Meta Ad Policy and AI Disclosure FAQ: Rejections, Appeals, Claims
How AI UGC meets Meta ads policy — disclosure, common rejections, appeals, restricted verticals, before-and-after rules, landing pages and account health. Not legal advice.
Meta does not ban AI UGC. It bans undisclosed synthetic people, unsubstantiated claims, restricted-vertical tricks and destinations that do not match the ad. This hub covers disclosure, rejections, appeals, before-and-after rules, landing pages and account health — with the advertiser still owning compliance.
01
AI Disclosure
1.Does Meta require AI disclosure for UGC-style ads?
Often yes, especially when a realistic person appears to speak, demonstrate or endorse. Meta has been tightening labels for AI-generated or digitally created media, and the toggle sits in the advertiser’s workflow, not in our renderer. Treat disclosure as the default for talking-head avatar ads rather than hunting for an exception. Rules still move by region and by month, so the Ads Manager UI and Meta’s current commercial-content help pages beat any article. Klip Kanvas can surface a recommended disclosure prompt; we cannot certify that a given creative is labelled correctly in a given market. The advertiser owns the click.
#AI Disclosure
2.Who is responsible for the disclosure — us or the advertiser?
The advertiser. We generate the file; you (or your agency, acting for the brand) run the account, pick the special-ad categories, write the primary text and complete the AI or digital-media declarations Meta shows at publish time. Agencies should put that split in the statement of work so a rejection does not become a vendor dispute. If you are running client media, do not silently skip the toggle because the creative “looks real.” Looking real is exactly when disclosure is most relevant. Our flags are tooling, not legal advice, and a missing label is an integrity issue, not a creative-quality issue.
#AI Disclosure
3.Where do I actually disclose inside Ads Manager?
In the ad-creation flow, in the fields Meta currently uses for AI-generated content, digitally created media or similar commercial-content declarations — the label names change, the job does not. Complete them per ad, not once per page. Do not strip provenance from the file in the hope the system will not notice; detection and labelling run independently of what you type, and hiding the method is worse than labelling it. If your team uses a bulk uploader or a third-party API, confirm those fields are mapped. A spreadsheet of 40 variants with a blank disclosure column is how a healthy account collects 40 avoidable rejections in one afternoon.
#AI Disclosure
4.Will an on-screen “AI” caption hurt hook rate?
Usually less than advertisers fear, and far less than a rejection or a disabled account. Viewers already assume a lot of feed creative is produced; a short, factual label is not the thing that kills the first three seconds. What kills hook rate is a weak opening line, a mismatched avatar or a claim that reads as an ad. Target 30%+ hook rate (3-second views over impressions); below about 20% the problem is the hook, not the disclosure. If you want to test the caption, run it as a single extra variant in a 6-creative batch — 3 hooks × 2 avatars — rather than rebuilding the whole account around the label. Read out at 48–72 hours or ~1,000 impressions per variant.
#AI Disclosure
5.What happens if I skip disclosure on purpose?
You trade a checkbox for an integrity problem. Undisclosed realistic synthetic people, especially making health, finance or results claims, is the pattern that gets ads rejected, pages restricted and, in ugly cases, the whole ad account reviewed. There is no performance upside that compensates. If a partner asks you to “just not tick it so it feels more authentic,” that is a partner you should not run media for. Re-exporting with slightly different captions does not launder the issue. Disclose, then compete on hooks, offer and landing page like everyone else. This is a limitation of the channel, not of the generator.
#AI Disclosure#Account Health
6.Does Meta detect AI even when I do not label it?
Assume some combination of provenance signals, behavioural patterns and visual forensics exists, and assume it will improve. Designing a workflow around “maybe they will not notice” is not a media strategy. We do not provide a metadata-stripping feature, and asking for one is usually a sign the real problem is disclosure discomfort. Label it, then make the ad good. Detection false positives happen too — filmed UGC sometimes gets queried — which is another reason a documented production method in your own files helps an appeal. None of this is a promise that labelled ads always deliver; it is a promise that unlabelled realistic people are a bad bet.
#AI Disclosure#Account Health
7.Do Advantage+ or Advantage creative settings change the AI rules?
They change delivery and asset mixing; they do not relax disclosure or claims. If Meta’s systems recut, caption or combine assets, every component still needs to be a component you would be willing to run alone. Do not dump an unlabelled talking-head and a restricted before-and-after still into a catalogue and hope the optimiser picks the safe one. Keep Advantage-style automation for extra cuts of already-approved angles. If you cannot explain what the system is allowed to show, turn the extra creative enhancements off for the first test. Policy applies to the ad the user saw, not to the ZIP you meant. That limit is on the channel, and AI volume makes it sharper rather than softer.
#AI Disclosure#Account Health
02
Common Rejections
8.Why was my AI UGC ad rejected?
The rejection note is almost never “because it is AI.” It is usually personal attributes, misleading claims, before-and-after in a restricted vertical, a landing page that does not match, adult or medical content, or low-quality spam patterns from near-duplicate uploads. Open the reason code first. If it cites personal attributes, you wrote “for people who struggle with X”; if it cites landing pages, the URL does not carry the same offer. Fix the cited issue, then resubmit. Regenerating the same script with a new avatar wastes a 3–5 minute render and does not address the policy layer. Keep a log of reason codes so the next batch is written clean rather than appealed clean.
#Common Rejections
9.What are personal-attribute rejections and how do I write around them?
Meta restricts ads that assert or imply personal characteristics — health status, finances, relationships, race, religion and similar. Avatar voiceover is not an exemption. “I know you’re anxious about your blood sugar” is still a personal attribute when a synthetic face says it. Write the problem as a product situation, not as a diagnosis of the viewer: “this is the afternoon crash I used to get,” spoken as the avatar’s own story, is a different shape from “you look tired.” Have a human read every hook aloud before you render. If you stumble into “you” plus a sensitive trait, rewrite. The 15-word sentence rule that helps delivery also helps policy: short lines are harder to overload with implied traits.
#Common Rejections
10.Do realistic avatars get flagged as misleading or inauthentic?
They can, when the ad implies a real named customer, a real doctor, or a real employee of your company. Licensed stock avatars are consenting performers, not your customers, and presenting them as such is both a rights problem and a policy problem. Use UGC-style delivery, not fake testimonial chrome — no invented five-star overlays, no “verified buyer” badges you cannot support, no screenshots of reviews you do not have. Disclosure plus honest framing (“creator-style ad,” not “real customer review”) is the combination that survives review most often. If the whole selling idea is a fabricated case study, film a real one or drop the idea. AI will not make a fake review acceptable.
#Common Rejections#AI Disclosure
11.Can uploading 40 near-duplicate AI ads look like spam?
Yes. Volume is a feature of the workflow and a risk if every file is the same hook with a different jumper. Meta’s quality systems notice near-duplicates, low-engagement repeats and sudden bursts from a new page. The starting batch we recommend is still 6 creatives — 3 hooks × 2 avatars on one body script — not 40 micro-edits of one line. Iterate on the hook first when fatigue hits around frequency 2.5–3.5. Name files so a reviewer can see they are structured tests, not a scrape. If you need 20–40 variants a month, vary the first two seconds and the angle, not just the end card. Quantity without difference is how AI UGC earns a quality penalty.
#Common Rejections#Account Health
03
Appeals
12.How do I appeal a rejected AI ad?
Appeal only when the reason code is wrong. If the code is right, edit and resubmit; an appeal that argues with a real claims issue wastes the 24–72 hours you could have spent fixing the script. In the appeal, state what the ad is, that the person is a licensed synthetic avatar, that AI disclosure was completed, and which page the destination uses. Attach nothing theatrical. Do not open a second ad account to dodge the decision — that is the behaviour that turns a creative rejection into an account problem. One appeal per ad, then move. If three ads in a row fail on the same claim, the claim is the product, not the reviewer.
#Appeals
13.How long do Meta appeals take, and should I keep spend running?
Treat an appeal as days, not minutes, and do not park the whole campaign on it. Keep the rest of a consolidated test set delivering against the same audience so you still get a 48–72 hour readout on the variants that were approved. Pausing everything “until policy replies” is how a one-ad issue becomes a missed week. If the rejected ad was your only hook, you under-tested: a proper first batch has three hooks, so two can still learn. Rebuild the rejected concept as a new ad with the cited issue removed rather than waiting for a yes that may never come. Appeals restore edge cases; they do not rescue a bad claim.
#Appeals
14.Should I resubmit a slightly recut version after a rejection?
Only after you change the thing the reason code named. A three-frame trim, a new avatar and the same prohibited before-and-after still fails, and repeated near-identical resubmits look like evasion. Change the claim, the overlay, the destination or the implied endorsement — then submit as a new ad with a new name so your internal log stays readable. If you are iterating a winner that was rejected on a caption safe-zone or a minor text overlay, that recut is legitimate. If you are iterating a loser that promised a 10 kg change in 7 days, the recut is how accounts get a reputation. Use the diagnostic order: policy first, then hook, then avatar, then offer.
#Appeals#Common Rejections
15.Can Klip Kanvas certify that an ad will pass Meta review?
No, and we will not. Policy is Meta’s, enforcement is automated plus human, and both move. We can tell you the failure patterns we see, recommend disclosure, keep stock avatars on licensed likenesses, and refuse to clone a celebrity without consent. We cannot warranty an approval, a CPA, or that a restricted vertical will treat your claim kindly. Anyone selling “guaranteed to pass Ads Manager” is selling you a future dispute. The practical bar is a human review against this checklist: disclosure, no personal attributes, no fake testimonials, no before-and-after morphs, landing page match, substantiated claims. Then you launch a 6-creative test and you read the policy result the same way you read hook rate.
#Appeals
04
Restricted Verticals
16.Which verticals are hottest for AI UGC on Meta?
Health, supplements, cosmetics, weight management, finance, credit, housing, employment, and anything that looks like a miracle medical device. AI does not loosen those rules; it makes them easier to violate because generation is cheap and the script model will happily draft a claim you cannot substantiate. If you sell in one of those verticals, put a human claims pass on every script before you spend credits. Restricted or Special Ad Categories still apply regardless of whether the face is stock or cloned. For a vertical-specific angle list, pair this hub with the supplements or skincare FAQs rather than hoping a disclosure toggle covers a substantiation gap.
#Restricted Verticals
17.Can an avatar give health, supplement or finance advice?
Not as a professional. An avatar is a licensed performer reading your script, not a doctor, nutritionist or adviser, and dressing it in a white coat does not create a qualification. Stick to non-personal, non-guaranteed product statements you can support, and keep advice-shaped language out of the first two seconds where reviewers and models both look. Finance ads that promise returns, “guaranteed approval,” or debt erasure are the fast way to a restricted page. If your category legally requires a disclaimer, put it on the ad and the landing page, and still do not ask us to certify it. The generator will say whatever you type; that is a limitation, not a feature you should use.
#Restricted Verticals
18.Do Special Ad Categories change if the creative is AI?
No. Housing, employment, credit and social-issue categories are about the product and the targeting, not about how the video was made. You still declare the category, you still accept the targeting limits, and you still cannot use lookalike-style shortcuts those categories forbid. An avatar saying “people like you in this postcode” does not become acceptable because the speaker is synthetic. If your media team has been using creative to imply an audience they can no longer target, AI volume will only multiply the policy debt. Declare honestly, then test hooks inside the allowed targeting. Disclosure and Special Ad Category are separate fields; complete both.
#Restricted Verticals#AI Disclosure
05
Before & After Claims
19.Are before-and-after visuals allowed in AI ads?
Often no, particularly for weight, body, hair, skin and medical outcomes. Meta has long restricted before-and-after imagery and “miraculous” results, and an AI morph is not a clever exemption — it is usually a clearer violation. Do not generate a slim-down face swap, a skin-texture miracle or a split-screen “day 1 / day 30” on an avatar. If you have real, consented, non-misleading customer documentation, that is a different asset, and even then the vertical’s rules still apply. For AI UGC, talk about the product ritual and the problem framing, and put proof on the landing page in the form your counsel already approved. A banned visual with a disclosure label is still a banned visual.
#Before & After Claims
20.Can the avatar talk about results they “personally” got?
That is a testimonial shape, and it is one of the easiest ways for AI UGC to become deceptive. A stock avatar did not use your supplement. A cloned founder can talk about their own experience if that experience is true and if you would be willing to stand it up without the AI layer. Invented personal results, invented labs, invented “my doctor was shocked” beats — those are the ads that get both policy and consumer-protection attention. Write in product language, not in a fake diary. If the winning human UGC in your account is a genuine customer story, keep paying for that story and use AI for hook volume around it. Do not launder a fake diary through a pretty face.
#Before & After Claims#Restricted Verticals
21.What about weight-loss, acne-clear and hair-regrowth claims?
Treat them as high-risk even when your product is real. Meta’s health and wellness line items, plus the before-and-after restrictions, catch exactly the transformations AI image and video tools are good at faking. Avoid timelines (“in 7 days”), guaranteed magnitudes, and any visual that implies a body or skin change on the avatar. If you have substantiation, keep the on-ad claim smaller than the on-page evidence, not the other way around. This is a place human creators with real bodies still win, because the proof is the footage. Hybrid is the grown-up setup: avatar for the hook, real demo or real customer for the result beat. We will not claim a given script is approvable.
#Before & After Claims
06
Landing Page Rules
22.Does the landing page have to match the AI ad?
Yes. Destination mismatch is one of the highest-volume rejection reasons we see, and AI makes it worse because teams generate 20 angles and send all of them to the homepage. If the ad promises a specific SKU, price, bundle or trial, the first screen of the URL must offer that thing. If the avatar speaks Spanish, the page should not be English-only. If you disclosed AI in the ad, do not ambush the user with a fake “meet our customers” grid of the same faces. Match offer, language, geo and claim intensity. A 1%+ cold CTR into a bounce is not a creative win; it is a page problem the algorithm will eventually punish.
#Landing Page Rules
23.Can I send traffic to a page that looks like a news article or a quiz?
That format is a classic policy magnet. Clickbait article skins, fake infographics, and quiz funnels that only reveal a price after three steps get reviewed as misleading even when the product is legitimate. AI-written landing copy with sensational headlines plus an AI spokesperson is a combination reviewers have seen too many times. Use a product page, a clean offer page or a shop collection that a reasonable person would expect from the ad. If your funnel is a quiz, make it obviously a brand quiz and put the identity on screen one. Policy is not the only issue: hook rate can look fine while CPA collapses because the click was curiosity, not intent.
#Landing Page Rules
24.What checkout and tracking issues get ads disapproved?
Broken URLs, immediate app-store redirects the ad did not disclose, malware warnings, and checkout flows that surprise the user with extra products or unclear pricing. Creative cannot rescue a destination that fails the basic user-experience bar. Before you launch a 6-creative test, click the final URL on a phone, in the target country, without your office IP. Confirm the SKU, the language and the price. Server-side tracking and CAPI are a measurement topic; they do not replace a working page. If you localise an English winner into 30+ languages, localise the destination in the same batch or you have built a rejection machine with extra credits.
#Landing Page Rules
07
Account Health
25.Will using AI creative hurt my ad-account health?
Not by itself. Account health moves on policy strikes, hidden-tax quality, landing-page experience, and payment issues — not on “this file came from a generator.” What does hurt is the behaviour AI makes easy: duplicate spam, claims nobody reviewed, skipped disclosure, and blasting 50 near-copies into a new page. Keep a human QA pass: claims match the approved script, disclosure completed, URL live, captions in the safe zone, no before-and-after morphs. Frequency on cold audiences still fatiguing around 2.5–3.5 is a media fact, not a policy fact; refresh the hook rather than cloning the same ad into a second page. A clean AI account looks like a clean filmed account, just faster.
#Account Health
26.Is it safer to run AI ads from a new Page or a new ad account?
Usually no. New accounts have less trust, thinner history and a smaller error budget. A brand-new page dumping 40 synthetic testimonials is the profile that gets a quality review, not a mature shop adding a labelled test set of six. Run AI UGC inside the account that already holds your catalogue and pixel, with disclosure on, and let the existing conversion history carry learning. If the current account is already restricted, a second account is not a policy strategy — it is a way to lose both. Fix the claim and the page. Agencies should be especially careful: a “clean” client account used as a dump for rejected angles becomes the next restricted asset on the roster.
#Account Health
27.How many policy strikes before I should stop generating?
Stop at the second strike on the same issue, not at the tenth file. Two rejections with the same reason code means the script system, the offer or the vertical rules are the constraint; more renders will not negotiate with Ads Manager. Freeze that angle, rewrite with a human claims pass, and launch a smaller batch. Account-level reviews are slower and uglier than ad-level rejections, which is why the kill rule exists. Credits still cost the same whether the file runs or dies, and unused credits only roll over one billing cycle on paid plans — do not spend the allowance proving a banned claim. If you cannot describe the substantiation in one sentence, do not generate it.
#Account Health#Appeals
28.What should I check before I hit publish on an AI UGC batch?
Seven items, in this order. One: AI disclosure completed per ad. Two: no personal-attribute phrasing in hook or captions. Three: no invented testimonials or professional credentials. Four: no before-and-after or body-change visuals. Five: landing URL matches offer, language and SKU. Six: claims match the approved script and whatever substantiation you actually have. Seven: the batch is structurally a test — 3 hooks × 2 avatars — not 20 clones. Watch each 15–60 second file once with sound on. That pass takes minutes and prevents the two-week appeal. Export 9:16 at 1080p unless you have a reason to add 1:1. Then launch and wait for 48–72 hours or ~1,000 impressions per variant before you judge anything but policy.
#Account Health#Landing Page Rules
Ready to put this into practice?
Create your first AI UGC video ad in minutes — no filming, no actors, no editing.
