Can I Make AI Video Ads for Dietary Supplements? FTC, TikTok, Meta

You can, but making the video with AI changes none of the supplement rules. The ad is judged on its claims, who it targets and where it runs. What AI adds is a new way to say something you cannot support, because a model will happily voice a claim you did not write. So write the script from claims you can substantiate, then read back what the finished clip actually says.

Read on 2026-10-03: the FTC's Health Products Compliance Guidance, Meta's Health and Wellness standard and TikTok's healthcare ad policy. This is not legal advice.

What the three sources require

The FTC guidance says health claims need competent and reliable scientific evidence and that randomized, controlled human clinical trials are generally what experts would require. It adds that "Results not typical" does not cure a testimonial that reports more dramatic results than users can expect, and that everyone who participates in marketing, agencies included, must check the support for the claims.

Where AI-made video can go wrong

Two risks are specific to generated video. A model with native audio can speak lines you never wrote, and a generated person who says "this changed my life" is a made-up testimonial. Use a script-driven voice or avatar so the words are yours, keep generated people out of experience claims, and never show a generated body transformation as a result.

Read back the spoken words

Sume's video inspect can transcribe a clip at $0.01 per audio minute and returns the text. This script submits the finished ad, waits for the transcript and flags phrases from your own list. A keyword list is a tripwire, not a legal review: a person who knows your substantiation file still approves the final cut. The clip must already be on media.sume.com.

import os, re, time, requests
API = "https://api.sume.com"
H = {"Authorization": f"Bearer {os.environ['SUME_API_KEY']}"}
BANNED = [r"\bcures?\b", r"\btreats?\b", r"\bclinically proven\b",
          r"\blose \d+", r"\bguaranteed\b"]
body = {"video_url": os.environ["CLIP_URL"], "frames": False,
        "transcribe": True, "duration_seconds": 60, "mode": "async"}
r = requests.post(f"{API}/v1/video-inspect", json=body,
                  headers={**H, "Idempotency-Key": "claims-audit-001"})
r.raise_for_status()
rid = r.json()["data"]["request_id"]
for _ in range(40):
    d = requests.get(f"{API}/v1/video-inspect/{rid}", headers=H).json()["data"]
    if d.get("transcript"):
        break
    time.sleep(3)
else:
    raise TimeoutError("transcript not ready")
text = d["transcript"]["text"]
print(text)
print("review needed:", [p for p in BANNED if re.search(p, text, re.I)] or "none")

Keep the transcript with the approved claim sheet. For the labels on the AI parts, see the disclosure checklist; for transcript review on bought creator clips, see reviewing a clip transcript before reuse.