A step-by-step guide to reviewing and managing content that has been flagged for moderation
Review queues are essential tools for reviewing, refining, and approving content flagged by automated moderation. They help you:
• Keep track of flagged items that need attention.
• Decide on the appropriate action for each item (e.g., approve, reject, escalate).
• Provide feedback to the model so it can learn from manual corrections over time.
Keeping your queue tidy—i.e., frequently reviewing and resolving items—ensures you’re addressing all flagged content promptly. This article will show you how content enters a queue, how to navigate charts and the list view, and how to review, resolve, and filter items effectively.
Content is added to a queue when submitted to any of the /moderate endpoints. It shows up in a specific queue if it meets that queue’s configuration criteria. By default, only flagged items appear in the queue. Refer to how to configure queues for more details.
When you open a queue, you’ll first see a chart showing the number of items over time:
• Each bar represents items submitted on a specific day.
• Dark green portions show resolved items.
• Light green portions show items still awaiting review.
Click and drag to zoom in on a specific date range, and the list below will filter to that range.
Selecting a date range using the chart
Below the chart, you’ll find labels and a circular gauge denoting progress toward resolving items with that label. You can click on labels to filter the queue by that label.
Below the charts is the list view, displaying items that need your attention. In the list view, you can:
• Click an item to open its detail view.
• Click the checkmark icon to resolve an item immediately.
• Select multiple items using checkboxes to take bulk actions, such as “resolve” or other custom actions.
The queue is a list of items that need to be reviewed
If you need to resolve multiple items quickly, select the relevant items and click “Resolve.” The queue automatically refreshes, removing those resolved items from view.
The detail view for an item in the queue
Clicking an item in the list opens its detail view, which includes:
• Original content.
• Labels the model applied.
• Whether the item was flagged.
• Metadata submitted with the item.
• Item history (useful for context or auditing).
Any content submitted as HTML is fully rendered in the detail view, including images, videos, and audio. This makes reviewing rich media straightforward.
Metadata that looks like a URL becomes a clickable link. This is especially helpful if you need to reference the content in your own system.
If you disagree with the model’s labeling:
Removing all labels is interpreted as NEUTRAL
, indicating no flag was warranted. Any corrections you make are captured for use in your next adaptive model optimization training. Over time, this feedback loop refines the model’s predictions.
Correcting an AI decision by removing a label and adding a new one
You can also execute actions in the detail view or mark the item as resolved. Once resolved, the detail view automatically navigates to the next pending item in the queue.
In the detail view, use Cmd + Enter to resolve an item, and Cmd + 1-9 to quickly run an action.
Resolving content means you’ve reviewed it and taken the desired action (e.g., approving, rejecting, or escalating). Once resolved, the item is removed from the queue.
You can resolve content in several ways:
Resolving an item only affects that queue, which is important if you have multiple queues. For instance, if you escalate content from the main queue to a senior moderator queue, you would then resolve it in the main queue. However, it remains active for senior moderators to review in the escalation queue.
Use the filter button to narrow down items displayed in the queue. You can refine your view by context ID, author, or labels.
If you submit a context ID with your content, the queue can be filtered to display items from that specific context (e.g., a forum thread), making it easier to see and review conversations in context.
If you submit an author ID, you can click the author ID in the detail view to filter and see all items from that user.
You can also filter by labels. Selecting multiple labels shows all items that match any of those labels.
By default, resolved items are hidden. To see these, check “Show all items” in the filter menu. You can also unresolve items here if you need to review them again.
One of the biggest advantages of review queues is the ability to provide feedback that improves model performance:
By consistently reviewing items and correcting any misclassifications, your moderation processes will improve and become more aligned with your specific needs and guidelines.
A step-by-step guide to reviewing and managing content that has been flagged for moderation
Review queues are essential tools for reviewing, refining, and approving content flagged by automated moderation. They help you:
• Keep track of flagged items that need attention.
• Decide on the appropriate action for each item (e.g., approve, reject, escalate).
• Provide feedback to the model so it can learn from manual corrections over time.
Keeping your queue tidy—i.e., frequently reviewing and resolving items—ensures you’re addressing all flagged content promptly. This article will show you how content enters a queue, how to navigate charts and the list view, and how to review, resolve, and filter items effectively.
Content is added to a queue when submitted to any of the /moderate endpoints. It shows up in a specific queue if it meets that queue’s configuration criteria. By default, only flagged items appear in the queue. Refer to how to configure queues for more details.
When you open a queue, you’ll first see a chart showing the number of items over time:
• Each bar represents items submitted on a specific day.
• Dark green portions show resolved items.
• Light green portions show items still awaiting review.
Click and drag to zoom in on a specific date range, and the list below will filter to that range.
Selecting a date range using the chart
Below the chart, you’ll find labels and a circular gauge denoting progress toward resolving items with that label. You can click on labels to filter the queue by that label.
Below the charts is the list view, displaying items that need your attention. In the list view, you can:
• Click an item to open its detail view.
• Click the checkmark icon to resolve an item immediately.
• Select multiple items using checkboxes to take bulk actions, such as “resolve” or other custom actions.
The queue is a list of items that need to be reviewed
If you need to resolve multiple items quickly, select the relevant items and click “Resolve.” The queue automatically refreshes, removing those resolved items from view.
The detail view for an item in the queue
Clicking an item in the list opens its detail view, which includes:
• Original content.
• Labels the model applied.
• Whether the item was flagged.
• Metadata submitted with the item.
• Item history (useful for context or auditing).
Any content submitted as HTML is fully rendered in the detail view, including images, videos, and audio. This makes reviewing rich media straightforward.
Metadata that looks like a URL becomes a clickable link. This is especially helpful if you need to reference the content in your own system.
If you disagree with the model’s labeling:
Removing all labels is interpreted as NEUTRAL
, indicating no flag was warranted. Any corrections you make are captured for use in your next adaptive model optimization training. Over time, this feedback loop refines the model’s predictions.
Correcting an AI decision by removing a label and adding a new one
You can also execute actions in the detail view or mark the item as resolved. Once resolved, the detail view automatically navigates to the next pending item in the queue.
In the detail view, use Cmd + Enter to resolve an item, and Cmd + 1-9 to quickly run an action.
Resolving content means you’ve reviewed it and taken the desired action (e.g., approving, rejecting, or escalating). Once resolved, the item is removed from the queue.
You can resolve content in several ways:
Resolving an item only affects that queue, which is important if you have multiple queues. For instance, if you escalate content from the main queue to a senior moderator queue, you would then resolve it in the main queue. However, it remains active for senior moderators to review in the escalation queue.
Use the filter button to narrow down items displayed in the queue. You can refine your view by context ID, author, or labels.
If you submit a context ID with your content, the queue can be filtered to display items from that specific context (e.g., a forum thread), making it easier to see and review conversations in context.
If you submit an author ID, you can click the author ID in the detail view to filter and see all items from that user.
You can also filter by labels. Selecting multiple labels shows all items that match any of those labels.
By default, resolved items are hidden. To see these, check “Show all items” in the filter menu. You can also unresolve items here if you need to review them again.
One of the biggest advantages of review queues is the ability to provide feedback that improves model performance:
By consistently reviewing items and correcting any misclassifications, your moderation processes will improve and become more aligned with your specific needs and guidelines.