How to Make Supplement Video Ads That Pass Policy Review

List every sentence the pack, the PDP and your counsel will support, market by market. Structure/function language, ingredients, format, when to take it, and routine are the usual survivable zone. Disease treatment, implied regulator approval, guaranteed weight loss or muscle, and invented labs are the death zone. If the whole hook is a medical outcome, you do not have an ad problem yet. You have a claims problem.

The advertiser owns this stack, and often the agency. Speed is not a defence. Keep versions of the script. Do not tell a client the avatar is 'cleared'. Tell them you have a claims process. 'Supports immunity' and 'boosts energy' are not automatically safe: they are softer than disease claims and still regulated. Boosts is spikier than supports; immunity is spikier than a specific authorised wording if you sell in the EU. Do not copy a US winner into the UK or India unchanged. Passing Ads Manager is not the same as being authorised.

If you would not put the sentence on the label, do not put it in the first two seconds of a 9:16 ad. Write the spoken line as a product fact a pack could support. Put any stronger evidence on a landing page that actually carries it, not as a shouted opener. A claims sheet per market is the brief. The renderer will say whatever you type; that is a limitation, not a feature you should use to draft a medical outcome you cannot substantiate. If your growth plan depends on a line the pack cannot carry, stop generating and fix the offer.

Strip anything that treats, cures, prevents or diagnoses a disease; guaranteed transformations; 'clinically proven' without documents you actually hold; and 'replaces your medication'. Drop fake titles: doctor, nutritionist, pharmacist, unless that real person consented and is allowed to speak. Plan as if you cannot use before-and-after body or energy shots, even when a platform lets a cut through.

Meta is aggressive on before-and-after for personal attributes — weight, muscle, skin, hair, ageing. AI-generated 'after' bodies are worse: you are inventing a result. A spoken 'day 1 versus day 30' with a calendar can still be a before-and-after, because reviewers look at the net impression, not at whether you used a split-screen. 'I have more energy' in a stock face is still a testimonial-style result claim. Safer: this is how I take it in the morning, plus an ingredient or format fact. If the ad only works because of a 30-day miracle, it is the wrong ad for this vertical.

An avatar is a scripted presenter, not a customer. First-person medical diaries in a generated face are the pattern platforms and regulators hate. A quiet 'results not typical' footer will not rescue a strong spoken miracle. A white-coat stock face is a fake expert, and it is one of the fastest ways to a rejection or a worse problem. If you have a real formulator, clone them with verified consent or film them. Otherwise do not costume your way into authority. Casual-creator casting outperforms fake expertise on both policy and trust here more often than brands expect.

Write three openings only your SKU could say: routine (this is the part of my morning), format and taste, or a specific ingredient explained without a disease. Objection hooks (I hate pills) work if the payoff is the format, not a cure. Avoid borrowed medical TikToks and competitor transformation memes. Specificity is the legal strategy and the performance strategy.

Keep each hook under 15 words and deliverable in under three seconds. Put the product in the first two seconds. A talking head promising a feeling without a pack looks like a scam, and the pack is the proof the avatar cannot be. Show the format — scoop, capsule, stick, drink — even if the hook is spoken. Founder or formulator stories work when the person is real and the story is a reason to trust the product, not a disease claim in costume. Authority is not a claims holiday. If the formulator cannot say it on a label, the clone cannot say it in Reels either.

Do not cast the avatar as unwell. Exhaustion, distress and body shame read as a personal-attribute before-and-after even without photos, and they invite the policy bucket you are trying to avoid. Problem language can live in the words (I kept forgetting to take anything) without a performance of illness. If the brief says 'make them look tired so the energy claim hits', rewrite the brief. Shame is not a hook family this vertical needs, and it is miserable to watch besides. Cast a functional, ordinary person having an ordinary moment with the product.

Cast someone who looks like the buyer a few years into using the product, in the room where they actually take it — kitchen, car, gym bag, bathroom. Not a medical-ad extra and not a fitness-model fantasy unless that is truly the customer. Then film the pack. Hybrid is the default: face for the line, real scoop and mix shots for the proof. Do not ask the avatar to pinch a capsule between generated fingers.

A practical 30-second shape is a short avatar hook, real product footage in the middle, avatar CTA. Pure talking-head can still test a hook cheaply. Do not scale a pure-head winner that never shows the product; it will fatigue into looking like every other wellness scam in the feed. If you have no pack footage, film three phone clips today: pack on the counter, scoop or pop, mix or swallow-adjacent. That is more important than a fourth hook line. Age band and energy matter more than aspirational. Match market: a US face reading a UK or India claim set is both a performance and a policy miss.

Stock library faces are enough to learn. Clone a founder, formulator or real creator when that person is the trust signal and you have consent. Do not delay the first 6-ad test on a clone. If you clone a customer, you still cannot invent their medical outcome. A consented clone of a fake story is still a fake story. Face rights and claim rights are different problems, and neither is solved by a more realistic smile. For a lot of SKUs the library face plus real pack footage is the whole first test.

Whenever a reasonable viewer could think a real person is reporting a real bodily result — which is most supplement UGC — treat disclosure as required. Use platform toggles and a readable on-screen line in the language of the video. Then send the click to a PDP that offers the same SKU, the same claim intensity, and no fake customer grid of the same faces.

Ambiguity is not a performance edge. An on-screen AI caption usually hurts hook rate less than advertisers fear, and far less than a rejection. Target 30%+ hook rate on cold traffic; below about 20% the problem is the hook, not the label. Missing branded-content or synthetic-media declarations are a common rejection path on their own. We can surface a label; we cannot certify it satisfies every regulator. In a health-adjacent category, label more than you think you need, in the language of the video, on screen long enough to read.

Destination mismatch is the other high-volume failure. If the ad promises a specific SKU, flavour, bundle or subscription, the first screen of the URL must offer that thing. If the avatar speaks Spanish, the page should not be English-only. Clickbait article skins and quiz funnels that hide the price are policy magnets and CPA traps. A 1%+ cold CTR into a bounce is not a creative win. Match offer, language, geo and claim intensity, then click the final URL on a phone in the target country before you launch.

Six files: 3 claims-safe hooks × 2 everyday avatars, one body that shows the real pack, conversion objective to a matching PDP, disclosures on. No lab coats, no day-1/day-30 bodies, no cures. Read hook and CTR at 48–72 hours or about 1,000 impressions; keep what approaches 30%+ hook and 1%+ outbound CTR without a rejection. Then iterate inside the winning family.

Use the same working floors as the rest of the canon, not a made-up supplement industry average: hook 30%+ on cold (weak below ~20%), hold 12–20%, outbound CTR 1%+ cold. This category can look clicky on miracle hooks and then fail CPA — that is a quality problem, not a reason to lower the CTR target. If conservative, compliant hooks sit at 18% and miracle hooks sit at 40% with terrible CPA and rejections, you do not have a performance mystery. You have a claim you should not be running. A rejected ad has no CPA. A miracle ad with refunds has a fake CPA. Build refund rate into the allowable number.

If Meta approved it, you can scale the file inside that platform's current mood; you cannot treat approval as a legal opinion. Approvals reverse. Competitors complain. A UK or EU complaint does not care that Ads Manager was green. Keep the conservative cut as the scale cut. On cold traffic, watch frequency 2.5–3.5. Refresh the hook first, then the face, then the proof beat. If you can only run one approved script, you have a compliance bottleneck, not a fatigue mystery — write more safe lines, not more adjectives on the unsafe one.