Content Policy Generator
Answer nine questions about what your AI feature allows and get an acceptable-use policy in Markdown, plain text or HTML, with every unanswered question left as a visible hole.
Unanswered sections are printed into the document as visible holes rather than filled with plausible wording. A policy you cannot enforce is worse than no policy, and boilerplate is how you end up with one.
- Sections answered
- 4
- Sections left blank
- 4
- Words
- 284
- Characters
- 1,618
- Owner
- Effective from
- How this is enforced
- How to report something
- How to appeal a decision
- When this policy changes
The reason acceptable-use policies read like boilerplate is that most of them are boilerplate: copied from a company with different users, a different product and a different enforcement capability, and therefore promising things nobody intends to do. The test for every line you write above is whether you could act on it. If the policy says repeat breaches lead to suspension, someone has to be able to see repeat breaches. If it promises an appeal, someone has to read appeals.
The two sections people skip
Enforcement and appeal. They are the ones that cost money, which is why they are the ones left vague, and vagueness there is exactly what turns a moderation decision into a complaint you cannot answer. Saying “an automated filter blocks X, and a person reviews any block you disagree with, within five working days” is a harder promise than “we reserve the right” — and it is the one that makes the rest of the document credible.
Specific beats broad
A prohibition on “harmful content” cannot be complied with in good faith, because nobody knows in advance what it covers, and it cannot be enforced consistently either. A list of specific things — each one something a reasonable person could check themselves against before pressing send — is shorter to write, easier to defend, and much less likely to be quietly ignored by everyone including you.