TikTok bans some AI content even if labeled: lint prompts first

TikTok does not allow AI-generated content that shows fake authoritative sources or crisis events, falsely shows public figures being bullied, endorsing or endorsed, or uses the likeness of anyone under 18 or of an adult private figure without permission. A label does not fix any of these, so check the prompt before you pay for the render.

The list on TikTok's page

TikTok's support page says some AI content can still cause harm even when properly labeled, and lists what it does not allow. The first group is content showing fake authoritative sources or crisis events, or falsely showing public figures in certain contexts, including being bullied, making an endorsement or being endorsed. The second is the likeness of young people under the age of 18, and the likeness of adult private figures used without their permission.

The page adds that TikTok acts on content that violates its Community Guidelines regardless of whether AI was used, including its policies on impersonation, misinformation and hate speech.

Why check before the render, not after

Sume reserves the cost of a video job from your balance when you submit it, at the provider list price times 1.25. A prompt that violates a platform rule is an avoidable spend, because the clip may be unusable for that platform. A text check costs nothing and runs in milliseconds.

A keyword check is not moderation. It cannot understand context, and it will miss a prompt that avoids the obvious words. What it does well is catch the careless case: a template that someone filled in with a headline or a name without thinking.

A small linter you can run in CI

The script below reads prompts from a file, one per line, and prints any that contain words from the four categories. It exits with status 1 if anything matched, so a pipeline can stop before it submits. Extend the word lists with the terms your team actually uses.

import re, sys

RULES = {
    "authority or crisis": r"breaking news|news anchor|press conference|official statement|earthquake|explosion",
    "endorsement": r"endorses|endorsement|recommends this|sponsored by",
    "under 18": r"\b(child|children|kid|kids|teen|teenager|toddler|baby|schoolboy|schoolgirl)\b",
    "real person": r"\b(president|senator|celebrity|ceo of)\b",
}

bad = 0
for n, line in enumerate(open(sys.argv[1]), 1):
    for name, pattern in RULES.items():
        if re.search(pattern, line, re.I):
            print(f"line {n}: {name}: {line.strip()}")
            bad += 1
sys.exit(1 if bad else 0)

What to do with a flagged prompt

Do not reword around the rule. If the idea needs a child, a real person or a newsroom, it needs a different treatment: an illustrated character, a clearly fictional setting, or real footage with real consent. For reference images, only use ones you have the right to use.

Then label what remains. The TikTok switch is covered in the AI-generated content switch post, and the under-18 rule is covered in the AI child character post.

Limits of this check

TikTok's page is a summary, and its Community Guidelines are the full text. The linter only enforces the shape of that summary. Review high-stakes ads with a person.