONCE client experiment. Start with one question and document the conditions the comparison is intended to hold steady.
| Field | What to record |
|---|---|
| Question | A specific uncertainty, such as whether the delivery makes a usage condition understandable |
| Common conditions | Product facts, core copy, duration, call to action and checked disclosure presentation |
| Change | The intended creative difference and any unavoidable accompanying changes |
| Decision task | Who reviews, what they should explain back, and which findings justify further work |
Holding disclosure steady must remain truthful to both samples. Do not mislabel live-action content to make a comparison tidy. If platform presentation differs or cannot be controlled, document that limitation. The exercise may still compare complete executions, but it cannot claim to isolate a label or a creative element.
A hypothetical voice test with a useful answer
Suppose an appliance brand has an AI ad requiring disclosure. The product action has been checked, but the voice delivery feels mechanical. Keep the picture, factual wording, disclosure and invitation unchanged, and compare two pacing treatments of the same voice approach. This is a planning example; no results are implied.
Ask reviewers to explain the product’s intended use and relevant conditions before asking what sounds natural. If one version makes the condition easier to repeat accurately, that supports further speech work. A preference for a pleasant voice alone does not establish greater trust or purchase intent.
Even a small change can affect other elements
A new pause can alter subtitle timing. A slower delivery can extend the film. Record these effects and keep the rest comparable where possible. If the revision requires substantial recutting, call it a combined voice-and-edit treatment rather than attributing the result to one pause.
Replacing a synthetic presenter can change performance, identity cues and composition together. A bundled comparison can help choose a direction. Its explanation simply needs to stay at the level of the bundle. Not every practical production test is a controlled causal experiment.
Use sample review for diagnosis and campaign testing for outcomes
An internal viewing can expose unclear references, confusing product movement or stiff delivery. A handful of preferences does not represent a target market or translate into sales. A paid campaign comparison requires its own agreed audience, allocation, observation window and primary measure, with the actual platform conditions documented.
Do not choose a favorable metric after reviewing the results and describe it as the original objective. Nor should one comment saying “AI” become evidence of universal rejection. Clarify whether the objection concerns a person, a product, a voice or the overall idea, then use it to form the next question.
Procure revisions with an explanation of what they can show
Ask the supplier for the common-condition record, identified sample changes, product-fact checks, required disclosure execution and a result note that preserves uncertainty. If several elements must change together, agree on that before reviewing the outcome.
For ONCE, the reusable asset is a record of the question, revision and limits of the conclusion. That can make the next brief more specific; actual time savings and advertising gains still need project measurement. Explore related planning in ONCE’s video marketing guides, then bring one unresolved creative question into the sample brief.
