Skip to main content

Open in AthenaHQ

app.athenahq.ai/knowledge-base/review

​
Purpose

This page allows users to triage AI-extracted facts (claims) before they are officially added to their Knowledge Base. Athena holds back certain claims for manual review if it cannot independently verify them, such as low-confidence extractions, claims from third-party sources, or comparisons against competitors. Support agents can guide users here to actively manage their brand’s truth by approving valid claims into pillars, rejecting incorrect claims, or deleting facts they don’t want stored.

​
What’s on the page

​
Page header and intro

  • Page title: “Review pending claims” with a “Back to Knowledge Base” link.
  • Intro copy: Explains why claims require human review. If the queue is larger than one page, it displays the current pagination range (e.g., “Showing 1–100 of 250.”).

​
Filter bar

  • Confidence tabs: Filters the queue by Athena’s extraction confidence (All, High, Medium, Low).
  • Source tabs: Filters the queue based on where the claim was found (All sources, Web research, AI responses, Uploads, Your site).
  • Selection toolbar: Appears automatically on the right side of the filter bar when one or more claims are selected. It displays the selected count and provides bulk action buttons (Clear, Approve, Reject, Delete).

​
Pending claims list

The main interface is a stacked card listing all claims on the current page.
  • Select all row: A header bar with a checkbox to select all visible claims. It also shows a count of how many claims on the page are “reviewable” versus “without pillar.”
  • Select-all-matching banner: If a user selects all visible rows on a page but more claims match their current filters on other pages, a banner appears offering to “Select all {N} matching claims”.
  • Claim row: Each individual row represents a single fact awaiting review, alongside its associated metadata and action buttons.
Pending claims list row elements:
  • Select: A checkbox to include the claim in a bulk action.
  • Claim text: The extracted fact or statement.
  • Pillar: A clickable link to the assigned pillar (topic). If the claim isn’t filed yet, this shows an Assign a pillar button instead.
  • Confidence: A badge indicating the AI’s confidence in the extraction (e.g., “High confidence”).
  • Source: A badge labeling the claim’s origin (Web research, AI response, Upload, Your site, Manual).
  • Source link: An external link labeled “Source” that opens the original web page, if available.
  • Source quote: A verbatim snippet of text from the source that supports the claim.
  • Row actions: Quick-action buttons to Approve (✓), Reject (✕), or Delete (🗑) the individual claim.
Shows the current page number and total pages, with Previous and Next buttons to navigate through the queue.

​
Assign a pillar dialog

A modal that appears when a user clicks “Assign a pillar” on an unfiled claim. It contains a searchable dropdown to select an existing published pillar so the claim can be approved.

​
What you can do here

  • Filter claims: Click any Confidence or Source tab to narrow down the queue. Changing filters resets your current selection and returns you to page 1.
  • Select claims: Check individual boxes, use the header checkbox to “Select all visible claims”, or click the banner link to “Select all {N} matching claims” across the entire filtered queue.
  • Approve claims (bulk): Click the Approve button in the selection toolbar. This approves all selected claims into their respective pillars. Tooltips explain why this might be disabled:
    • “Approve works on the selected page only, because each claim is approved into its own pillar.”
    • “Approve {N} claims across {X} pillars”
    • “{N} selected claims can’t be approved without a pillar”
  • Reject claims (bulk): Click Reject in the toolbar to hide selected claims and block them from future re-import. If rejecting the whole queue, it asks for confirmation (“Reject all {N} matching claims?”). A success toast appears with an Undo option.
  • Delete claims (bulk): Click Delete in the toolbar to permanently remove selected claims. It prompts for confirmation, noting that deleted claims might be re-imported by future research runs.
  • Clear selection: Click the Clear button in the toolbar, click “Clear selection” in the banner, or press the Escape key on your keyboard.
  • Approve/Reject/Delete (single): Use the small icon buttons on the right side of any row to take action on just that claim.
  • Assign a pillar: Click Assign a pillar on a row that lacks one. Select a published pillar from the dropdown and click Assign.
  • Navigate pages: Click Previous or Next at the bottom of the page.

​
Data shown

The page displays pending (unreviewed) facts extracted for the active website. The data is pulled directly from the user’s Knowledge Base queue. The “Assign a pillar” dropdown displays published pillars the user has already created for this website.

​
Common workflows

Approving valid claims:
  1. Filter the queue by “High” confidence.
  2. Review the visible claims. If they look accurate, click the header checkbox to select all visible claims.
  3. Click Approve in the selection toolbar.
  4. The claims are moved into their assigned pillars and the next page of claims loads.
Mass rejecting incorrect claims:
  1. Filter the queue by “Low” confidence.
  2. Click the header checkbox to select all visible claims.
  3. A banner appears. Click Select all {N} matching claims.
  4. Click Reject in the selection toolbar.
  5. Confirm the action in the dialog to reject all low-confidence claims at once.
Filing an unassigned claim:
  1. Locate a claim that says “Assign a pillar”. (The Approve button will be disabled).
  2. Click Assign a pillar.
  3. In the dialog, select a relevant published pillar from the dropdown.
  4. Click Assign.
  5. The row updates with the new pillar name, and the Approve button is now enabled.

​
Empty, loading, and error states

  • Empty: If the queue is totally empty, it shows “Nothing to review. All extracted claims have been reviewed.” If filters are active but yield no results, it shows “No pending claims match these filters. Try different filters.”
  • Loading: Displays gray skeleton placeholder rows while fetching the claims list.
  • Error: Actions that fail (like a failed bulk deletion) will trigger a red error toast. If a mass reject/delete operation is interrupted because another user or background process is mutating the same rows, a yellow warning toast explains how many claims were successfully processed and asks the user to try again for the remainder.
  • Linked from: The Knowledge Base main page (usually via a “Review {N} pending” banner/button).
  • Links to: The Knowledge Base main page (via the back button). Clicking a pillar name takes the user to the Pillar detail page (/knowledge-base/{topicId}). Clicking “Source” opens the original third-party web page in a new tab.

​
Common support questions

Why is the Approve button disabled for some claims? A claim cannot be approved until it has a designated home in the Knowledge Base. If a claim lacks a pillar, the user must click “Assign a pillar” to file it before they can approve it. Why can’t I bulk approve all matching claims across the whole queue? The “Approve” action is restricted to the current page’s selection. Because claims are approved into specific individual pillars, the system only processes them in visible batches. Users can mass reject or mass delete across the whole queue, but approvals must be done page-by-page. What is the difference between rejecting and deleting a claim? Rejecting a claim hides it and tells the AI never to import that specific fact again. Deleting a claim removes it entirely, which frees up space, but means the AI might re-import the exact same fact during a future web research run if it finds it again. I accidentally rejected a batch of claims. Can I get them back? Yes. Immediately after rejecting claims, a success toast appears at the top of the screen with an Undo button. Clicking this restores the claims back to the “pending” review queue.