What is an AI advertisement? A guide to AI ads

What counts as an AI advertisement?

An ad is an AI advertisement when AI models generate part of what the viewer sees or hears. That ranges from one AI background in a filmed ad to an ad where every shot, voice and sound effect is generated.

The amount matters, because the ad platforms treat light AI edits and fully generated content differently. TikTok's ad policy separates "insignificant" AI edits, such as lighting, colour or background changes, from content that is significantly modified, including content that is completely AI-generated.

This guide is about the third level, and the second where the AI parts carry the ad.

The main formats of AI video ads

AI video ads come in the same formats as filmed ads, plus styles that are slow to film or animate by hand, such as talking mascots and claymation. Each format below links to a page of public examples of AI ads posted by creators on X. Those examples were not made on or for Anglr; each one links to the original post.

The full set is on the examples hub. Use it to study how each format opens, how it frames the product and how long it runs. The posts show what an AI ad looks like; they do not show whether one sold anything.

For the brand side, meaning how brands get AI video ads made, what goes in the brief and how to check one before it runs, see AI commercials.

How AI ads are made

A finished AI ad is rarely one generation. It is a script split into shots, each shot generated (often several times), then cut, voiced, captioned and checked in an editor, the same way a filmed ad is assembled.

The workflow most skilled AI editors follow has seven steps:

1. Start from a script. A script that already sells in one format is the safest start, because its angle and proof are tested. The method is in One winning script, made in every ad format.
2. Split it into beats. Hook, problem, proof, product, call to action. Each beat becomes one or more shots.
3. Build references. A character sheet (face, wardrobe, voice), a product sheet (front, side, label text) and a location still. Every shot uses them.
4. Generate shot by shot. Short shots are quicker to redo when one part fails. Current models generate up to 30 seconds per clip (Seedance 2.5, Kling 4.0) or 8 seconds (Veo 3.1).
5. Edit. Cut on movement or line breaks, level the dialogue, add music.
6. Add text in the edit, not the prompt. Captions, prices, offers and legal lines change per version; text in the editor can change without a new generation.
7. Check and label. Run a QC pass and apply the platform's AI label before the ad goes live.

A shot plan for a 30-second UGC selfie ad might look like this:

Format: UGC selfie   Length: 30 s   Aspect: 9:16
References: [character sheet], [product sheet], [kitchen still]

Shot 1  0–3 s    Close face, mid-action        Line: "[hook line]"
Shot 2  3–9 s    Face, then cut to problem     Line: "[problem line]"
Shot 3  9–18 s   Product in hand, close        Line: "[proof line]"
Shot 4  18–25 s  Product in use               Line: "[what it does]"
Shot 5  25–30 s  Face, pointing down          Line: "[call to action]"

Edit: captions on every line, [offer text] on shot 5, AI label at upload

AI ad generators vs skilled AI editors

There are two ways an AI ad gets made: a self-serve app makes it from a template, or a person makes it with AI models. The pages that rank for AI video ads mostly describe only the first.

Self-serve apps are built around stock AI presenters. Arcads, for example, lists more than 1,000 AI actors on its home page, and HeyGen's pricing page lists avatar videos, voice cloning and translation. These tools are fast for one format: a person talking to camera.

The edit limits are worth reading before you choose a self-serve tool. Arcads' terms grant a non-exclusive licence on each video and limit edits to cropping, music, text, subtitles, B-roll, sound effects, format changes and adding the client's name and logo. Any other change needs written approval.

A skilled editor uses the models directly and combines them: stills in one tool, motion in another, voice in a third. Which tool suits which format is covered in Best AI video generator for ads, by ad format.

What separates an AI ad from an AI video

An AI ad has a job: get a specific viewer to act. An AI video only has to look good. The difference shows up in five places.

• The first three seconds. The hook line or image has to stop the scroll and set up the angle. Ad hooks has patterns by format.
• The angle and proof. The ad argues one thing and backs it up. A model will not write a better angle for you; start from one that already works.
• The product. Shape, colour and label text have to match the real product in every shot. If a label changes between shots, re-generate that shot rather than fixing it in the edit.
• The sound. Lips match the words, the voice stays the same across shots, and there are no added voices or music you did not ask for.
• The versions. One ad becomes many: new hooks, new formats, cut-downs for each placement. Volume of good versions is what lets a buyer find the one that runs.

AI ad labels on Meta and TikTok

Both platforms label AI content in ads, and TikTok can reject an AI ad that is not labelled. Plan the label into the workflow, not after upload.

TikTok. AI-generated content in ads is allowed only if you "Apply the AIGC label, or by adding a clear disclaimer, caption, watermark, or sticker of your own", and undisclosed AI content can be rejected or restricted (ad policy). In Ads Manager the label is a toggle that resets on every new or duplicated campaign (disclaimer help).

Meta. Meta labels ad images created or significantly edited with its own generative AI features, or with third-party AI tools, with "AI info" (Meta help). Social issue, election and political ads have a separate disclosure rule for realistic AI people, events and audio (Meta policy).

The full rules, including what counts as a significant edit, are in AI-generated content labels on Meta and TikTok ads.

Common faults in AI ads and how editors fix them

These are the faults to look for in AI video, with the usual fix for each. Check for every one of them on every version.

A QC checklist before an AI ad goes live

Every AI ad needs a check by someone looking for AI faults and rule problems, ideally not the person who made it. Use this list on every version.

[ ] Face, hair and wardrobe match in every shot
[ ] Hands: right number of fingers, natural grip on the product
[ ] Product shape, colour and label text match the product sheet
[ ] Lips match every word; no extra voices, music or sound you did not plan
[ ] Captions match the audio word for word
[ ] Every claim is one the brand has approved
[ ] No real person's likeness used without permission
[ ] Correct aspect ratio and length for the placement
[ ] AI label applied (TikTok toggle or own disclaimer; Meta AI info)

Who makes AI ads for brands

AI ads are made by brands' in-house teams, by independent AI creators and editors, and by AI agencies that run teams of editors. The self-serve apps above are one option; skilled people using models are the other.

Anglr pays AI creators, editors and agencies on performance to remake winning ad scripts in every format: UGC, podcast, drama, animation and brand film. Anglr provides the winning scripts and angles; the creator's job is to scale them in every format, in volume. Every video gets a QC and compliance review before it goes live. If you already make full AI ads and know they perform, see how it works for creators.