Before your AI avatar speaks, approve these five things

An AI avatar can turn one approved recording into many videos, but speed creates a new problem: a business can publish words, claims, or facial expressions that the represented person never reviewed.

Before publishing, approve five separate things: the identity being used, the evidence behind the claims, the exact script, the disclosure plan for each destination, and the final rendered video. Approval of the script alone is not approval of the finished asset.

Keep one approval packet for each video

Use a document, project record, or protected spreadsheet. One row per video is enough.

Video ID and working title:
Person represented by the avatar:
Approved accounts and placements:
Identity permission record:
Script version and approver:
Claims and supporting evidence:
Synthetic-media disclosure by platform:
Commercial-relationship disclosure, if applicable:
Final render link and approver:
Publication URLs and date:
Expiration or review date:

The packet is an operating record, not a substitute for a contract, legal advice, or a platform's current publishing controls. It gives the business a way to answer a basic question later: who approved what actually appeared?

1. Approve the identity and its boundaries

Record whose face and voice the avatar represents. Then state where the business may use it, who can request a new video, and when the permission must be reviewed. If an employee, customer, contractor, or creator is represented, do not treat access to an avatar tool as permission to publish every future script.

A useful permission record separates four decisions:

This is a practical control, not a statement that one form satisfies every law or contract. When a likeness belongs to someone outside the business, or the video covers a regulated subject, get appropriate professional review.

2. Build a claim sheet before polishing the script

An avatar does not make a marketing claim more accurate. Pull every price, result, comparison, testimonial, guarantee, and product capability into a small claim sheet. For each one, record the source, the person responsible for it, and the date it was checked.

The Federal Trade Commission's small-business advertising guide explains that qualifying information needed to prevent deception should be clear and conspicuous. It also warns that fine print cannot reverse the impression created by the main claim.

Use three claim statuses:

• Supported: the business has current evidence and can state the claim accurately.
• Needs qualification: the claim is true only with conditions that the audience needs to understand.
• Remove: the claim cannot be supported, is outdated, or creates a misleading impression.

If the offer itself is still unclear, complete the one-page offer worksheet before writing more versions of the video.

3. Lock the script and pronunciation

Give the approved script a version number. Include spoken words, captions, prices, on-screen text, and the call to action. Also approve pronunciations of names, technical terms, locations, and branded phrases.

A script marked final should not silently change during generation. If someone changes a price or adds a stronger outcome claim, create a new version and send it through the claim check again. This prevents a small edit from bypassing the person responsible for the offer.

4. Separate two different disclosure questions

A synthetic-media disclosure tells the viewer that media was generated or meaningfully altered. A commercial-relationship disclosure tells the viewer about a material connection between a promoter and a brand. One does not automatically answer the other.

YouTube says creators should disclose realistic altered or synthetic media that viewers could mistake for a real person, place, scene, or event. Its examples include a synthetically generated voice and realistic changes to a person's likeness. YouTube also distinguishes those cases from production assistance such as idea generation or captions in its official disclosure guidance.

Meta explains that its AI information labels may be triggered by industry signals or a person's own disclosure, and that label presentation can differ between generated content and content edited with AI. Review the current destination controls rather than assuming a label will appear because the creation tool added metadata. Meta documents that approach in its AI-content labeling guidance.

If a creator, spokesperson, or affiliate has a financial, employment, personal, family, free-product, or other material relationship with the brand, review the FTC's social-media disclosure guidance. It says a disclosure should be hard to miss and placed with the endorsement. For video endorsements, the FTC recommends disclosure in the video, not only in its description.

Platform controls and laws change. Record the official page you checked, the date, and the setting used for each publication destination. When the answer is unclear, pause and verify it instead of copying a setting from another platform. The platform-change verification card provides a simple way to document that check.

5. Approve the final render, not a preview image

Watch the complete exported file with sound, then watch it muted. Check the first and last frames, captions, price, call to action, disclosure, pronunciation, lip movement, background, logos, and every visual that appears while a claim is spoken.

Use a short final-render test:

1. Does the video match the locked script?
2. Could a reasonable viewer mistake it for an unaltered recording?
3. Are required platform and relationship disclosures present where viewers will notice them?
4. Does any visual imply a result, customer, location, credential, or product feature the business cannot support?
5. Does the destination page match the offer and price in the video?

Keep the approved render link in the packet. If the generator creates a new render after approval, treat it as a new asset even when the script is unchanged.

Follow one hypothetical example

Suppose a roofing-company owner creates an avatar of himself for a short video about inspections. The first draft says, "We find every hidden problem before it costs you thousands." The business cannot support the word every or predict each viewer's future cost.

The claim sheet marks that sentence for removal. The revised script explains what the inspection covers, what it does not cover, and how a homeowner requests an appointment. The owner approves the script, the pronunciation of the company name, the platform disclosure setting, the on-screen synthetic-media notice chosen by the business, and the final render.

That process does not guarantee legal compliance, platform acceptance, or campaign performance. It produces a traceable decision instead of an unreviewed talking head.

Use a publish, revise, or stop decision

The cheapest correction is the one made before generation credits and ad spend are committed. A useful AI-video workflow should make approval visible, not make accountability disappear. If you want help planning the channel and customer path around the video, review Ocean Media's social-media service.