Meta AI Ad Terms: A Creative Approval Workflow for Brands

This change is first a responsibility signal, not another creative feature. Meta’s preview says the updated Self-Serve Ad Terms take effect on October 30, 2026. The preview states that advertisers assume the risks associated with AI features and resulting ad creative. Meta’s Ad Creative Generative AI Terms separately require users to evaluate an output’s accuracy, suitability, completeness and appropriateness for its intended use.

The preview also says Meta may run tests covering areas such as formatting, ad creative, relevance, pricing and reporting. Platform generation or testing does not mean the platform has approved the brand’s product facts, rights or expression. The practical gap for brands is therefore not a missing “AI used” checkbox. It is the absence of a named person who reviews the runnable ad, authorises launch and can stop it when the result leaves the approved scope.

This article is operational guidance, not legal advice. Meta may still revise wording before the effective date, so advertisers should review the final terms presented in their own account and obtain advice for the markets in which they operate. The evidence shows what Meta has changed; it does not prove that every platform or jurisdiction has adopted the same allocation of responsibility.

Why an approved master is no longer the same as an approved ad

A conventional production handoff has a familiar acceptance point. The client approves a particular video, image or copy deck, the production partner retains that file, and the media team uploads it. If the approved file is what runs, the decision can usually be traced back to the master.

With platform AI, the master may become an input. Text can vary, imagery can be modified and components can appear in different combinations. Platform tests can also affect what the audience encounters. One campaign now contains two separate approvals:

Is the creative delivered by the production partner accurate, licensed and on brand?
Are the automated features and runnable variants in the ad account still inside that approved scope?

A contract that says “the project is accepted when the client approves the final film” answers only the first question. The brand may assume the agency will review every generated option, the agency may treat account defaults as client-authorised, and the production partner may never see Ads Manager. The failure is often not deliberate negligence. It is an approval that everyone assumes happened somewhere else.

The ONCE two-baseline model

Creative baseline

The creative baseline is the brand-approved advertising unit, not merely a file name. It records the picture and copy, applicable SKU, markets, claim support, talent and music rights, permitted languages and the elements that AI must not change.

A brand might allow a background extension while prohibiting changes to the product model, ports, package contents and safety language. It might permit a headline rephrase while preventing the system from inventing a price, deadline, certification or performance figure. These permissions are more useful than a blanket instruction to “allow AI optimisation.”

Runtime baseline

The runtime baseline records what was actually enabled when the ad was authorised: platform AI features, automatic enhancements, placements and dimensions, destination, markets, budget authority, preview result and the person who can pause delivery. A production company does not need to take over media buying to make this interface explicit.

Both baselines are necessary. Without the creative baseline, the team cannot tell what the platform changed. Without the runtime baseline, it cannot tell whether the campaign left the source approval. When something goes wrong, the first diagnostic question becomes simple: did the issue exist in the approved input, or did it arise through generation, assembly or testing at runtime?

Four sign-offs, even if one person holds two roles

Claim sign-off: A product, legal or compliance owner confirms specifications, pricing, performance, conditions and disclosures.
Rights sign-off: A production or rights owner checks talent, voice, music, fonts, source material and the permitted use of AI tools.
Brand sign-off: A brand owner confirms tone, visual identity, audience and prohibited expressions.
Launch sign-off: The person with account control verifies actual settings, preview, geography, destination, budget permissions and the emergency stop path.

This does not require four long meetings. A one-page record can let each role check its own red lines against the same version number. When a platform cannot preview every possible variation, the launch owner must explicitly choose whether to disable that function, narrow the market or add early monitoring.

Four additions to the production agreement

Define the AI scope. State which third-party tools the supplier used, which source assets the platform may transform, and which elements cannot be regenerated. “Includes AI production” is too vague to allocate later changes.

Name the approval object. Is the client approving a concept, a delivery master or the final runnable unit in Ads Manager? If the production partner has no account access, launch approval belongs with the client or media agency and should be stated as such.

Cap the review volume. Do not treat unlimited platform variants as included by default. A quote can cover a defined set of priority variants, markets and one launch review, with additional review handled as another batch. The right quantity depends on SKU count, geography, claim risk and enabled features; there is no honest universal industry average.

Assign rollback. Specify who can pause an ad, expected response time, which screenshots and version records are retained, and who authorises relaunch after an incorrect claim, likeness issue, rights concern or destination mismatch.

ONCE’s article on AI ad provenance and change records answers where an asset came from and what changed. This workflow answers a different question: who approved the unit that actually ran? Provenance helps an investigation, but it is not itself an approval.

Compare suppliers by cost per safely runnable unit

As generation becomes cheaper, price per exported asset becomes a weaker purchasing metric. A more useful internal model is:

Cost per safely runnable unit = (creative production + variant review + rights records + platform preview and launch QA + monitoring and rollback maintenance) ÷ ad units that passed sign-off and can run.

This is a decision model, not an industry pricing benchmark. It exposes two forms of false economy: a large volume of generated variants that nobody checks, and a polished master whose downstream automation is treated as “the client’s problem.” The valuable supplier capability is the ability to connect production volume, review capacity and operating responsibility.

When to use the full workflow

Use the complete two-baseline, four-sign-off approach when the ad contains specific performance, price, promotion dates, certifications, health or safety statements, real-person likenesses, or multiple country versions. An incorrect generated result in these categories can create more than an aesthetic problem.

A stylised, low-risk, single-market ad with no concrete claims can use a lighter version: retain the creative baseline, require brand and launch sign-off, and inspect what actually runs on day one. The point is not to turn every small campaign into a legal project. Review effort should follow the consequences of error.

A minimum action before October 30

Choose one active Meta campaign rather than auditing the entire account. Export the approved master, list the generation and enhancement features currently enabled, name the four roles, and rehearse a case in which the platform produces an incorrect price or product detail. If the team cannot quickly find the pause owner, supporting fact and next approved version, the workflow is not yet operational.

Meta’s terms cannot design this division of labour for the brand. The more a platform can generate, assemble and test, the more important it becomes to manage “creative completed” and “ad authorised to run” as separate events. A production partner’s role is not to accept unlimited downstream liability; it is to deliver a clear creative baseline, rights record and reviewable version for the media team.