A practical labeling workflow for AI UGC and synthetic ads on Meta and Google, plus how to think about FTC and state disclosure — without treating this as legal advice.
If an avatar, cloned voice, generated background, or AI-written testimonial could be taken as a real person or a real endorsement, labeling is your problem — not the vendor's. Platforms auto-tag some files; they will not take the enforcement meeting for you. This tutorial is an operational checklist for Meta, Google, and how to think about FTC and state disclosure when you ship AI UGC. It is not legal advice, and following it does not make an ad lawful. Live help centers and counsel win when they disagree with this page.
Takes 30 minutes
Write it into the brief: whoever hits Publish is responsible for disclosure, claims, and likeness rights, even when a tool rendered the face. A renderer, an agency, or an editor can help you attach a label. They cannot accept the platform strike or the regulator letter for you. If that sentence is not true in your contract, fix the contract before you scale AI UGC.
Separate three risks that get mashed into 'AI compliance.' One: advertising law — misleading claims, fake reviews, undisclosed paid relationships, health and income claims. Two: platform policy — Meta and Google rules on generative content, restricted categories, and the disclosure toggles in their UIs. Three: IP and publicity — using a real person's face, voice, or trademark you do not have rights to. Labeling addresses part of two and sometimes part of one. It does not wash a false 'clinically proven' line or a stolen likeness. Teams that think 'we added AI Info' means the ad is cleared are the teams that get the next rejection.
Assign a named owner on every account: usually the media lead plus whoever approves claims. Give them a one-page map of what you generate (avatars, voices, B-roll, copy, backgrounds) and where labels go. Do not leave it to 'the tool will flag it.' Some Meta generative features self-label; third-party video often needs you to disclose. Google has been rolling an AI label setting and allowing on-creative labels so advertisers can meet regional rules. If nobody owns the toggle, it stays off until a review queue catches you.
Pro tip:Put 'advertiser owns labeling' on the creative brief next to the claims list. If legal has not signed the claims list, do not argue about the badge.
Takes 10 minutes per ad
For each ad, list what a person might reasonably believe is real: face, body, voice, testimonial identity, product demo, location, before/after, and on-screen quotes. Mark each as filmed, licensed stock, or generated/edited with AI. If you cannot complete the list, you cannot choose a label.
AI UGC usually trips the 'this looks like a real customer' wire. That is the point of the format, and it is also why disclosure and endorsement rules matter together. A talking head that never existed, reading a five-star story that never happened, is not just unlabeled AI — it can read as a fake endorsement. Use first-person language only when you can stand behind it as advertising, not as a named real customer's statement, unless you have that customer's permission and a real experience underneath. Generated 'reviews' with invented names are a hard no.
Also flag political, social-issue, housing, credit, employment, and health creatives. Those categories often have stricter disclosure and substantiation than a simple DTC gadget ad. When in doubt, treat photorealistic people, cloned voices, and event footage as needing a disclosure path, then confirm against the current Meta, Google, and counsel guidance — not against a blog post, including this one.
Takes 5 minutes per ad
In Meta Ads Manager, disclose AI-generated or significantly AI-edited photorealistic content using the controls Meta currently provides (AI info / digitally created style labels, including for third-party tools). Do not assume that because you rendered off-platform, Meta will never ask. For ads about social issues, elections, or politics, Meta has long required disclosure when image, video, or audio was created or edited with AI. Check the live Help Center at upload time — label names have already changed once from 'Digitally created' toward 'AI info.'
Meta auto-labels some of its own generative features (background generation, image generation, add animation, and similar). That is a floor, not a strategy. If you upload a talking-head video from an external avatar tool, complete the disclosure step as if detection might miss it. Missing disclosure is itself a policy problem even when the rest of the ad would have been approvable. Rejections for undisclosed AI are cheaper than an account-quality spiral after repeated misses.
After publish, open the ad as a user: three-dot menu, About this ad, and confirm the AI info treatment you expected. If it is absent and the file is photorealistic synthetic, fix it before you scale spend. Do not hide the disclosure behind a hope that a 9:16 crop is 'too native' for a badge. Native is a creative goal. Transparency is a policy goal. They can coexist; your CTR is not a defense.
Pro tip:Save a screenshot of the Ads Manager disclosure state with the ad ID. That is the artifact you want if someone asks what you declared.
Takes 15 minutes
Google has been allowing text or visual AI labels on image and video ads, and rolling an AI label setting across Google Ads, DV360, Campaign Manager, Merchant Center, and Ads Editor so advertisers can meet regional rules (including EU, India, and New York-style requirements). Use the official setting when it is available on the campaign type you run; add an on-creative label when a jurisdiction or your counsel wants the disclosure in the file itself, not only in a UI chrome the viewer may never open.
On-creative labels need to be readable on a phone in the first seconds: short, high contrast, not covered by platform UI. 'AI-generated' or the wording counsel specifies beats a tiny footnote at second fourteen. Safe-zone the label the same way you safe-zone captions. A disclosure that sits under the mute button is not conspicuous. For Demand Gen and YouTube, watch both in-stream and Shorts/in-feed crops — a label that works on 16:9 often dies on 9:16.
Google's policy pages change in dated drops (for example, labeling updates described for July 2026). The workflow is: open the current Advertising Policies Help article on AI labeling, apply what it says, and keep a dated note in the campaign folder. Do not copy a competitor's on-screen badge as your legal strategy. Their counsel is not yours. Their country mix is not yours.
Takes counsel-dependent
US federal advertising law still turns on deception and unfairness, not on whether you used a particular software brand. Material facts — that a person is not real, that a testimonial is not a real customer's, that a paid relationship exists — are the facts a reasonable consumer may need. Some states add express synthetic-performer or AI-interaction disclosures. This is the step where you talk to counsel, not where a media buyer invents a slogan.
A practical, non-legal way to brief counsel: show the file, say what is synthetic, say where it will run, and ask what must be on the creative versus in platform UI versus in a landing-page terms line. Ask specifically about endorsements (fake customers), about health claims, and about any state where you have material spend. New York's commercial-ad rules around synthetic performers are an example of a state overlay that can require a conspicuous disclosure when an ad depicts an AI-generated human who is not a real identifiable person. Other states regulate AI in other ways. Do not generalize one state's statute into a national policy without advice.
Industry blogs describe 'double disclosure' (paid relationship plus AI involvement) as if it were a single new FTC form. Treat that as a reminder that two different facts can both be material, not as a statute you can cite from a how-to. The Endorsement Guides still exist. Truth-in-advertising still exists. If your UGC script invents a user and a result, a small 'made with AI' watermark will not save the claim. Fix the script, then label the format.
Takes 20 minutes per batch
Before upload: claims list, synthetic inventory, Meta disclosure, Google label, on-creative text if required, landing page consistency. After 48 hours: open the live ad on a phone and confirm the badge or on-screen line is present. Once a month: sample ten live ads. Drift happens when a new editor clones an old file and strips the caption layer.
Store the checklist next to the brand kit, not in a legal folder nobody opens. Make it binary: disclosed in Meta (yes/no/not applicable), Google AI label (yes/no/not applicable), on-creative disclosure (wording + timecode), endorsement status (real customer / actor / synthetic persona clearly advertised). If a row is blank, the ad does not ship. Agencies should include this in the trafficking ticket so a producer cannot 'clean up' the frame by deleting the only disclosure.
When you generate AI UGC in Klip Kanvas or any other renderer, assume the export is unlabeled until you add platform and on-creative disclosures yourself. The render is production. Labeling is trafficking. Confusing the two is how a six-creative test goes live with one labeled file and five clean-looking violations. If a platform later auto-detects and dual-labels, you live with redundancy. Missing a required disclosure is the expensive direction.
Pro tip:Never let 'we'll add the label in the next iteration' ship. The iteration that spends is the one that needs the label.
Labeling is trafficking, not a debate about whether avatars count as 'real UGC.' Use Meta's disclosure controls, Google's AI label settings, and a conspicuous on-creative line when a destination or counsel requires it. The advertiser still owns deception, fake endorsements, and likeness risk. Audit live ads the way you audit pixels. A badge does not fix a claim you cannot substantiate, and this tutorial is not a legal opinion.
Not necessarily. Auto-label covers some Meta generative tools and detections. Off-platform avatar video, cloned audio, and other photorealistic edits may still need you to disclose in Ads Manager. Confirm the live ad, and ask counsel about on-creative requirements in your states.
Do not assume so. Federal advertising law is about whether the ad deceives a reasonable consumer, including fake endorsements. A watermark does not fix a false claim or an invented reviewer. Get advice for your creative and category.
In practice the advertiser is still in the blast radius — account, spend, and brand. Write responsibility into the SOW, but do not outsource accountability. Check the toggle yourself before you scale.
Script assistance alone is often treated differently from a synthetic person or a fake testimonial, but it depends on the facts and the platform. If the only AI is a first draft the talent rewrote, document that. If the face or voice is generated, disclose that path.
Create your first AI UGC video ad in minutes — no filming, no actors, no editing.
Try Klip Kanvas freeFeed ChatGPT real customer language, lock a four-beat UGC structure, force spoken English, and strip the model tells before anyone films or renders the ad.
Use Gemini in Asset Studio, Demand Gen, and Performance Max as a draft engine: feed a real brief, rewrite every line against claims, and keep human UGC as the YouTube-native asset.
Paste a product link and Klip Kanvas writes the script, casts the creator and renders the ad — no filming, no actors, no editing.
