Community rules

We moderate behaviour, not beliefs. A comment is hidden for what it does: attacking a person or a group, threatening, exposing private information, or spamming. It is never hidden for what it believes. Every side argues under the same rules, in all six lanes.

Public figures (politicians, officials, executives, celebrities) can be criticised harshly and accused of corruption. Claims about institutions and powerful people are allowed, with a label when they're unverified. Threats are never allowed, against anyone.

• Attacking another reader

  Hidden

  Insulting, harassing or demeaning another reader, or a private individual.

  • “You're a waste of oxygen, @Sally”

  Still fine: Disagreeing with what someone wrote, even bluntly. Argue the point, not the person.

• Attacking a group of people

  Hidden

  Dehumanising or blaming people for who they are: religion, ethnicity, nationality, race, sex, sexual orientation, gender identity or disability. That includes coded versions, like conspiracy theories that blame a group.

  • “ALL MEXICANS ARE CROOKS”
  • Blaming a religious group for a “deep state” plot

  Still fine: Criticising institutions, governments or powerful people, and quoting hate in order to criticise it.

• Threats

  Hidden

  Threatening or encouraging violence or self-harm against anyone, public figures included.

  • “They should be hanging from the same rope”
  • “Wipe them off the map”

  Still fine: Figures of speech that aren't about violence, like “crush them at the polls”.

• Private information

  Hidden

  Sharing, or asking for, someone's private information: their address, phone number, workplace or family.

  • “I know where you live… your kids go to…”

• Spam

  Hidden

  Ads, scams, link farms and repetitive off-topic promotion.

  • “Earn $500/day → bit.ly/…”

• Dangerous misinformation

  Hidden

  Claims or instructions likely to cause concrete harm, such as skipping medical treatment, dangerous “cures”, or the wrong voting date, place or eligibility.

  • “Vote on Nov 6 this year, the 3rd is cancelled”

• Unverified claims

  Shown with a label

  Presenting a contested or conspiratorial claim as fact. These stay up, in every lane, with a small “Unverified claim” label.

  • “The poll was rigged by donors”
  • “The CIA ran the coup”

• Rudeness

  Shown; very rude comments are collapsed behind “Show”

  Profanity, insults or contempt aimed at public figures, institutions or ideas, or a generally coarse tone. Harsh criticism of politicians is allowed.

  • “Trump is a whiny loser”
  • “This policy is garbage”

What you'll see

• Labelled: shown normally with a small “Unverified claim” chip, and a link to the fact-check when the story has one.
• Collapsed: “This comment may be offensive. Show”. One click reveals it. Its author always sees it normally.
• Hidden: nobody but its author sees it. The author sees why.
• Waiting for a moderator: when our models aren't sure, a person decides. Until then only the author sees it.

Before you post

If a comment looks like it breaks one of these rules, we ask before posting: Want to rephrase? You can edit it or post it anyway; the normal checks still apply. We only count that we asked, never what you wrote.

Appeals

If your comment was hidden or collapsed, you'll see the reason under it, with an Appeal button (one appeal per comment). A moderator reviews it and either restores it, restores it with a label, or keeps the decision. If it's restored, the hidden comment no longer counts against your account.

How comments are checked

A blocklist and our own moderation model (Laya, which we host) check each comment. When Laya isn't sure, a second model from an outside provider (DeepSeek, through OpenRouter) checks it. If neither is sure, the comment is collapsed, or held for a moderator. See Privacy for what is shared and kept.

We don't judge which side a comment reads as: that analysis is currently off, because our model was not accurate enough and showed a bias (neutral comments mentioning religion read as far-right). Cross-side upvotes all count double. It stays off until the model is retrained and passes a bias test (details on the privacy page).