Use AI to address a specific creative need, such as a clearer product explanation or a different opening. The production method does not establish better advertising performance.

Verify product and placement

Check packaging, proportions and demonstrations against approved references. Ask the person running advertising to confirm current placement requirements and inspect the final export in account previews. AI Vidia's Start, Grow and Scale plans supply 9:16 videos up to 30 seconds; additional requirements need scoping.

Record AI use and plan the test

Meta's transparency post (3 February 2025, updated 1 June 2026) says ads made with Meta's own tools that show an AI-generated photorealistic person get a label next to Sponsored. Meta also says it will detect ads created or edited with third-party AI tools through industry-standard signals and show "AI info" under About this ad, and that the rollout is gradual and may vary by region. Check current account requirements before publication.

What each platform and market requires is in AI disclosure rules for Meta, Google and TikTok, and who has to label under the EU AI Act is in the Article 50 guide. How AI Vidia agrees the label with you per platform and market before delivery is on AI ad governance.

Keep a record of generated and edited elements. Labeling does not replace accurate claims or rights checks. Agree a test metric with your media buyer and investigate performance changes in context rather than assuming a fixed fatigue window.

Illustration of a phone showing an ad with AI information. The actual display can change.
Illustration of a phone showing an ad with AI information. The actual display can change.
Illustrated overview of ads and budget split. It shows no documented account data.
Illustrated overview of ads and budget split. It shows no documented account data.

Give each variant a reason to exist

Start from a missing explanation or demonstration in the current ads. A changed opening, clearer product action or alternative substantiated argument can answer a real question. Generating near-identical videos without a hypothesis produces files without necessarily producing useful learning.

Check the entire clip for packaging changes and implausible hand contact. A generated demonstration cannot substantiate an untested product effect. Retain information about the references, voices and generated elements for the approver. When performance changes, investigate spend, audience, offer, landing page and measurement alongside creative. A fixed replacement day cannot by itself demonstrate that every ad has become fatigued. To compare a new variant fairly, see how to test a new AI video against your best ad. The production itself is described on AI video ads.