What should a human approve before an AI-made ad goes live?

An AI-made ad is ready to launch only when a named person approves the exact version that will run, the facts it states, the rights behind its source material, the destination it opens, and the disclosure settings required for that placement. A prompt history or a platform's green approval status is not that record.

Use one approval record for each release family: the master concept plus the precisely identified sizes, crops, captions, voiceovers, headlines, and destination combinations covered by the decision. If a generator, editor, platform, or person changes a controlled field later, create a new version or explicitly approve the change.

This workflow governs publishable advertising creative across copy, image, and video. If a synthetic presenter or cloned voice is involved, also use the AI-avatar video approval checklist for identity permission, script, pronunciation, and final-render review.

1. Give the approval a stable identity

Start with a record that another person can find without opening an advertising platform:

Do not call a folder named final the version. Preserve the exported file or exact text, its identifier, and the settings that affect what a viewer sees. A platform-created crop, translated headline, generated background, expanded text, or alternate destination is a different output when it can change meaning.

2. Prove the source rights

List every meaningful input: photography, footage, logo, product image, music, typeface, testimonial, review, spokesperson, customer quote, voice, likeness, dataset, and generated element. For each one, record its source, license or permission, permitted uses, territory, expiration, and any required credit or restriction.

Do not treat an AI tool's output as proof that every input and use is cleared. The record should make it possible to answer: Who supplied this material? What are we allowed to do with it? When does that permission end? Who removes it if permission changes?

If the answer depends on a contract or law, route it to the qualified person responsible for that review. This checklist organizes evidence. It does not replace legal advice.

3. Freeze the offer facts

Copy the customer-facing facts into the approval record instead of relying on memory:

• Product or service name and included scope
• Price, currency, fees, billing interval, and taxes where relevant
• Discount amount, eligibility, start date, end date, and redemption limit
• Geographic availability, inventory, appointment limits, or other conditions
• Cancellation, renewal, guarantee, trial, or refund terms shown or implied
• The exact call to action and destination

Then compare those facts with the current destination. TikTok's current policy prohibits price, discount, and promotion mismatches between an ad and its landing page. LinkedIn's current review guidance likewise requires accurate offers and a destination that matches the ad. Treat that comparison as a release test, not a copywriting preference.

Use the one-page offer worksheet when the business has not settled the offer, then run the landing-page preflight on the final destination.

4. Build a claim ledger

Pull every objective or implied claim out of the creative. Include on-screen copy, narration, demonstrations, before-and-after images, charts, badges, testimonials, comparisons, guarantees, and visual suggestions about likely results.

A person should be able to trace each retained claim to current evidence. Do not use tiny text to reverse the main impression. TikTok's current ads policy rejects exaggerated or absolute results, and LinkedIn requires truthful claims. Platform approval does not turn an unsupported statement into a supported one.

5. Record AI involvement and disclosure decisions

Write what AI changed, not just which tool was used. Separate drafting assistance from a generated or significantly altered person, voice, scene, product demonstration, testimonial, result, event, or other realistic media. Then record the platform control, the label or disclaimer selected, who verified it in the final placement, and the official policy page checked on that date.

Google's July 2026 policy update permits advertisers to identify AI-generated or AI-modified image and video creative through its label setting and says certain assets made with Google AI tools may be labeled automatically. Google also states that using the setting does not by itself establish compliance with applicable laws. Meta says AI information can appear for ads created or significantly edited with its generative tools and that it is expanding detection through third-party industry signals. TikTok requires disclaimers for covered AI-generated, synthetic, or manipulated media.

These controls differ by platform, placement, region, and content. Do not assume embedded metadata, a creator-tool label, or one platform's selection follows the file everywhere. Verify the live or previewed placement each time. Record separate commercial-relationship, regulated-content, or election disclosures when they apply.

6. Check restricted categories and sensitive implications

Record whether the ad touches housing, employment, credit, financial products, health, politics, elections, alcohol, gambling, minors, personal attributes, or another restricted category. Document the applicable reviewer, permitted targeting, required certification, content limits, and region.

Also ask what the creative implies. A generated person can appear to be a customer, professional, employee, public figure, or real witness. A generated product scene can imply features or results the actual product does not have. Mark the role clearly in the record and remove a presentation that creates a false material impression.

7. Approve brand rules and every covered variant

Record the approved logo, colors, tone, product naming, pronunciation, prohibited phrases, safe areas, caption treatment, and required notices. Then list each covered combination rather than writing all variations.

Measure version coverage before release:

Approval coverage = approved live variants / total live variants x 100

Field drift rate = changed controlled fields / reviewed live fields x 100

Suppose a campaign has 18 live variants. Sixteen trace to an approval record, so approval coverage is 16 divided by 18, or about 88.9%. Those 16 variants contain 128 reviewed controlled fields, and four fields differ from the approved record. Field drift is 4 divided by 128, or 3.125%.

Neither percentage is a safety score. The two uncovered variants must be paused or approved, and the four changed fields must be reviewed. One changed price can matter more than many correct colors. The calculation makes the gap visible without pretending that all fields have equal risk.

8. Test the release and withdrawal paths

Run the actual output through a small release matrix:

Test with previews or non-spending methods when available. If launch is required to inspect a placement, use the smallest controlled scope authorized by the business. Do not spend merely to complete this checklist.

9. Keep approval connected to change control

Approval ends when a material input changes or the review date passes. Changes to price, offer, evidence, destination, claim, spokesperson, generated scene, disclosure requirement, region, account, category, or platform-created variant should reopen the affected checks.

Retain the approval ID with campaign and asset records. Log the old and new version, reason, requester, reviewer, test result, and deployment time. Use the conversion-tracking change audit when the release also changes forms, events, destinations, or measurement.

Make one release decision

Human approval should not be a ceremonial click after generation. It should bind a responsible person to an exact release, current evidence, a defined scope, and a practical stop path. If you want help turning the record into a controlled campaign workflow, review Ocean Media's paid advertising service.

For general education only. Platform details and examples can change. Verify important information before acting. Ocean Media Marketing is not responsible for decisions based solely on this resource. Suggest a correction