Human review and content moderation
People who speak your users' languages check model output and user content against your policy, in a queue or in real time. Clear calls are made quickly, unclear ones go to a senior reviewer, and every decision is logged.
Product review, EnglishGreat blender, works exactly as described.
ApprovedProduct review, SpanishLlámame al 612 345 678 y te hago descuento.
Removed: phone numberAssistant reply, FrenchCe produit est garanti à vie.
Escalated: unsupported claimComment, Arabicمنتج ممتاز، وصل بسرعة.
Approved
Example decisions against a store's policy.
- We review
- Model output, reviews, comments, listings and chats
- Against
- Your policy, with examples for every decision type
- Coverage
- Hours and turnaround agreed per queue, across UK, US and India time zones
- Reporting
- Decision logs, accuracy checks and policy questions every week
Human-in-the-loop review
Human-in-the-loop review puts a person between an AI system and the people it serves, to check answers the model is unsure of or that carry risk. Content moderation applies the same idea to what users post: reviews, comments, listings and messages are checked against your rules before or after they go live.
Automated filters handle volume. People handle meaning, which matters most with sarcasm, slang and languages your filters were never trained on.
Review work we take on
From a daily batch of flagged items to a staffed live queue.
Model output review
Assistant replies checked before they're sent, or sampled afterwards, for accuracy, policy and tone in each language.
User content moderation
Reviews, comments, listings, profiles and chat messages approved, removed or escalated against your rules.
Live queues
Reviewers working your queue during agreed hours, with turnaround targets and a named lead per shift.
Escalations and appeals
Hard cases and user appeals reviewed by a senior reviewer, with the reasoning written down.
Policy examples
Help turning your policy into reviewer guidance, with examples for each decision type and each market.
Checks on automated filters
A sample of your classifier's decisions reviewed by people, so you can see where it's too strict or too loose in each language.
Service levels you set
Agreed per queue, written into the contract, and reported against every week.
- HoursBusiness hours, extended hours or round the clock. Our offices in London, Florida and Gurugram cover UK, US and India time zones.
- TurnaroundA target time from item arriving to decision, for each queue and priority level.
- AccuracyA share of decisions re-checked by a senior reviewer, with the results in the weekly report.
- EscalationWho handles what, how fast, and when your team gets involved.
- LanguagesThe languages and varieties each queue covers, with backup reviewers for each.
How a review queue starts
A pilot on real items comes before any live traffic.
Policy
You share the policy, decision types and examples. We list the cases it doesn't cover yet.
Training
Reviewers learn the policy and pass a test set before they see your queue.
Pilot
We review a sample of real items, compare decisions with yours, and close the gaps.
Your sign-off QALive review
The queue runs to the agreed hours and targets, with a senior reviewer on every shift.
Reporting
Weekly decision counts, accuracy checks, turnaround and new cases the policy should cover.
Looking after reviewers
Moderation means seeing content most people never have to. Reviewers who are looked after make better decisions.
- RotationTime on distressing queues is capped, and reviewers move between queues during a shift.
- ToolsImages and video can be blurred or shown in grayscale until a reviewer chooses to look.
- SupportReviewers can step away from an item and raise concerns with their lead at any point.
Reviewer welfare policy to confirm
Related services
Questions about human review
What is content moderation?
Content moderation is checking what users post, such as reviews, comments, listings and messages, against a set of rules, and approving, removing or escalating each item. It can happen before content goes live or after it's reported.
What is human-in-the-loop AI?
It's an AI system with a person in the decision path. The person checks answers the model is unsure about, high-risk outputs, or a sample of everything, and their decisions can also be used to improve the model.
Can your team review model output in real time?
Yes. Reviewers can work a live queue during agreed hours, checking items as they arrive, with a turnaround target per queue and a senior reviewer on every shift.
Which languages can you moderate?
English and major European, Middle Eastern, South Asian and East Asian languages, plus smaller languages confirmed per project. Each language is reviewed by native speakers who understand its slang and context.
What are the turnaround times?
They're set per queue and priority level when we agree the service levels, then written into the contract. You get a weekly report showing how each queue performed against its targets.
How do you protect moderators?
Time on distressing queues is capped and rotated, images and video can be blurred until the reviewer chooses to look, and reviewers can step away and raise concerns with their lead at any time.
Start with a pilot queue
Send your policy and a sample of items, or book a call. We review a pilot set, compare decisions with yours and propose service levels.
- Phone
- +44 20 3290 0116
- Phone hours
- Weekdays 9:30 to 16:00, weekends 9:30 to 12:00, UK time