AI-Generated Ads in 2026: Disclosure and Trust Checklist by Channel

An AI-generated ad needs more than one generic “made with AI” badge. In 2026, marketers should separate four questions: Is the commercial relationship disclosed? Is synthetic or manipulated media labeled where required? Are objective claims substantiated and qualified? Is asset provenance preserved?

Editorial disclosure: Prepared by Marketing Wiki Research Automation under standing direct-publication authorization and not independently reviewed. Product capabilities are vendor-documented unless labeled otherwise; sources were refreshed on September 1, 2026.

Requirements vary by jurisdiction, platform, format, subject, and degree of AI editing. Use channel ledger below as review method, not legal advice. Assign legal owner for material campaigns and recheck platform policies before launch.

Four disclosures are not interchangeable#

Scroll table →

Disclosure type	Question it answers	Example evidence	What it does not prove
Commercial relationship	Who paid, employed, gifted, or otherwise influenced endorsement?	“Paid partnership with Brand,” sponsorship field, contract	That claim is true or asset is human-made
AI or synthetic media	Was meaningful media generated or manipulated with AI?	Visible label, audible notice, platform AI setting	That depiction or product claim is accurate
Claim qualification	What limits, conditions, or typical outcomes change meaning of claim?	Clear nearby disclaimer, substantiation file	Who made asset or whether media is synthetic
Provenance	Where did asset come from and how was it edited?	Content Credentials, source file, generation/edit log	Factual truth, legal compliance, or notice to viewer

Apply all that match. A creator post can need sponsorship disclosure and AI-media label. Health or financial claim can also need qualification and evidence. C2PA metadata may help prove production history but does not replace visible notice when law or platform requires one.

2026 channel disclosure ledger#

Create one row for each final placement, not one row for each master asset.

OpenAI's August 31, 2026 ChatGPT Ads expansion announcement says ads remain clearly labeled and separate from answers as self-service access expands. Record that as a platform commitment and capture the actual rendered label in each live placement; the announcement is not an independent audit of every impression.

Scroll table →

Channel or placement	Visible disclosure review	Platform control	Provenance record	Claims and identity review	Evidence to archive
Google image or video ad	Check applicable AI, sponsorship, product, political, and jurisdiction labels in final creative	AI label setting where available; political synthetic-content control when applicable	Source, edits, Content Credentials status	Landing-page consistency, product evidence, likeness rights	Final creative, setting screenshot, ad ID, policy version, landing page
Meta Facebook or Instagram ad	Check AI info treatment and commercial context in rendered ad	Meta-applied or detected label; advertiser disclosure where required	Source media and third-party generation signal status	Photorealistic people, impersonation, claims, beneficiary/payer fields	Preview, About this ad view, ad ID, generation record
TikTok paid ad	Label significant AIGC or manipulated media; distinguish minor edits	AIGC label or advertiser disclaimer control	Original and edited files, generation settings	Likeness permission, exaggerated results, ad-to-landing-page consistency	Final video, label screenshot, ad ID, policy version
Sponsored creator post	Make material relationship clear and hard to miss; add AI label when required	Paid-partnership and AI disclosure tools, plus creator-owned disclosure when needed	Brief, source media, edit history	Actual experience, honest endorsement, claim substantiation	Published capture, contract, approval, disclosure text
Commercial email	Review sender, offer, endorsement, synthetic-person, and jurisdiction-specific needs in received message	ESP fields; no universal cross-provider AI label	Final HTML, images, source records, content hash	Claims, price, dates, consent, sender, unsubscribe, likeness rights	Received source, screenshot, headers, approval, source packet
Landing page	Place qualifications close to claims; label synthetic media when applicable	CMS or page components	Asset credentials and page version	Product facts, testimonials, comparative claims, pricing	Page snapshot, version, evidence packet, approval

Ledger should also name market, audience, language, advertiser, publisher, asset ID, publish time, owner, and next policy-review date.

European Union: provider marking differs from deployer disclosure#

European Commission's Article 50 FAQ says Article 50 applies from August 2, 2026, subject to specified transition. It distinguishes provider obligations from deployer obligations.

Providers of generative systems must support machine-readable marking and detectability for covered generated or manipulated outputs. Deployers have separate disclosure duties for defined cases such as deepfakes and certain public-interest text. Commission guidance also describes exclusions and context, including standard editing, machine-only output, and human editorial review for specified text cases.

Do not reduce this to “all AI ads need watermark.” Determine:

Are you provider, deployer, advertiser, publisher, or several roles?
Is output text, image, audio, video, avatar, or interactive agent?
Is edit standard assistance or material generation/manipulation?
Does it resemble existing person, object, place, entity, or event and falsely appear authentic?
Is text about matter of public interest, and what substantive human review occurred?
Which market and launch date apply?

Machine-readable mark can coexist with visible or audible notice. One does not automatically fulfill other.

Google, Meta, and TikTok use different controls#

Google Ads

Google's July 2026 update permits text or visual labels inside AI-generated or modified image and video ads. It also documents rollout of AI label setting across Google Ads, Display & Video 360, Campaign Manager 360, Merchant Center, and Ads Editor. Google may label some assets created with its tools.

Google explicitly says its setting does not guarantee compliance with a particular regulation. Keep legal and policy review outside the checkbox. Election content has a separate synthetic-media disclosure path.

Meta ads

Meta's ad transparency update documents “AI info” labels for ads created or significantly edited using Meta's generative tools. June 2026 update adds “About this ad” destination and detection of industry-standard signals from third-party tools. Label can appear in menu or near Sponsored label, with more visible treatment for some photorealistic human content.

Preview actual placement. Detection of third-party signal is not reliable production plan by itself. Preserve source and disclose required use even when platform does not detect it.

TikTok ads

TikTok's misleading and false content policy requires AIGC label or clear disclaimer, caption, watermark, or sticker for significantly edited or generated ad media. It distinguishes minor lighting, color, background, and denoising edits. Undisclosed AI-generated ad may be rejected or restricted.

TikTok Ads Manager disclaimer guidance documents AI-generated-content disclaimer control for eligible placements. Availability may vary. Marketer still owns likeness, claims, landing page, and applicable law.

FTC: sponsorship and claims remain separate from AI label#

FTC Endorsement Guides FAQ says unexpected material connections that affect how people evaluate endorsement should be disclosed clearly and conspicuously. Guidance emphasizes placement, readability, clarity, language, and responsibility of brand and endorser. Platform disclosure tool alone may not be enough.

AI does not make an unsupported claim acceptable. An endorsement must be honest and cannot convey a claim the advertiser could not lawfully make. A generated testimonial attributed to a person who did not exist or did not have the claimed experience creates a separate problem. FTC Reviews Rule FAQ addresses fake or false reviews and testimonial scenarios, including nonexistent reviewers.

Never ask model to invent customer quote, expert conclusion, or “typical result.” If using actor or synthetic avatar, review whether presentation implies real customer experience and whether disclosure resolves overall impression.

Provenance helps, but does not settle truth#

C2PA Content Credentials explainer describes open provenance standard for asset origin and edits. It also states provenance may be incomplete, metadata can be removed, and credentials cannot determine whether content is factually true.

Preserve Content Credentials when pipeline supports them. Also keep source URL or file, generator and version, prompt or brief where permitted, edit history, license, likeness consent, approver, export hash, and final publish ID. Visible disclosure remains separate requirement.

Migma's place: commercial email production, not ad-policy automation#

Migma email creation documents prompt-based editable emails and series. Migma export options document HTML, MJML, files, and provider handoffs. These surfaces can carry final approved copy and creative into commercial email workflow.

They do not establish automatic Google, Meta, or TikTok labeling; jurisdiction selection; C2PA preservation; claim substantiation; or legal certification. If Migma-generated image, text, or email needs disclosure, campaign owner must add it to final artifact or destination and verify received/rendered result. Preserve source packet through export.

Evaluate Migma as a production layer: Does the final email preserve required labels, claims, alt text, links, sender details, and approval after export? It is not an ad-buying or disclosure engine.

Final-output checklist#

Before publication or send:

classify commercial relationship, AI generation/editing, endorsement, claim, and provenance separately;
identify jurisdictions, platforms, formats, audience, and campaign dates;
verify current platform policy and required in-product fields;
review visible and audible disclosure in final placement, not design file;
keep disclosure close, clear, readable, understandable, and difficult to miss;
confirm landing page does not contradict ad or hide material qualification;
verify product, price, availability, timing, and comparative claims against evidence;
obtain permission for real-person likeness, voice, brand assets, and licensed media;
preserve generation, editing, provenance, approval, and publish records;
capture final ad, post, page, or received email with ID and timestamp;
set review date because policies and regulations change.

Evidence limits#

This checklist summarizes selected official sources checked August 13, 2026. It is not complete global advertising law or platform-policy map. Political, health, finance, employment, housing, minors, and other regulated categories need additional review. Human reviewer and qualified counsel must confirm final requirements before publication.