Content moderation policy: how to write it, what the DSA requires, and a template
Good moderation starts with a good text. Your moderation policy tells readers what they can post, guides your moderators through the hard cases and now also serves as the instructions for the AI that sorts your comments. Here is how to build it, what the law requires, and a template to get started.
What the DSA requires
The EU Digital Services Act (DSA) requires all providers of intermediary services to include in their terms and conditions information on any restrictions they impose on content provided by their users (Article 14). That information must cover:
- the policies, procedures, measures and tools used for content moderation;
- algorithmic decision-making and human review;
- the rules of procedure of the internal complaint-handling system.
The text must be written in clear, plain and unambiguous language and be publicly available, and users must be informed of any significant change. Finally, moderation must be applied in a diligent, objective and proportionate manner, with due regard to freedom of expression and the freedom and pluralism of the media.
The sections of a content moderation policy
The spirit of the conversation
Why the comment section exists and the tone you expect. Two or three sentences are enough, but they guide every difficult decision.
What is not allowed
Insults and personal attacks, hateful or discriminatory speech, threats, harassment, spam and advertising, scams, sharing personal data, illegal content. Each rule works better with an example.
What is still allowed
Disagreement, sharp criticism of an idea, an article or a public figure, irony that targets no one. Spelling out what is allowed keeps you from stifling debate out of excess caution.
How moderation works
Who moderates, with which tools, and the share of automated and human decisions. The DSA requires you to tell your users exactly this.
Sanctions
Comment removal, temporary suspension, permanent ban: a graduated scale that everyone knows makes each decision easier to accept.
How to appeal
Where and how a user can ask for a decision to be reviewed, and how soon they will get an answer.
Writing rules an AI can apply
An AI reads your policy like a new moderator: it applies what is written, nothing more. A few principles help a great deal.
- Sentences, not word lists. "No insults towards another reader, even disguised ones" covers far more cases than a list of banned words, which lets irony and deliberate misspellings slip through.
- Examples on both sides. For each rule, give one refused example and one similar accepted example. It is the best way to draw the line.
- Context matters. The same word does not carry the same weight under an article about a fundraiser and under a divisive story. Say so, and use a tool that passes the topic of the post to the AI.
- An acknowledged grey zone. Write down which cases should be reviewed by a human rather than decided automatically: harsh criticism of a journalist, an ongoing court case, borderline political speech.
- Instructions for the news cycle. During an election or after a major incident, add a targeted temporary instruction instead of tightening the whole policy.
Content moderation policy template
A template to adapt, here for the comment section of a news website:
Community guidelines
The comments are a space for debate around our articles. You are welcome to disagree, including with our newsroom, as long as you respect the other participants.
Not allowed: insults and personal attacks, including disguised ones; hateful or discriminatory speech; threats and calls for violence; harassment of a participant; advertising, spam and misleading links; publishing other people's personal data; any content that breaks the law.
Allowed: criticism, even sharp, of an idea, an article or a public figure; irony that does not target a specific person; minority opinions expressed without attacks.
How we moderate: comments are analysed by an automated tool that applies these guidelines. Uncertain cases are reviewed by our team. Every removed comment comes with an explanation.
Sanctions: removal of the comment, then temporary suspension of the account for repeat offences, then a permanent ban.
Appealing a decision: you can ask for a decision to be reviewed by replying to the message that notified you of it. We will get back to you as soon as possible.
Keeping the policy alive
A policy is never finished. Moderation decisions show where it is unclear: when our team or yours keeps correcting the AI on the same kind of comment, a rule is missing or an example is badly chosen. In OpenModeration, the system learns from these decisions and suggests adjustments to your rules, which you apply in one click. Every change creates a new version, so you always know which rule applied to each decision.
Source: Regulation (EU) 2022/2065 on a Single Market For Digital Services, Article 14. This page is for information only and does not constitute legal advice.