ComplianceChecklistMeta AdsReference

Meta Ad Policy Checklist: The Pre-Submission Review Sheet

A pre-submit review sheet for Meta ads: personal-attribute traps, banned-phrasing swaps, restricted-vertical gates, landing-page match, AI disclosure and the appeal path. Not legal advice — the advertiser owns compliance.

Updated 2026-02-1612 min read

LanguageEnglish

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 is a pre-submission review sheet, not a warranty and not legal advice. Policy is Meta's, enforcement moves, and the advertiser owns compliance — passing Ads Manager is not the same as being authorised.

Read the reason code, not the myth that 'AI got us 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 a spam pattern from near-duplicate uploads. Open the 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. Regenerating the same script with a new avatar wastes a 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. Two strikes on the same issue and the angle is frozen until a human rewrite — more files will not negotiate with Ads Manager.

Personal attributes and fake proof are the expensive habits

Meta restricts ads that assert or imply personal characteristics — health, 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 the speaker's own story, not as a diagnosis of the viewer. Licensed stock avatars are consenting performers, not your customers: invented five-star overlays, 'verified buyer' badges you cannot support, and 'my doctor was shocked' beats are both a rights problem and a policy problem. Before-and-after morphs on weight, body, hair, skin or medical outcomes are high risk; an AI slim-down is not a clever exemption. A banned visual with a disclosure label is still a banned visual.

Disclosure, destination and account health sit with the advertiser

We generate the file; you (or your agency, acting for the brand) run the account, pick Special Ad Categories, write primary text and complete the AI or digital-media declarations Meta shows at publish time. Klip Kanvas can surface a recommended disclosure prompt; it cannot certify that a given creative is labelled correctly in a given market, and it does not issue SOC 2 letters, legal opinions or 'guaranteed to pass Ads Manager' stamps. Anyone selling that guarantee is selling a future dispute. New Pages and second ad accounts are not a policy strategy. Volume without difference — 40 near-copies of one hook — looks like spam. The practical bar is a human review against the locked checklist, then a 6-creative test whose policy result you read the same way you read hook rate.

How to use the sheet below

Work the locked tables in order before you hit publish: personal-attribute phrasing, the swap list for high-risk lines, restricted-vertical and Special Ad Category gates, landing-page match, synthetic-media and proof rules, then the appeal path. If a row fails, fix that row and resubmit; do not appeal a real claims issue. This sheet is an operational QA, not counsel. Your lawyer, your substantiation file and Meta's current commercial-content help pages still beat any article.

6 tables: personal-attribute traps, banned-phrase → safer-shape swaps, restricted-vertical and Special Ad Category gates, landing-page match, AI/proof/before-after rules, and the appeal / account-health path. Not legal advice.

Personal-attribute traps (rewrite before render)

Health status aimed at 'you': Fail: 'you look tired / your hormones / I know you're anxious about X' · Pass shape: speaker's own past situation, no diagnosis of the viewer
Body, weight, skin, hair, ageing as the viewer's trait: Fail: 'for people who struggle with acne/weight' · Pass shape: product ritual + problem framing without naming the viewer's body
Finance / credit / debt: Fail: 'you can't get approved / drown in debt' · Pass shape: product feature, no implied personal financial status
Relationships, race, religion, sexual orientation: Fail if asserted or implied about the viewer · Do not use as targeting-by-creative
On-screen captions: Same rule as voiceover · a line that is legal spoken can still fail as text · read the hook as primary text
15-word check: If you stumble into 'you' + a sensitive trait when reading aloud, rewrite · short lines are harder to overload
Interest stacking to imply a trait: Special Ad Category rules still apply · creative cannot restore targeting the category forbids

High-risk phrasing → safer shape (still not a legal clearance)

'Clinically proven' / 'doctor recommended': Do not use unless you have current substantiation you would put in a file · swap to specific, sourced product facts without a white-coat avatar
'Guaranteed' / 'risk-free' / 'works for everyone': Strip · swap to offer terms that actually exist (returns window stated on the page)
Timelines ('in 7 days', '10kg', 'week-one glow'): High risk on health/beauty · remove magnitude + clock · put any real evidence on the landing page your counsel already approved
'My doctor was shocked' / fake diary results: Fail · stock avatar did not use the product · founder may speak only to true own experience
Income / 'quit your job' / 'guaranteed approval': Fail on finance and info-product ads · swap to what the product does, not the viewer's bank account
Competitor naming / 'we replaced our agency': Avoid unless legal signed the comparison · no invented metrics
Before/after language in captions: Treat as the visual rule · 'day 1 / day 30' overlays fail even when the footage is mild
Advertiser owns the line: Generation does not certify a claim · if you cannot describe the substantiation in one sentence, do not ship it

Restricted verticals and Special Ad Categories

Hottest verticals for AI UGC: Health · supplements · cosmetics · weight · finance · credit · housing · employment · miracle-device claims
Special Ad Category (housing, employment, credit, social issues): Declare it · targeting limits apply · AI creative does not loosen the category · complete the field separately from AI disclosure
Avatar as professional: Fail: white coat, 'I'm a nutritionist/adviser' · avatar is a licensed performer reading a script, not a qualified expert
Weight / acne / hair-regrowth: High risk even when the product is real · no body-change visuals on the avatar · hybrid: avatar hook + real demo if you have one
Human claims pass: Required on every script in restricted verticals before credits are spent · two strikes on one issue → freeze the angle
Regional rules: US winner is not automatically UK/EU-safe · do not copy a claims set across markets unchanged
Passing Ads Manager: Not the same as being authorised by a regulator · advertiser owns both layers

Landing-page and destination match

SKU / offer / price: First screen of the URL must offer the thing the ad promised · homepage dumps of 20 angles are a high-volume reject
Language / geo: Spanish ad → Spanish page · localised winner needs a localised destination in the same batch
Claim intensity: On-ad claim ≤ on-page evidence · do not ambush with a 'meet our customers' grid of the same stock faces
Fake article / quiz skins: Policy magnet · if you use a quiz, it must be obviously a brand quiz on screen one
Broken URL / surprise checkout: 404, undisclosed app-store hop, unclear pricing, malware warnings → disapproved · click the final URL on a phone in-market
Phone QA: Target country · no office IP · confirm SKU, language, price before a 6-creative launch
CTR into bounce: ≥1% cold CTR into a bounce is a page problem the algorithm will punish · not a hook win

AI disclosure, proof assets, before-and-after

AI / digitally created media field: Complete per ad as the default for talking-head avatars · Ads Manager UI names change, the job does not
Bulk / API uploads: Disclosure column mapped · 40 blank cells = 40 avoidable rejections
On-screen 'AI' caption: Usually less damaging than a rejection · if you test it, one extra variant, not a rebuild of the account
Skip disclosure on purpose: Integrity issue · no performance upside · do not strip provenance hoping detection will miss it
Fake testimonial chrome: Fail: invented reviews, verified-buyer badges, named customer who is a stock avatar
Before-and-after / body morphs: Often no on weight, body, hair, skin, medical · AI morph is usually a clearer violation · disclosure does not save it
Real consented customer proof: Different asset · still subject to typicality and vertical rules · safer on the landing page than as the thumbstop
Advantage+ creative mix: Every mixed asset must be runnable alone · do not dump a restricted still into a catalogue and hope the optimiser picks the safe one
What we will not certify: Approval · CPA · labelling in a given market · SOC 2 · 'guaranteed to pass' · the advertiser owns the click

Appeals, resubmits and account health

When to appeal: Only if the reason code is wrong · if the code is right, edit and resubmit · appeals take days, not minutes
Appeal contents: What the ad is · licensed synthetic avatar if applicable · disclosure completed · destination URL · no theatre
Resubmit rule: Change the cited issue · a 3-frame trim with the same before/after still fails · new name so the log stays readable
Keep spend running: Do not park the campaign on one appeal · a proper batch has 3 hooks, so 2 can still learn
Two strikes, same issue: Freeze the angle · human rewrite · more renders will not negotiate · account reviews are uglier than ad rejects
Near-duplicate volume: Start at 6 creatives (3 hooks × 2 avatars) · 40 micro-edits of one line looks like spam
New Page / new ad account to dodge: Usually worse · thin history, smaller error budget · not a policy strategy · agencies: do not dump rejects into a 'clean' client account
Pre-publish 7-item gate: Disclosure · no personal attributes · no fake testimonials · no morphs · URL match · claims match substantiation · batch is a test not 20 clones

Locked

Unlock the full cheatsheet

Enter your email and we'll instantly unlock the complete cheatsheet — plus send you fresh ad playbooks every month.

No spam. Unsubscribe anytime.

Ready to put this into practice?

Create your first AI UGC video ad in minutes — no filming, no actors, no editing.

Try Klip Kanvas free