Does This AI-Edited Ad Need a Label?

01

1. What exactly did AI change?

Classify the asset by its highest-risk edit. Do not call an entire project “AI-assisted” without recording whether AI corrected pixels, replaced a background, generated a voice, changed a face, or created the depicted event.

TikTok distinguishes insignificant edits—including adjustments to lighting, brightness, saturation, backgrounds, and denoising—from significant AI modification. Its advertising policy requires disclosure for AI-generated content and substantially AI-edited media covered by the policy. TikTok’s advertising policy should be checked again when the asset is uploaded because platform rules and publishing controls can change.

A background removal used to isolate a photographed product is generally lower risk than generating a realistic kitchen, laboratory, outdoor environment, or customer home that changes what viewers infer. Ask whether the new setting merely presents the product or appears to prove where, how, or under what conditions it was used.

1. Usually lower risk: cropping, stabilization, denoising, sharpening, exposure correction, color correction, audio cleanup, or simple background removal.
2. Review for disclosure: generated locations, extended scenes, synthetic objects, altered events, generated speech, realistic avatars, face swaps, or substantial video regeneration.
3. Escalate regardless of labeling: any edit that creates misleading evidence about the product, endorsement, result, or offer.

02

2. Does a real person appear to say or do something that did not happen?

If yes, route the ad for disclosure and consent review. This includes cloned speech, rewritten dialogue presented in the original speaker’s voice, lip-syncing to new words, face replacement, or generated actions attributed to a real person.

TikTok treats making a primary subject say or perform something they did not actually say or perform as significant modification. YouTube likewise requires creators to disclose realistic altered or synthetic content when a person appears to say or do something that did not occur. YouTube’s altered-content guidance includes examples involving face replacement and cloned voices.

Keep written permission covering the intended person, voice, script, edits, channels, markets, and usage period. Disclosure and consent solve different problems: a visible AI label does not establish permission. Google’s unacceptable-business-practices policy addresses deceptive impersonation and misleading suggestions of affiliation or endorsement. Review Google’s policy before using a recognizable person’s identity.

1. Did AI replace or materially alter a face? Label and verify likeness permission.
2. Did a real person appear to speak new words? Label and verify approval for the script and synthetic voice treatment.
3. Is the person entirely synthetic but presented as real? Check the platform’s AIGC rule and label where required.
4. Could viewers infer an endorsement that was never given? Stop publication; a label does not fix the endorsement problem.

03

3. Did AI materially change the product scene or claim?

Escalate the asset when AI changes the evidence supporting a product claim rather than merely changing its presentation.

Check whether the edited asset changes product size, color, finish, fit, ingredients, quantity, packaging, accessories, visible results, or use conditions. Compare the final creative with the actual item and landing page. TikTok prohibits exaggerated product effects, distorted before-and-after results, and material inconsistencies between the advertisement and destination.

Objective advertising claims need support before publication. The FTC’s Advertising Substantiation Policy Statement says advertisers must have a reasonable basis for express and implied objective claims before disseminating them. An AI disclosure does not turn a fabricated demonstration into substantiated evidence.

1. Compare the generated product with current photography, specifications, packaging, and the landing page.
2. Identify every express claim and every likely implied claim created by the scene.
3. Check whether AI improved the apparent result, created a before-and-after state, or simulated conditions that were never tested.
4. If the edit changes claim evidence, send the asset to the appropriate legal or compliance reviewer—not only the creative approver.

04

4. What does the destination platform require?

Check the policy and disclosure control for every publishing destination. A decision made for one platform should not automatically be copied to another.

For TikTok Ads, significantly edited realistic media and AI-generated content covered by the policy must use TikTok’s AIGC label or a clear disclosure such as a caption, watermark, sticker, or disclaimer. TikTok says advertisers should label content when uncertain and may reject or restrict undisclosed AIGC.

For YouTube uploads, disclose realistic and meaningful synthetic changes, including fabricated actions, speech, events, or places. YouTube generally distinguishes these from minor aesthetic edits and production assistance.

Google’s policy page describes AI-labeling controls and a July 2026 rollout across named Google advertising products, with additional visible treatment for certain covered locations. Because timing and campaign coverage matter, verify the current status in the account rather than assuming every AI-assisted edit receives the same treatment. Google also states that using its labeling setting does not guarantee legal compliance. See Google’s AI-label update.

1. Record the platform, placement, media type, target markets, and publishing date.
2. Read the current platform definition of generated, altered, realistic, and material content.
3. Apply the native disclosure control when available; add another clear disclosure if the policy requires one.
4. Save a screenshot or export showing the selected setting and final published treatment.

05

5. What should the approval record contain?

Create one provenance record for each materially different asset. Provenance means the documented origin and editing history of the creative—not a claim that the asset is automatically compliant.

Store the original inputs, licensed source files, tool and model, prompts, settings, generated outputs, edit description, creator or actor consent, claim evidence, approved markets, reviewer, decision date, and final label text or platform setting. If a generated scene or line changes during iteration, update the record rather than relying on the first approval.

ATIYO can organize the source asset, brief, AI-edit history, uploaded consent records, approvals, iterations, and labeling decision when the team records them. Media performance remains in the ad platform, and ATIYO preserves creative context and learnings. ATIYO does not connect to ad accounts, buy media, calculate ROAS, or independently know performance.

1. Name the unedited source and final export clearly.
2. Describe each AI change in plain language.
3. Attach permission and claim-support files.
4. Record “label,” “no label,” or “escalated,” plus the reason, reviewer, platform, market, and date.
5. Preserve the published version and evidence of the platform disclosure setting.