Blendpipe homeSend a brief
Menu

AI disclosure on video ads: what YouTube, Google, Meta and TikTok require

AI disclosure for ads, platform by platform: when YouTube, Google Ads, Meta and TikTok require a label, who applies it, and which AI edits need none.

Reader question
When does a video ad made with AI-assisted tools need an AI label, and who applies it: the advertiser, the platform or the file's own metadata?
A wooden crate of lemons at an outdoor market stall in morning light, where one lemon is a flat drawn marigold cartoon lemon and is the only one with a blank paper tag tied to it.

A video ad needs an AI label when AI has made it show something realistic that did not happen: a real person saying or doing something they did not, a real place or event altered, a lifelike scene that never occurred or, on YouTube, AI music at the center of the video. Three parties can apply that label. The advertiser does it with a setting at upload or in the ad platform, the platform does it on its own when it detects signals such as C2PA Content Credentials in the file, and some laws require a disclosure inside the ad itself. Cleanup, color, upscaling, captions and script help need no label under the current YouTube and TikTok rules, but metadata your tools wrote into the file can still bring one.

This guide puts the four platforms’ rules side by side, as checked on November 22, 2026, and then turns them into a shot-by-shot AI disclosure plan. The rules change often, so recheck each one for every campaign. The platform decides what its label rules mean, and the client’s counsel makes the final call wherever a law applies.

Who applies an AI label: the advertiser, the platform or the file

Any of the three can, and one campaign can trigger all three at once. Knowing which route applies tells you who has to act and when: before upload, at trafficking, or back in the edit.

Route 1: the advertiser discloses

Every platform covered here gives the advertiser a switch, but each switch covers a different scope:

  • YouTube: an “AI use” setting in the Attributes section of the upload flow in YouTube Studio (YouTube Help).
  • Google Ads: a checkbox in the “Altered or synthetic content” section of campaign settings, required for election ads only (Google Ads political content policy).
  • Meta: a disclosure in the advertising flow for ads about social issues, elections and politics (Meta political ads AI disclosure).
  • TikTok: a toggle in Ads Manager labeled “This ad contains AI-generated content” (TikTok Ads Help).

Route 2: the platform detects

Platforms also read signals in the file and label it themselves. YouTube says it may apply a label automatically to content made with its own generative AI tools, content that contains C2PA metadata, and content its internal systems detect as AI generated or altered. Meta relies on “industry standard indicators”, and TikTok has read C2PA Content Credentials since May 9, 2024 (TikTok Newsroom).

C2PA is the open standard behind Content Credentials: signed records, attached to a file, that state how it was made and edited. How those records survive the edit, the grade and the export is covered in our guide to C2PA Content Credentials for brand video. For labels, one fact matters most: a credential written by a generative tool can produce a label even when no rule required one.

Route 3: the ad itself carries a disclosure

Some disclosures have to sit inside the creative as a super, a caption or a spoken line. Google Ads asks for one in election ad formats where it does not generate its own. TikTok accepts “a clear disclaimer, caption, watermark, or sticker” of your own instead of its label. New York’s and California’s synthetic performer laws require a disclosure in the ad itself, whatever the platform does.

The routes do not cancel each other. A platform toggle does not satisfy a statute, and an automatic label does not count as the advertiser’s disclosure. Treat each route as a separate line in the plan.

The four platforms’ AI label rules side by side

The platforms agree on the core idea, realistic content that could mislead, but differ on scope, mechanism and whether the label can be removed. The table below compares the rules as written on each platform’s own pages.

Platform and scopeWhat triggers itHow it is appliedExemptCan it be removed?
YouTube, every upload (including the video an ad plays)Realistic content: a real person saying or doing something they did not; altered footage of a real event or place; a realistic scene that did not occur; music that is the main focus"AI use" setting in YouTube Studio; automatic labels from YouTube's tools, C2PA metadata or internal detectionNon-realistic content, minor aesthetic edits, production help such as scripts, captions and upscalingUsually, if detection erred; not when the label came from C2PA metadata, YouTube's tools or manual review
Google Ads, election ads onlySynthetic or digitally altered images, video or audio that inauthentically depict real or realistic-looking people or eventsCheckbox in "Altered or synthetic content"; Google generates the in-ad disclosure for some formats, the advertiser adds one for the restChanges inconsequential to the ad's claims: resizing, cropping, color, red-eye, background editsNot applicable: the advertiser's setting
Meta, all adsAds made or significantly edited with Meta's own generative AI tools; third-party AI detected through industry-standard signals"AI info" in the three-dot menu ("About this ad") or next to "Sponsored" for photorealistic peopleMeta's tools used without significant edits and without a photorealistic humanNot described on Meta's pages
Meta, social issue, electoral and political adsPhotorealistic image or video, or realistic audio, digitally created or altered to depict people or eventsAdvertiser discloses in the ad flow; Meta adds information to the ad and the Ad LibraryInconsequential or immaterial changes, such as resizing and color correctionNot applicable: undisclosed ads are rejected
TikTok, non-Spark adsCompletely AI-generated images, video or audio; real material significantly modified by AI"This ad contains AI-generated content" toggle, or the advertiser's own disclaimer, caption, watermark or stickerLighting, brightness and color; removing or modifying backgrounds; denoisingToggle locked after submission unless the video or image changes
TikTok, Spark Ads and organic postsAI-generated or significantly edited content showing realistic-looking scenes or peopleCreator's label on the original post; TikTok may label from C2PA Content CredentialsContent that is not realistic or not significantly editedUnlabeled content may be removed, restricted or labeled by TikTok
Rules as published on each platform's own help, policy and newsroom pages. Source: YouTube Help (answer 14328491), Google Ads political content policy, Meta newsroom and political ads AI disclosure page, TikTok Ads Help and Advertising Policies, checked November 22, 2026.

Three differences change what a production team does:

  1. Scope. YouTube’s rule covers every upload. Google’s checkbox covers election ads only, and so does Meta’s advertiser disclosure. TikTok’s ad policy covers every ad that is not a Spark Ad.
  2. Realism. YouTube and TikTok’s Community Guidelines turn on whether content looks realistic. TikTok’s ad policy names “completely AI-generated” images, video or audio without that qualifier, so a fully generated stylized shot can need a TikTok ad label and no YouTube disclosure.
  3. Permanence. A YouTube label that came from C2PA metadata cannot be adjusted by the creator. A TikTok toggle, once submitted, stays on until the creative changes. Both decisions effectively happen before upload.

The rules have also moved quickly. Every dated change in the timeline below falls between January 2024 and November 2026, and several pages were revised in 2026, which is why a plan written for a spring campaign needs rechecking before an autumn flight.

How AI label rules for ads changed, January 2024 to November 2026A vertical timeline of ten dated changes: Meta's political ad AI disclosure (January 11, 2024), TikTok reading C2PA Content Credentials (May 9, 2024), Meta renaming its label to AI info (July 1, 2024), Meta moving the label for AI-edited content into the menu (September 12, 2024), Meta's ad labels for its own AI tools (February 3, 2025), New York signing S8420-A (December 11, 2025), Meta's About this ad with third-party AI detection (June 1, 2026), New York's disclosure duty taking effect (June 9, 2026), California signing SB 1050 (September 16, 2026) and TikTok's ad policy update (November 2026). Jan 11, 2024 May 9, 2024 Jul 1, 2024 Sep 12, 2024 Feb 3, 2025 Dec 11, 2025 Jun 1, 2026 Jun 9, 2026 Sep 16, 2026 Nov 2026 Meta: political and social issue ads must disclose AI-made or altered photorealistic content TikTok starts labeling content from other platforms by reading C2PA Content Credentials Meta renames its "Made with AI" label "AI info" Meta moves "AI info" into the post menu for content only edited with AI (announced) Meta explains "AI info" labels on ads made with its own generative AI creative tools New York signs S8420-A: synthetic performers in commercial ads must be disclosed Meta adds "About this ad" and begins detecting ads made with third-party AI tools New York's disclosure duty takes effect (the 180th day after signing, by our count) California signs SB 1050: ads that prominently include a synthetic performer must disclose it TikTok's ad policy on edited media and AI content shows "last updated November 2026"
Blue marks Meta, orange TikTok, yellow state laws. Source: Meta newsroom (April 5, 2024 and February 3, 2025 posts, with their updates), Meta political ads AI disclosure page, TikTok Newsroom (May 9, 2024), TikTok Advertising Policies, New York Senate S8420-A, California SB 1050 (chaptered text), checked November 22, 2026.

YouTube: realistic content, the AI use setting and labels you cannot undo

YouTube requires creators to disclose AI-generated or altered content that seems realistic, using the “AI use” setting during upload. It also labels some content on its own, and a label that came from C2PA metadata cannot be removed by the creator.

The four triggers and YouTube’s own examples

YouTube Help lists four kinds of content that must be disclosed. It “makes a real person appear to say or do something they didn’t do”, “alters footage of a real event or place”, “generates a realistic scene that didn’t actually occur” or “creates music that’s the main focus of the video”. The help page adds that this “could include content that is fully or partially altered or created using AI tools.”

One of YouTube’s examples reads like an ad brief: “AI generated extra footage of a real place, like a video of a surfer in Maui for a promotional travel video.” Others include realistic videos of a match between two real tennis players, making it appear that someone gave advice they did not give, and a realistic tornado moving toward a real city. AI-generated music is on the list too.

What YouTube says needs no disclosure

The same page says creators do not need to disclose non-realistic content or minor edits, which it defines as edits that “are primarily aesthetic” and do not mislead the viewer “about what actually happened.” Its examples map closely onto post-production work:

  • Not realistic: someone riding a unicorn, a green-screen shot of someone floating in space, an AI-generated or altered animation of a missile in a fully animated video.
  • Minor: beauty filters, color adjustment or lighting filters, background blur or vintage effects.
  • Production assistance: AI help with an outline, script, thumbnail, title or infographic; caption creation; idea generation.
  • Repair: “Video sharpening, upscaling or repair and voice or audio repair.”
  • Backgrounds: “AI generating or extending a backdrop to simulate a moving car.”
  • Voice: “Cloning one’s own voice to create voice overs or dubs.”

The voice example is narrower than it looks. It covers a creator cloning their own voice. A brand ad that clones a performer’s voice to say new lines makes a real person appear to say something they did not, which is the first trigger.

Where the label appears

After a creator selects “Yes”, YouTube adds a label. For AI-generated music or photorealistic content, a label “in the video player may also appear.” For non-music AI audio and for non-photorealistic or animated AI content, the label may appear in the expanded description instead. YouTube states that disclosure alone “won’t limit a video’s audience or impact its eligibility to earn money.”

Automatic labels, and the ones you cannot change

YouTube may label content automatically when it was made with YouTube’s generative AI tools, contains C2PA metadata, or is detected by its internal systems. If detection is wrong, creators can usually change the answer in the disclosure survey. The exception is the one that matters for advertisers: “content made with YouTube’s AI tools, content containing C2PA metadata, or content labeled after manual review can not be adjusted.”

A separate page, “How this content was made”, says YouTube carries forward disclosures from “secure Content Credentials (C2PA) 2.1 or higher that indicate the entire video was made with AI.” The same section can show a “Captured with a camera” note for unedited footage recorded with C2PA-capable devices. In practice, the provenance data in your master can either add an AI label or support a camera-origin note, and you decide which data the master keeps.

Does this apply to YouTube ads?

A YouTube video ad plays a video that sits on a YouTube channel: Google Ads asks you to search for a video you have uploaded or paste its YouTube URL when building a video campaign (Google Ads Help). That upload goes through the same YouTube Studio flow, with the same AI use setting and the same metadata reading. YouTube’s help page does not describe how labels display when the video runs as an in-stream ad, so check a test placement before launch.

Creators who consistently choose not to disclose may get a label applied manually or face penalties, including removal of content or suspension from the YouTube Partner Program, according to the same help page.

Google Ads requires a disclosure only for election ads that contain synthetic or digitally altered content. For an ordinary brand spot on YouTube, the checkbox does not apply, but the YouTube upload rule above still does.

The Google Ads political content policy defines the scope as “images, video, or audio that inauthentically depict real or realistic-looking people or events.” Its examples are an ad “that makes it appear as if a person is saying or doing something they didn’t say or do” and one that “alters footage of a real event or generates a realistic portrayal of an event to depict scenes that did not actually take place.”

Out of scope are changes “inconsequential to the claims made in the ad”: image resizing, cropping, color or brightening corrections, defect correction such as red-eye removal, and “background edits that do not create realistic depictions of actual events.”

How the checkbox works

Advertisers must disclose every election ad with such content by selecting the checkbox in the “Altered or synthetic content” section of campaign settings. Google then generates the in-ad disclosure itself for three formats: feeds on mobile phones, Shorts on mobile phones, and in-stream on computers, mobile phones, mobile web and TV screens.

For every other format, the advertiser ticks the box and also places “a prominent disclosure” that is “clear and conspicuous.” Google’s example wording includes “Altered or synthetic content”, “This audio was computer generated”, “This image does not depict real events” and “This video content was synthetically generated.” Google says it will warn at least 7 days before suspending an account under this policy.

What counts as an “Election Ad” depends on the region, and in some regions the advertiser must be verified first. Brands rarely run election ads, but issue campaigns and public affairs work can come close, so the plan should record whether the campaign falls under the policy at all.

Meta: AI info labels, “About this ad” and political ad disclosure

Meta labels ads made or significantly edited with its own generative AI tools, has begun detecting third-party AI tools through industry-standard signals, and requires advertisers to disclose AI-altered photorealistic content in social issue, electoral and political ads. The label is called “AI info.”

Organic posts

Meta renamed its “Made with AI” label to “AI info” on July 1, 2024, after finding that labels based on industry indicators sometimes caught minor retouching (Meta newsroom). An update dated September 12, 2024 moved the label into the post’s menu for content Meta detects “was only modified or edited by AI tools”, while content detected as generated by an AI tool keeps the visible label.

Ads made with Meta’s tools

Meta’s ads transparency post (Meta newsroom, February 3, 2025, updated June 1, 2026) sets out three cases for its in-house generative AI creative features:

  • No significant edit and no photorealistic human: no AI label.
  • A significant edit: a label behind the three-dot menu or next to the “Sponsored” label.
  • An AI-generated photorealistic human: the label appears next to “Sponsored”, not behind the menu.

Ads made with other tools

The June 1, 2026 update begins rolling out “About this ad”, a destination in the three-dot menu of every ad, which will hold the “AI info” labels. Meta writes that it “will also begin automatically detecting ads created or edited using third-party AI tools through industry-standard signals,” and will apply an “AI info” label there when it detects them. It adds that the experience “may vary in certain regions due to legal requirements.”

The pages we read describe detection, not an advertiser toggle, for ordinary brand ads made with third-party tools. That puts the decision back in the file: what provenance signals the delivered master carries. Check Ads Manager for your own account, since features roll out unevenly.

Social issue, electoral and political ads

Since January 11, 2024, advertisers must disclose when an ad about social issues, elections or politics “contains a photorealistic image or video, or realistic sounding audio, that was digitally created or altered” (Meta political ads AI disclosure). That covers a real person shown saying or doing something they did not, a realistic-looking person or event that does not exist, and altered footage of a real event.

Changes that are “inconsequential or immaterial”, such as resizing and color correction, are exempt. Meta adds information to the ad and the Ad Library when the advertiser discloses. It rejects ads that should have been disclosed, and repeated failures can bring penalties. The page notes that the self-disclosure label “will soon change from ‘digitally created’ to ‘AI info’.”

TikTok: the AI-generated content toggle, Spark Ads and the ad policy

TikTok’s ad policy allows significantly edited media and AI-generated content only when it is labeled, either with TikTok’s toggle or with the advertiser’s own visible disclaimer. Undisclosed AI content gets the ad rejected or restricted.

What the ad policy covers

TikTok’s Misleading and false content policy, last updated November 2026, defines “significantly modified by AI” as content changed “beyond minor tweaks or enhancements.” Its examples:

  • Content with images, video or audio that are completely AI-generated.
  • Showing the primary subject doing something they did not do.
  • Making the primary subject say something they did not say, using AI voice cloning.

It lists as insignificant: adjustments to lighting, brightness or color saturation; “removing or modifying backgrounds”; and denoising. The policy adds that if you are uncertain, “it’s best to use caution and label it accordingly.” Its identity rules also bar using a public figure’s likeness without permission, including a false endorsement.

How the toggle works

The TikTok Ads Help page (last updated September 2025) sets three conditions for the AI-generated content disclaimer: the ad is a TikTok ad, it is not a Spark Ad, and it contains image, video or audio that is “completely AI-generated or significantly modified by AI.” The advertiser checks “This ad contains AI-generated content” during ad creation.

Three mechanics catch trafficking teams out:

  1. After you submit, the toggle cannot be changed or unselected unless you change the video or image.
  2. Any new campaign, including a duplicated one, resets the toggle, even when you reuse an ad that had it on.
  3. The feature may not be available to every account; TikTok says to contact your sales representative if it is missing.

TikTok’s overview of ad disclaimers lists AI-generated, synthetic or manipulated media under “mandatory disclaimers” and says the label shows at the bottom of the in-feed video for the duration of the ad.

Spark Ads follow the organic rules

Spark Ads run from organic posts, so TikTok sends them to its Community Guidelines instead. Those guidelines (released August 25, 2026, effective September 24, 2026) require creators “to label AI-generated or significantly edited content that shows realistic-looking scenes or people.” Unlabeled content “may be removed, restricted, or labeled” by TikTok. If a creator’s post becomes a Spark Ad, the label belongs on that original post.

TikTok also labels on its own. Since May 9, 2024 it has read C2PA Content Credentials to label AI-generated content uploaded from other platforms, and it describes itself as the first video sharing platform to do so.

Which production steps need a label: a step-by-step test

Ask two questions of every AI-assisted step: does its output appear in a delivered file, and does it change what a viewer would believe happened? If both answers are yes, plan a label on every platform. If the output is fully generated but not realistic, TikTok’s ad policy can still require one.

Does this AI-assisted step need a label?A four-question flow. If the AI output is not in any delivered file, no label is needed and the step is logged. If it changes what a viewer would believe about a real person, place or event, or shows a realistic scene that did not happen, plan a label on every platform and check state laws. If it is completely generated imagery, audio or music, label it on TikTok ads and on YouTube when it is realistic or the music is the focus. If it is only a minor edit, no label is required by the rule text, but check what the file's metadata will trigger. 1. Does the AI output appear in any delivered file (not only in previs, boards or planning)? No: no label. Log the step and its reviewer. 2. Does it show a real person, place or event differently from what happened, or a realistic scene that never occurred? Yes: label on YouTube, TikTok and Meta where required; check state laws and consent. 3. Is it completely generated imagery, audio or music, even if stylized? (not a fix to real footage) Yes: TikTok ad label; YouTube if realistic or music is the focus. 4. What is left is a minor edit: cleanup, color, upscaling, denoise, mattes, backdrop extension, captions, script help. No label in the rule text. Check what the file's metadata will trigger on upload. Yes No No
A planning aid built from the rule text, not a platform tool; the platform interprets its own rules and counsel decides where a law applies. Source: YouTube Help (answer 14328491), TikTok Advertising Policies (Misleading and false content), Meta newsroom, checked November 22, 2026.

The table below applies the same test to the AI-assisted steps that show up most in mixed-media production. “Label” means the rule text points to a disclosure; “no” means the platform’s own examples put the step outside its rule.

AI-assisted step YouTube TikTok ads Meta ads Notes
Concept frames, previs, boards No (not in the file) No No Log it; nothing reaches the viewer
Script, outline or title help No (production assistance) No No Claims in the script still need substantiation
Caption and subtitle drafts No (caption creation) No No A fluent person reviews the text
Wire, sign or power line cleanup No, if aesthetic and not misleading No (background edits are minor) Possible if the file carries AI signals Ask whether the removal changes a fact the ad relies on
Rotoscoped mattes No (adds no imagery) No Possible from signals The matte separates real footage; it shows nothing new
Upscaling, sharpening, denoise No (named as minor) No (denoising named as minor) Possible from signals Keep the source plate for reference
Backdrop extension behind real action No (YouTube’s moving-car example) No if it only modifies the background Possible from signals Becomes a label if it fakes a real place
Generated footage of a real place Label (YouTube’s travel video example) Label AI info if detected Common in travel and tourism spots
Fully generated stylized shot No if clearly not realistic Label (completely AI-generated) AI info if detected The main split between YouTube and TikTok
Synthetic narrator voice Depends: no trigger unless it imitates a real person Label (completely AI-generated audio) AI info if detected California’s SB 1050 names a voice that provides the narration
AI-generated music Label (named in YouTube’s examples) Label (completely AI-generated audio) AI info if detected YouTube’s trigger names music that is the main focus
Synthetic performer (a person who does not exist) Label Label Next to “Sponsored” if made with Meta’s tools New York and California disclosure; never as a fake customer
Digital replica or cloned voice of a real performer Label Label AI info if detected Written consent and union terms first

The “possible from signals” entries are where most surprises come from. A cleanup tool that writes an AI action into the file’s Content Credentials can lead Meta or YouTube to label a shot that no written rule required you to disclose. Whether to keep those credentials is a separate decision; the C2PA guide covers what survives export and transcoding.

Consent and copyright travel with the last three rows. A cloned voice or replica needs the performer’s written consent and, for union talent, the contract’s digital replica terms, set out in our guide to digital replicas and AI voices in ads. Generated material also changes what the brand can own, covered in what the Copyright Office protects.

Laws that sit on top of platform rules: New York, California and the FTC

Platform labels are terms of service, not law. New York and California now require a disclosure when an ad uses a synthetic performer, and federal advertising law applies to AI-made content the same way it applies to anything else.

New York’s synthetic performer disclosure

New York Senate bill S8420-A was signed on December 11, 2025 as chapter 617 of the laws of 2025. It amends section 396-b of the General Business Law. Anyone who deals in goods or services and, for a commercial purpose, creates an advertisement for them must “conspicuously disclose” a synthetic performer in it, where that person has “actual knowledge” that one is there.

The law defines a synthetic performer as a digitally created asset, made with generative AI or a software algorithm, that is intended to appear as a human performer and is not recognizable as any identifiable real person. Penalties are $1,000 for a first violation and $5,000 for each later one.

Three exemptions matter for producers:

  • Ads for expressive works such as films, television programs and video games, when the synthetic performer is used as it is in the work.
  • Audio advertisements.
  • Uses where AI “solely involves the language translation of a human performer.”

The section does not apply to the media that carry the ads, such as newspapers, streaming services and billboards, so the duty falls on the ad’s creator, not the outlet. The act takes effect on the 180th day after it became law. Counting from December 11, 2025 (20 days to December 31, plus 31 + 28 + 31 + 30 + 31 + 9 = 160 days to June 9, for 180 in all), that is June 9, 2026.

The bill text we read does not define how large or how long a “conspicuous” disclosure must be. The plan proposes the wording and placement; the client’s counsel decides whether the law reaches the campaign and whether the proposal meets it.

California’s synthetic performer disclosure (SB 1050)

California followed on September 16, 2026, when the governor signed SB 1050 (chapter 246 of 2026), adding section 17610 to the Business and Professions Code. It makes it unlawful to create and cause to be published an advertisement that “prominently includes a synthetic performer without a clear and conspicuous disclosure.”

Its definitions are more specific than New York’s:

  • Synthetic performer: “a digital figure, voice, or representation created in whole or in part using generative artificial intelligence” that gives the realistic impression of a human performer “who is not recognizable as any identifiable natural person.” A voice alone counts.
  • Prominently: in the foreground demonstrating or illustrating the product or service; providing or voicing the on- or off-camera narration; or illustrating or reacting to that narration.
  • Wording: “substantially similar to ‘this performance features a synthetic performer’ or ‘no human performer is depicted.’”

Ads for expressive works are exempt when the use matches the work, as are ads where AI “solely involves the language translation of a human performer or other accessibility features.” A violation counts as a violation of section 17500, California’s false advertising statute, and an outlet must stop running an ad only once a court order is served on it. The chaptered text sets no special effective date; under California’s usual rule for regular-session statutes, that points to January 1, 2027, which counsel should confirm.

The FTC: no AI exemption

When it announced Operation AI Comply on September 25, 2024, the FTC said its actions “make clear that there is no AI exemption from the laws on the books” (FTC press release). Two existing rules bear directly on AI-made people in ads:

  • The Endorsement Guides say ads presenting what are represented as “actual consumers” should use actual consumers “in both the audio and video, or clearly and conspicuously disclose that the persons in such advertisements are not actual consumers” (16 CFR 255.2(c)).
  • The FTC’s rule on fake reviews, announced August 14, 2024, bars reviews and testimonials that misrepresent that they are by someone who does not exist, “such as AI-generated fake reviews” (FTC announcement).

A platform label does not turn an invented customer into a real one. The simplest plan does not put synthetic people in testimonial roles at all.

Election advertising carries its own federal and state rules on top of these. If a campaign touches candidates, ballot measures or political issues, counsel reviews it before any AI decision is made.

Writing the AI disclosure plan, shot by shot

An AI disclosure plan is a table, written before production starts, that lists every AI-assisted step, the shot and files it reaches, and the label decision for each platform. It turns the rules above into instructions for the editor, the person who uploads the files and the person who builds the campaigns.

What each row records

  1. The step and the tool category (for example, “AI-assisted cleanup”), not a brand name.
  2. The shot it touches, by number and timecode.
  3. Whether the output appears in delivered files, and which ones.
  4. Whether it is realistic, and what it changes about a real person, place or event.
  5. The decision for YouTube, Google Ads, Meta and TikTok, with the rule it rests on.
  6. Whether Content Credentials will be kept on the master, and what they will say.
  7. The exact wording and placement of any in-ad disclosure.
  8. The person who reviewed the output, and who signed off the decision.

Example (fictional). Orlenna is an invented Colorado brand, and its Ridge Bottle hero spot is an illustration, not a client project. The 30-second spot at 24 fps has 8 shots and four AI-assisted steps. A generative image model made previs frames for shot 1 that never reach a delivered file: no label, logged. AI-assisted cleanup removes a power line from the sky in shot 1 (0 to 4 s), reviewed frame by frame: proposed as a minor edit, with no YouTube disclosure and no TikTok label. AI-assisted rotoscoping makes the hand matte in shot 5 (17 to 20 s), finished by a compositor: production help that adds no imagery, no label proposed. Machine translation drafts the Spanish captions, checked by a fluent reviewer: text only, no label.

The plan then flags the metadata question. If the cleanup tool writes an AI action into the shot 1 plate’s Content Credentials and those credentials survive into the masters, YouTube may label the upload automatically, and that label cannot be adjusted. The plan states this in advance, so the client chooses knowingly between keeping the credentials and accepting a possible label, or delivering without them and recording the AI step in the provenance log only.

Counting matters as much as deciding. The fictional Orlenna matrix has 3 lengths (30, 15 and 6 s) x 4 aspect ratios (16:9, 9:16, 1:1, 4:5) x 2 languages (English, Spanish) = 24 deliverables. Each one gets a row in the disclosure sheet, because a cutdown that drops shot 1 also drops the cleanup step.

Example (fictional). Suppose the TikTok buy uses only the 9:16 files: 3 lengths x 1 ratio x 2 languages = 6 ads. Suppose, too, that Orlenna’s team later asks for a variant where a generated runner replaces the filmed one in the Spanish versions. That variant reaches 3 lengths x 4 ratios x 1 language = 12 files. Each of those 12 needs a YouTube “Yes” at upload, and the 3 Spanish 9:16 files need the TikTok toggle on. If counsel decides New York’s or California’s law reaches the campaign, all 12 also need an on-screen disclosure. If the TikTok campaign is duplicated for a second flight, the 3 toggles reset and must be set again. The English 12 files are unchanged.

Wording and placing an in-ad disclosure

When a disclosure has to sit inside the creative, treat it like any other super: written, timed, translated and checked in every aspect ratio. New York’s text does not say what “conspicuously” means. California’s defines a clear and conspicuous disclosure as one that is “difficult to miss, easily understandable,” and noticed by a reasonable consumer given “the medium, format, and context.” The FTC’s Endorsement Guides use similar words for their own purposes: “difficult to miss (i.e., easily noticeable) and easily understandable by ordinary consumers” (16 CFR 255.0(f)).

The same definition says a visual disclosure should stand out “by its size, contrast, location, the length of time it appears, and other characteristics,” and that a claim made in the audio calls for a disclosure in the audio. Production decisions follow from that:

  • Placement. Keep the super inside the safe area of every ratio, and clear of the band where TikTok places its own disclaimer at the bottom of in-feed ads.
  • Duration. Propose that it stays on screen while the generated or altered material is on screen, and state that choice in the plan.
  • Language. The Spanish versions get Spanish wording, reviewed by a fluent person like the captions.
  • Wording. Short and literal works best. For synthetic performers, California’s suggested phrases set the register: “this performance features a synthetic performer” or “no human performer is depicted.”

Counsel reviews the wording and placement before the versions are made, because a change after versioning multiplies across every file that carries it.

Two rules keep the plan honest. First, it travels with the files: the editor, the trafficker and the media buyer each work from the same sheet. Second, it is rechecked when anything changes, such as a new cutdown, a new placement or a new platform policy date.

What to do before launch

Run the disclosure decision before the edit locks, not at upload. By upload, a label from metadata may already be baked into the master, and a locked TikTok toggle can mean re-exporting a file.

  1. List every AI-assisted step with its shot, tool category and reviewer, including steps that never reach a delivered file. The AI disclosure checklist walks through the same questions.
  2. Apply the two-question test to each step: is it in a delivered file, and does it change what a viewer would believe happened? Then check TikTok’s broader “completely AI-generated” wording.
  3. Check each platform’s current page and write the date you checked into the plan. The policies above were last updated between September 2025 and November 2026.
  4. Decide on Content Credentials for each master, knowing that C2PA metadata can bring a YouTube label you cannot remove.
  5. Clear consent and terms first. Any real person’s face, body or voice needs written consent, and every generative tool’s terms need checking for commercial use and training on uploads; our guide to AI video tool terms covers the second part.
  6. Write the in-ad wording for any disclosure that must sit inside the creative, and have counsel review it where a law such as New York’s or California’s may apply.
  7. Put the decisions in the trafficking sheet: YouTube “AI use” answer per upload, TikTok toggle per ad, Google checkbox per election campaign, with a note that duplicating a TikTok campaign resets the toggle.
  8. Deliver the record. The provenance log and rights sheet belong in the handover package, so the people who traffic the ads next year can see why each label was or was not applied.

Blendpipe is a production studio in Colorado Springs that combines live action, animation, 3D and AI-assisted tools, and it writes an AI disclosure plan into every production plan. How AI-assisted steps are used and reviewed is set out on AI practices and AI-assisted production. If you have a brief with AI-assisted shots in it, you can send it for a written plan.