ad-policy-check
Used by: make-static-ads (paid ads) and run-paid-campaigns (creative it checks at launch). Organic, email and creator content: load disclosure.md instead (or as well, when a paid ad uses a creator).
Craft knowledge, not legal advice. Policies change: when a rule decides whether an ad ships, tell the human to confirm it on the platform's live policy page. Claims are a separate check (the claims-gate skill, ../claims-gate/SKILL.md); run that first — a line cut there is not reworded here.
Reads: the ad's planned or rendered text (copy, on-image or on-screen words, end cards, captions), what the picture shows, the people in it, the platforms and placements from the brief, the brand's category, and the brand's dont / must learnings (past disapprovals).
Writes: one policy line per ad and platform, and any set-level flag.
1. Category first
Regulated (health, supplements, food with health, nutrition or free-from claims, cosmetics with efficacy language, alcohol, cannabis/CBD, tobacco/vape, financial products, crypto, weight loss, kids' products, regulated professions, gambling, dating, political) or a special ad category on Meta (housing, employment, credit, social issues)?
- No brand rule on file for it → flag the whole set "needs legal review before launch".
- A special ad category → write it in the hand-off so whoever launches sets the flag; copy may not target protected traits.
- Gambling, dating, political → most platforms need pre-authorisation: flag and stop.
- Market matters (melatonin is prescription-only in the UK and Australia; the EU allows only authorised health-claim wordings; Canada needs licensed claims) — know the market before writing.
| Category | Rule of thumb |
|---|---|
| Health, supplements | Structure/function wording only with substantiation ("supports restful sleep"); never disease language ("treats", "cures", "for insomnia", "anxiety relief") |
| Cosmetics with efficacy | "Treats acne", "heals", "boosts collagen" are drug-like; approved wording only |
| Alcohol, cannabis/CBD, tobacco/vape | Age-gated; often banned or pre-approved; no health benefits |
| Financial products, crypto | Certification, risk disclosures, no guaranteed returns |
| Weight loss | No before/after, no time-bound promises; banned on Pinterest |
| Kids' products | No direct appeals to children; no children in melatonin ads, no "candy"/"treat" |
| Regulated professions | Local rules (e.g. no patient testimonials for some medical services) |
2. Per ad, per platform
Meta (Facebook, Instagram)
- Personal attributes: no assertion or implication that the viewer has a trait — health condition, body, age, finances, religion, orientation, ethnicity. "Struggling with your acne?", "Can't sleep?" → rejected. Situational ("11am, and skin already feels tight") and identity-by-behaviour ("for people who rehearse conversations in bed") pass. First person passes the platform rule but, from a generated person, must also pass the speaker test of the maker playbook.
- Body image and before/after: nothing that makes people feel bad about their body; no before/after for weight loss; for beauty and health treat it as needing the human's OK — same conditions, no retouch, typical result; prefer old-vs-new routines, never faces or bodies.
- Misleading and sensational: no fake play buttons, close-X, notification badges, fake DMs; no shocking imagery.
- Branded content: a real creator's face or name needs their permission and the paid-partnership label.
- Text-heavy images get less delivery — keep image words short: a hook and at most one proof line (a number, rating or claim), whatever slots the layout has. Other slots carry a few plain words each (list items of 2–4 words, labels).
TikTok — stricter on supplements, weight management and "too good to be true" results, varies by market; no exaggerated before/after; must not imitate organic UI (fake "For You" chrome, fake like counts). Music: only TikTok's Commercial Music Library or generated/licensed music.
Google (Display, Demand Gen, Shopping, Performance Max) — misrepresentation (unclear pricing, missing terms, unavailable offers), healthcare and finance certification, trademarks in text; no gimmicky caps or symbols; Shopping/PMax product images carry no promotional text, watermark or border; image text and logos kept under about a fifth of the image.
LinkedIn — professional claims need the customer and the study behind them; no sensational images; no targeting language on protected traits.
Pinterest, Snapchat, X, Reddit — the same core rules; Pinterest bans weight-loss ads; Reddit punishes ads that pose as user posts.
Every platform
- Third-party trademarks and likenesses: no rival logos or celebrity lookalikes without permission; integration names in plain words only when true.
- Music: generated or licensed only — never "the song from the reference" or a sound-alike.
- Realistic generated people or voices: plan the platform's AI label (Meta "AI info", TikTok AIGC, YouTube altered/synthetic) in the launch notes.
3. Disclosures on the asset
- Offer terms that change the meaning ("Free shipping on orders $45+", not "Free shipping"); an end date on any time-limited offer.
- Material connection: paid creator, gifted product, affiliate → "Paid partnership" / "#ad".
- Required category disclaimers when the brand's rules say so (e.g. a supplement statement).
- A disclaimer never rescues an unsubstantiated outcome claim and never cures a fake testimonial — cut the claim instead.
4. Fix honestly
Rewrite the phrasing to the rule (situation framing instead of "you have…"), remove the fake element, add the disclosure. Never disguise a violation to get past review.
Disapproval fix: map the platform's stated reason to the exact line or element (it may be the ad's text, not the image or video); fix that; re-run the claims check and this check; offer to save the lesson as a brand rule through the save-learning skill (../save-learning/SKILL.md) (only with the human's own words).
5. Report
Per ad: pass (Meta, TikTok) · fixed (what, why) · flag (risk, rule, decision needed). Never "pass" a rule you could not check — say "not checked: , ". In chat, plain words ("rewritten for Meta's rule on personal traits"); rule names stay in the hand-off file.
