Starter
$99$199 per month, billed yearly $1,188per month, monthly
3 reviewer seats, 100,000 items a month through AI scoring.
Choose StarterContent moderation process from your policy to the weekly report
Every client gets the same six step process. Your policy decides, AI takes the clear cases, trained people take the rest, and every decision leaves a record you can show to users, app stores and regulators.
Queue 4
Profiles, Spanish
Reports, English
Listings, Portuguese
Lucía, 31
Madrid, joined today
Nurse, love the sea and salsa nights.
Not here much, write me on Telegram @lu_31
Policy
Romance scam pattern
Section 4.2 Moving users off the platform to ask for money
AI score
0.71
Audit log
09:40MD-48210 AI 0.55 → human review → Approved, reviewer 022
09:39MD-48209 AI 0.97 → removed under your threshold
09:38MD-48207 AI 0.71 → human review → Removed, reviewer 014
Six steps under one SLA
The process below runs for managed moderation and inside the moderation software when your own team does the reviewing. The difference is only who sits in the queue.
- I. Policy
- II. AI pre filter
- III. Human review
- IV. Escalation
- V. QA
- VI. Reporting
I. Your policy becomes a decision tree
You send your community guidelines, terms of service and any internal notes your team already uses. We turn them into a decision tree per content type and per market: what is removed, what is restricted, what is allowed, and what goes to a senior reviewer.
Grey areas get written examples, because two moderators reading the same sentence in a policy can decide differently. The examples are what keeps decisions consistent across shifts and languages.
You approve the decision tree before day one. After launch, policy changes go through a short change note, so every moderator works from the same version on the same day.
- Rules per content type: text, chat, images, video, livestream, profiles, listings
- Market specific rules where local law differs
- Severity levels that decide remove, restrict, warn or allow
- Escalation triggers written down before launch
II. AI pre filter on every item
Every item that enters the queue gets an AI score per policy category. Two thresholds decide what happens next. Above the upper threshold, a clear violation closes automatically. Below the lower threshold, a clearly safe item passes. Everything in between goes to a person.
You approve the thresholds and can move them. Spam, duplicates and obvious violations stop costing reviewer time, and people only see what needs judgement. More on what models do well and where they fail on AI content moderation.
- Scores per category, not one generic risk number
- Thresholds you approve per content type
- Automatic decisions logged with the score and the rule that applied
- Sampling of automatic decisions in QA, so the model is checked too
III. Human review in the user's language
Trained moderators review the items AI cannot decide. They work in the language of the content, with the policy section and the decision tree next to the item, and they choose a decision and a reason from your policy, not free text.
Graphic images and video open blurred and in grayscale. The moderator decides when to reveal more. Queues are split by risk, so a reviewer on profile photos is not suddenly looking at violent media.
- Native level reviewers per language
- Specialist queues for fraud, dating, gaming and graphic content
- Decision and policy reason recorded on every item
- SLA timers visible in the queue
IV. Escalation that is written down before launch
Some items cannot wait and some are not ours to decide. Threats to life, child safety signals, credible violence, legal requests and high profile accounts go to senior reviewers within minutes, and to your team when your policy says so.
Every escalation path names who receives the case, how fast, and through which channel. Nothing depends on someone remembering what to do at 3 a.m.
- Senior reviewer tier on every shift
- Client escalation contacts per category
- Legal and law enforcement paths defined with you
- Escalation time tracked in the report
V. QA and calibration every week
Senior reviewers re check a sample of every moderator's decisions each week. Disagreements are discussed in calibration sessions and turned into policy notes and new examples.
QA also samples automatic AI decisions, so a model that drifts is caught by people, not by your users.
- Weekly QA sample per moderator
- Agreement rate per policy category
- Calibration sessions on disagreements
- Policy notes shared with every shift
VI. Reporting and the audit log
A live dashboard shows the queue, decisions and SLA in real time. Every week you receive a report with volume, decisions by policy, automatic versus human decisions, appeals and their outcomes, QA results and SLA performance.
The audit log keeps who decided what, when, under which policy version and why. It is what you show when a user complains, an app store asks or a regulator requests transparency data.
- Live dashboard for your team
- Weekly report by email
- Exportable audit log
- Data for transparency reports under the DSA
Self serve onboarding in 30 days
After you build a plan and create your account, onboarding runs without sales calls.
Week 1
Share your policy and content samples. We draft the decision tree and the escalation matrix.
Week 2
You approve the policy and thresholds. We connect your content through the API or the queue integration.
Week 3
Moderators train on your policy and your samples. A shadow period runs with decisions checked but not applied.
Week 4
Live moderation starts with your coverage and SLA. The first weekly report arrives at the end of the week.
Get your moderation plan in two minutes
Tell us your platform type, content, volume, languages and SLA. The plan shows your team per shift, the AI and human split and a monthly estimate.
Get my moderation plan