Which content-moderation measures are visible to users, and which opaque ones raise concerns about censorship?

Visible: report buttons, redirect links, fact-check labels, sensitive-content warnings, information panels and "are you sure?" prompts. Opaque: content cartels, industry standard-setting, automated moderation, and delisting, downranking or demonetisation.

Visible changes made in the name of safety:

• prominent report buttons and ReDirect links
• fact-check labels and "sensitive content" warnings on Facebook
• YouTube's information panels on Covid content or state media
• Twitter's (X's) pop-up asking "Do you want to retweet this without opening the link first?"

Opaque measures that raise censorship questions:

• Content cartels (Inhaltskartelle, Douek 2020): agreements between platforms to work together on removing content or actors from their services, without adequate oversight.
• other mechanisms for setting industry standards
• automated moderation by algorithms
• delisting, downranking and demonetisation

The visible measures put friction in front of the user, who can see it. The opaque ones are decided between companies or by algorithms, and the affected user often never learns why their reach shrank.

Go deeper:

• Wikipedia: Content moderation — overview of human and automated moderation.
• Gorwa, Binns & Katzenbach (2020): Algorithmic content moderation — open-access paper on the technical and political problems of automating moderation.