AI-assisted production can turn one brief into many advertising assets. That speed also multiplies the number of factual claims, images, voices, likenesses, disclosures, destinations, and platform combinations a team must review. A release gate keeps creative volume separate from approval quality.
This checklist is an operating control, not legal advice and not a guarantee that a platform or regulator will accept an ad. Map it to the jurisdictions, contracts, products, and platform policies that actually govern the campaign.
1. Register the asset and the decision
Give every variant an ID, campaign, audience, destination, owner, model or tool used, creation date, and planned launch. Record whether AI generated, transformed, selected, translated, or merely assisted the asset. Decide who may approve, reject, revise, or pause it.
Write the risk appetite for this campaign. A regulated claim, a child-directed audience, a public figure likeness, or an automated optimisation path may need a stricter gate than a simple layout variation.
2. Lock the brief and the destination
Confirm objective, audience, offer, price or qualification condition, required disclosure, and landing-page promise. Compare the ad copy and visual with the destination on mobile and desktop. Check that the form, consent, routing, and response owner work before approval.
Do not approve a creative merely because its text is attractive. A destination that cannot deliver the promise turns a production success into a trust and compliance problem.
3. Verify every material claim
List claims in the asset: performance, price, availability, time, comparison, health, security, customer result, or implied guarantee. Attach the source, date, owner, scope, and limitation for each. Mark an illustrative example as illustrative; never let generated copy turn it into a customer result.
Keep a “not proven” state. If a claim cannot be verified for the intended audience and period, remove it or hold the asset. A human reviewer must read the actual rendered creative, not only the prompt or text layer.
4. Check rights and provenance
Record the source and permission for images, music, fonts, logos, testimonials, data, voice, and likeness. The IAB AI intellectual-property and transactions playbook provides an advertising-industry framework for discussing rights and transactions; it does not replace the campaign’s contract or jurisdictional review.
Preserve tool terms and license evidence at the time of approval. If a model or stock provider has restrictions on commercial use, training, attribution, or modification, attach the relevant record. Do not treat “generated” as synonymous with “unrestricted.”
5. Review identity and realistic representation
Flag a real or recognisable person, customer, employee, professional, location, event, or product condition. Verify consent, contractual permission, accuracy, and the risk that a viewer could mistake a synthetic scene for documentary evidence. Replace or clearly label an illustrative scene when the distinction matters to the decision.
For testimonials and creator assets, confirm that the speaker actually supports the stated experience. Do not make a synthetic person appear to have used a product or achieved a result they did not experience.
6. Apply the disclosure decision
Decide whether the audience needs a disclosure about AI involvement, synthetic media, a material connection, or a sponsored relationship. The IAB AI Transparency and Disclosure Framework offers a risk-based advertising framework; local law, platform rules, and the campaign context still control.
Make the disclosure visible where the audience sees the claim. Test it on the actual placement, device, language, and format. A disclosure hidden in a profile, caption fold, or terms page may not meet the campaign’s transparency goal.
7. Check platform requirements
List the ad platform, format, audience, geography, and automated features. Review current platform guidance for synthetic or altered media. Google’s Display & Video 360 AI-generated content labelling guidance illustrates why a platform’s own labelling controls must be checked separately from an internal approval.
Record the expected label, metadata, reviewer, and fallback if the platform rejects or alters the asset. Do not assume approval in one placement transfers to another.
8. Test language and accessibility
Review translations, claims, numerals, units, disclaimers, alt text, captions, contrast, and legibility at the smallest placement size. A generated translation can change a qualification or promise. Have a fluent reviewer check high-risk languages and local references.
Render the full asset, including overlays, animations, audio, and call-to-action. Keep a text alternative for review and accessibility, but do not approve from the text alternative alone.
9. Sign the human approval chain
Use separate approvals for claim owner, rights or brand owner, compliance or legal reviewer where required, and campaign owner. Each signer records what was checked, date, scope, exceptions, and the version approved. The campaign owner cannot silently replace an approved visual with a new generated variant.
| Gate | Required evidence | Stop condition | | — | — | — | | brief | objective, audience, offer, destination | promise is undefined | | claims | source, date, scope, limitation | material claim unverified | | rights | license, consent, provenance | origin or permission unknown | | identity | likeness and testimonial review | viewer could be misled | | disclosure | placement-specific decision | disclosure is hidden or missing | | platform | current policy and label path | policy status uncertain | | rendering | device, language, accessibility check | disclaimer unreadable | | approval | named human signers and version | no accountable owner |
10. Prepare post-launch monitoring
Create an exception log for disapprovals, complaints, rights challenges, misleading comments, broken destinations, and unexpected model output. Assign an owner, response time, pause authority, and evidence-preservation rule. Monitor a sample of live placements rather than trusting the uploaded file.
Pause all variants in the same family when a systemic issue is found. Record whether the fix is a prompt change, source change, human-review change, disclosure change, or platform setting. Do not delete the rejected version before the review record is complete.
11. Make scaling conditional
Scale production only when the control sheet shows repeatable claim verification, rights evidence, visible disclosures, platform compatibility, rendered QA, named human approval, and a working pause path. If review capacity is the bottleneck, reduce the number of variants or increase review capacity before increasing generation.
Keep the checklist and source evidence local and non-indexable until overlap review, current policy checks, technical QA, and approval are complete. Creative speed becomes an advantage only when the organisation can prove which asset was approved, why it was safe to run, and how it will stop when the evidence changes.
How did this article land?
Choose one reaction. You can change it anytime.
