What reviewers check
Advertising law does not change because software made the ad. In the UK, the CAP Code applies to AI-generated ads exactly as to any other, and the ASA enforces it. In the US, the FTC applies Section 5 of the FTC Act on deceptive practices whatever the production method. The core checks stay the same: claims true and substantiated before publication, comparisons fair, prices and offers accurate, required disclaimers present, nothing misleading by omission. AI adds its own failure modes: a product shown with a feature it lacks, lettering inside an image that nobody proofread, results depicted more dramatically than the evidence supports.
The questions that are new with AI
Three questions join the standard checklist. Does this asset need an AI disclosure? That depends on the market, the platform and whether the content is realistic enough to be mistaken for real footage or a real person. Is a real person's face or voice involved, and is there a signed consent covering this exact use, territory and period? Were the models and every input cleared for commercial use, including reference images, fonts, music and any third-party marks visible in a generated scene? Reviewers should not have to investigate these from scratch; the answers should arrive attached to the asset.
Routing by risk
Not every asset needs a lawyer. Build tiers. Low risk covers resizes, crops and translations of approved content with no new claims, which go through brand or market review only. Medium risk covers new copy built from pre-approved claims, or new scenes with approved products, which get a compliance checklist and spot checks. High risk covers new claims, comparative advertising, health, financial and environmental statements, real people, children, and anything political or sensitive, which get full legal review. Agree the tiers with legal in advance and write them into the workflow, so routing happens at intake rather than depending on who happens to be asked.
Templates and evidence trails
Two things make review fast and defensible. The first is templates: pre-approved layouts with fixed legal lines, a claims library holding the substantiation for each claim, and standard disclosure wordings per market. An asset built from approved parts needs checking only for what is new. The second is an evidence trail: for each asset, the brief, the models used, the source and reference files, the approvals with names and dates, and the published versions. In the studio, approved claims and banned words are enforced at generation, and each asset carries an audit trail and a record of the model that made it, so the answer to how it was made already exists when a complaint arrives.
Updated 25 September 2026 · General information, not legal advice. Rules change, so check the current text with your legal team before relying on it.