Moderation Policy

OurManifesto.uk is a public political-speech platform. We intervene in user-submitted text on a single, narrow basis: UK law. We do not police tone, civility, or any party- political line, and we do not down-rank, hide, or rewrite content for being unpopular, controversial, or rude.

What we block

Every string a user submits — comments, manifesto policies, party names and descriptions, profile bios and display names, feedback, and the optional reason field on a flag — is checked against the same five categories before it is saved:

• Child sexual abuse material. Strict liability; no contextual exceptions.
• Racial hatred and other unlawful hatred as defined by the Public Order Act 1986 (Part III & IIIA) and the Racial and Religious Hatred Act 2006 — text that is threatening, abusive or insulting and is intended to (or likely to) stir up hatred against a racial or religious group, or on the grounds of sexual orientation.
• Direct incitement to violence against an identified person or identifiable group.
• Terrorism content as defined by the Terrorism Act 2000 / 2006 — encouragement of acts of terrorism or support for a proscribed organisation.
• Defamation — specific, factually testable claims about a named non-public-figure that, if false, would harm their reputation. Strongly worded opinion and criticism of public figures acting in their public role is not defamation and is not moderated.

What we do not block

• Strong opinion, swearing, sarcasm, anger.
• Sharp criticism of named politicians, parties, governments and policies.
• Minority, fringe, or unpopular policy positions across the political spectrum.
• Criticism of OurManifesto itself, its staff, or its moderators.

If the AI is unsure, the call is escalated to a human admin and the content is held in a pending state until reviewed. We err on the side of letting speech through.

Three-strike suspension

If three of a user's submissions are blocked under the categories above within any rolling 30-day window, their posting permission is suspended pending appeal. Reading the site, downloading their data, and cancelling subscriptions is unaffected. A successful appeal voids the underlying strike and lifts the suspension automatically.

How to appeal

Every block carries a decision ID. Go to Settings → Moderation history to see your decisions and submit an appeal. A human admin will review within 7 days. We will email you (using your account email) when an appeal is resolved.

How the AI works

We use OpenAI's API as a classifier. Each submission is sent — with no identifying metadata about you beyond an internal user ID — and the model returns one of pass, borderline or block with a category and a short rationale. The model's decisions are recorded so admins can audit them and so repeated overturns calibrate future decisions. We do not use your submissions to train any third-party model. See our Privacy Policy for the full data-processing detail.

Transparency

The full list of moderation categories, the verbatim system prompt, and the appeals workflow are documented in the project repository. If you believe a block was wrong, please appeal it — admin overturns directly improve the AI's future calls via a learned-example feedback loop.