Content moderation glossary

The words of comment moderation, the DSA and AI moderation, explained simply.

A

AI moderation: AI moderation is the use of artificial intelligence models, such as classifiers and large language models, to understand user content in context and decide or suggest whether it breaks moderation rules.
Appeal: An appeal is the process that lets a user contest a moderation decision, such as a removed comment or a suspended account, and have it reviewed, ideally by someone other than the system that took it.
Astroturfing: Astroturfing is the practice of creating the appearance of spontaneous, grassroots support or opposition, through fake accounts, paid posters or coordinated messages that hide who is really behind them.
Audit trail: A moderation audit trail, or activity log, is a chronological record of moderation actions showing what was done to which content, when, by whom or by which system, and for what reason.
Automated moderation: Automated moderation is the use of software (keyword filters, rules, machine-learning or language models) to analyse user content and take or propose a moderation decision without a person reviewing each item.

B

Ban: A ban is a sanction that prevents a user from posting, commenting or accessing a community, either for a limited period (temporary ban or suspension) or permanently.
Brand safety: Brand safety is the set of practices that protect a brand's reputation by preventing it from being associated with harmful, illegal or offensive content, whether in ads or in the comment sections of its own pages.
Brigading: Brigading is a coordinated action in which a group of users, often organised from another community, descends on a page or post to flood it with comments, votes or reports in order to impose a view or target someone.

C

Comment moderation: Comment moderation is the review of comments posted under articles, videos or social media posts in order to remove or hide abuse, spam and illegal content while keeping legitimate opinions, including criticism, visible.
Community management: Community management is the work of building, animating and moderating an organisation's online community: publishing, answering comments and messages, handling complaints and enforcing the community's rules.
Confidence score: A confidence score is a value, often between 0 and 1, that indicates how certain an AI model is that a piece of content belongs to a category such as spam, insult or hate speech.
Content moderation: Content moderation is the process of reviewing user-generated content (comments, posts, images, videos) and deciding whether to keep, hide, remove or restrict it, based on a platform's rules and the applicable law.
Content moderator: A content moderator is a person who reviews user-generated content, such as comments, posts and images, and decides whether it complies with a service's rules and the law, then takes or confirms the appropriate action.

D

Digital Services Act (DSA): The Digital Services Act (Regulation (EU) 2022/2065) is the European Union regulation that sets rules for intermediary services, including hosting services and online platforms, on handling illegal content, moderation transparency and users' rights.
Disinformation: Disinformation is false or misleading information that is created or shared deliberately to deceive, gain money or influence, whereas misinformation is false information shared without the intention to mislead.
Doxxing: Doxxing is the publication of a person's private or identifying information, such as home address, phone number, workplace or family details, without their consent, usually to intimidate or expose them to harassment.

F

False negative: In moderation, a false negative is content that breaks the rules but is not detected, so it stays visible when it should have been hidden or removed.
False positive: In moderation, a false positive is legitimate content that is wrongly identified as breaking the rules and therefore hidden, removed or flagged when it should have stayed visible.

G

GDPR: The General Data Protection Regulation (Regulation (EU) 2016/679) is the EU law, applicable since 25 May 2018, that governs the processing of personal data, including the names, messages and identifiers handled in comment moderation.

H

Hate speech: Hate speech is content that attacks, demeans or incites violence or discrimination against people because of a protected characteristic such as origin, religion, gender, sexual orientation or disability.
Hide vs delete: Hiding a comment makes it invisible to the public while keeping it (and on Facebook, still visible to its author and their friends); deleting removes it permanently. Hiding is reversible, deleting is not.
Hosting service: Under the Digital Services Act, a hosting service is an intermediary service that stores information provided by, and at the request of, its users; an online platform is a hosting service that also disseminates that information to the public.
Human-in-the-loop: Human-in-the-loop (HITL) describes an automated system in which a person reviews, confirms or corrects some of the machine's decisions, typically the uncertain, sensitive or contested ones.
Hybrid moderation: Hybrid moderation is a moderation model in which software handles clear-cut content automatically and people review the ambiguous or sensitive cases, so each decision is taken by whoever is best placed to take it.

I

Illegal content: Under the Digital Services Act, illegal content is any information that, in itself or in relation to an activity, does not comply with EU law or the law of a member state; the DSA does not itself define what is illegal but refers to those laws.
Internal complaint-handling system: An internal complaint-handling system is the process that online platforms must provide under Article 20 of the Digital Services Act so that users can contest, electronically and free of charge, decisions to remove or restrict their content or suspend their account.

L

LLM (large language model): A large language model (LLM) is an artificial intelligence model trained on very large amounts of text to understand and generate language; in moderation, it can read content in context and apply rules written in plain language.

M

Model arbitration: Model arbitration is a technique in which several AI models analyse the same content and their verdicts are compared, so that a moderation decision on a delicate case does not rest on a single model's judgement.
Moderation policy: A moderation policy, often published as community guidelines or a comment charter, is the set of rules that defines what content is allowed in a space, what is not, and what action is taken when the rules are broken.
Moderation queue: A moderation queue is the list of content waiting for review by a moderator, such as comments held for approval, items flagged by the AI as doubtful, user reports and contested decisions.
Moderation threshold: A moderation threshold is the cut-off value applied to a model's confidence score that decides what happens to content: automatic action above it, human review or no action below it.

N

Notice and action: Notice and action is the mechanism, set out in Article 16 of the Digital Services Act, by which hosting services must let any person or entity report content they consider illegal, and must then process those reports in a timely, diligent, non-arbitrary and objective manner.

O

Online harassment: Online harassment is abusive behaviour directed at a specific person, often repeated or coordinated, that aims to intimidate, humiliate or silence them, through insults, threats, unwanted contact or exposure of personal information.

P

Phishing link: A phishing link is a link to a fraudulent website that imitates a trusted service, such as a bank, delivery company or brand, in order to steal login details, payment information or personal data.
Post-moderation: Post-moderation is a moderation method in which user content is published immediately and reviewed afterwards; content that breaks the rules is hidden or removed once it has been detected.
Pre-moderation: Pre-moderation is a moderation method in which user content is held back and reviewed before it becomes visible to the public; nothing is published until it has been approved.

R

Raid: A raid is a sudden, organised arrival of a large number of outside users in a comment section, live chat or stream, usually to disrupt it with spam, insults or shocking content.
Reactive moderation: Reactive moderation is a moderation method in which content is reviewed only after a user, a trusted flagger or an authority reports it, rather than being checked systematically.

S

Sandbox mode: In moderation, sandbox mode is a test setting in which the system analyses real content and shows what it would decide, without actually hiding or removing anything.
Scam: In moderation, a scam is content designed to deceive people into handing over money, personal data or account access, for example fake giveaways, fake customer service accounts or too-good-to-be-true investment offers.
Sentiment vs toxicity: Sentiment describes the emotional polarity of a message (positive, neutral or negative), while toxicity describes whether it is abusive; a negative comment is not necessarily toxic, and a toxic one can look positive.
Shadowban: A shadowban is a restriction in which a user's content is hidden from others or made less visible without the user being clearly informed, so they keep posting without realising nobody sees it.
Spam: Spam is unsolicited content posted in bulk or repeatedly, usually to advertise, drive traffic or deceive, such as promotional links, copied messages and bot-generated comments.
Statement of reasons: A statement of reasons is the clear and specific explanation that a hosting service must give a user when it restricts their content or account, setting out what was decided, on what facts, on what legal or contractual ground and how to contest it (DSA Article 17).

T

Toxicity: Toxicity is a broad label for rude, disrespectful or aggressive content likely to make other people leave a conversation, including insults, contempt, threats and personal attacks.
Transparency report: A transparency report is a public document in which a service discloses figures and information about its content moderation, such as orders received, notices handled, decisions taken and complaints, as required by Articles 15 and 24 of the Digital Services Act.
Troll: A troll is a person who deliberately posts provocative, insulting or off-topic messages in order to provoke reactions, derail a conversation or annoy other participants.
Trust and safety: Trust and safety (T&S) is the set of policies, teams and tools an organisation uses to protect its users and its service from abuse, fraud, harmful and illegal content, and to make sure people can rely on it.
Trusted flagger: A trusted flagger is an entity with particular expertise, awarded that status by the Digital Services Coordinator of its member state, whose notices of illegal content online platforms must process with priority and without undue delay (DSA Article 22).

U

UGC (user-generated content): User-generated content (UGC) is any content created and published by users of a service rather than by the organisation itself, such as comments, reviews, forum posts, photos and videos.

V

Very large online platform (VLOP): A very large online platform (VLOP) is an online platform with an average of at least 45 million monthly active recipients in the European Union that has been designated as such by the European Commission under Article 33 of the Digital Services Act.

Protect your brand in a few minutes.

Connect your site or your pages and tell us your rules and use cases. We set up the moderation and run it for you: our team handles the doubtful cases and reports, and you see every decision and its reason in real time.

📅 Book a call Try it for free

No credit card required · DSA-compliant · GDPR-compliant

Questions? contact@openmoderation.com

Book a call 60 min · free · no commitment