AI Content Audit

AI Content Audit scores every published page for Redundant, Outdated, and
Trivial (ROT) content, while separately identifying exact duplicate sets. No
content is changed until a site administrator approves each proposed action.

Features

• Per-page ROT scoring — every scanned page receives age, traffic, priority,
  similarity, and ROT classification data, including pages that do not belong
  to a duplicate set.
• Age and staleness reporting — the ROT report stores creation-based node
  age and last-updated age separately from the normalized 0–1 staleness score
  used for priority, so migration timestamps do not hide legacy content.
• Similarity evidence without transitive clusters — semantic search can
  identify near-duplicate pages for per-page ROT review, but related pages are
  not unioned into inferred duplicate groups.
• Exact duplicate detection — groups nodes only when their normalized title
  and full editable content match exactly. Shared topics, similar titles, and
  repeated campaign language do not create clusters.
• Duplicate prioritization — combines traffic (from uploaded GA4 CSV data)
  and age into a priority score for each exact duplicate set.
• LLM-generated summaries — optionally asks an AI model to summarize why
  each cluster is flagged and what action it recommends.
• Automatic draft edit proposals — optionally asks an AI model to draft
  updates on canonical pages for merge/refresh clusters and saves them as
  approval-gated revisions.
• Action queue with approval gate — proposed actions (prune, merge,
  retire, merge, flag for review, proposed edits) are queued for administrator
  approval before any node is touched.
• Recommendation workflow — the ROT Scores view can be sorted by
  recommendation, and each recommendation links to a ROT detail page with a
  review decision or an Action Queue path.
• Draft edit proposals — an AI agent can save proposed text edits as a
  draft revision so admins can review a diff and approve or reject.
• Incremental batch scanning — large sites are scanned in configurable
  batches, resuming across cron runs.
• Automated scheduling — optionally runs a full scan on a configurable
  cron interval.
• AI agent tool integration — exposes a set of structured tools so an AI
  agent (via the AI Agents module) can query clusters, propose actions, and
  read node content programmatically.

Requirements

• Backdrop CMS 1.x
• PHP 8.2 or later
• AI module with at least one
  configured text provider
• AI Search is not required for exact duplicate detection. It may still be
  enabled elsewhere on the site for search and related-content features.

Installation

1. Install and enable the AI module if you want AI summaries or agent tools.
2. Install and enable this module.
3. Visit Admin → Configuration → AI → AI Content Audit to configure.

Configuration

Navigate to Admin → Configuration → AI → AI Content Audit → Settings.

Key settings:

• Content scope — choose which content types are included in the scan.
• Cron scanning — enable automatic scans and set the interval and batch
  size.
• LLM summary model — model used to generate cluster summaries (can be
  left empty to skip summaries).
• Automatic draft edits — optional post-scan AI drafting for canonical
  pages in merge/refresh clusters, with a model choice and per-run cap.
• Individual AI edit suggestions — from an eligible page's ROT detail,
  Suggest AI edit drafts a revision for that page even when it is not a
  duplicate. The draft remains pending until an administrator reviews the
  diff and approves it from the Action Queue.
• Traffic data — upload a GA4 pageviews CSV to factor traffic into duplicate
  prioritization. Nodes with high traffic score lower for pruning.

Usage

Running a Scan

Go to Admin → Configuration → AI → AI Content Audit. Click Scan Now
to start an incremental scan. Progress is shown on the dashboard. Large sites
may require multiple cron runs to complete.

The scan produces a ROT score for every page. It also produces duplicate
sets only for exact matches. Similarity evidence contributes to the page's
ROT classification, but semantically related pages are intentionally left out
of the duplicate/consolidation workflow unless their normalized full content
matches exactly.

If automatic draft edits are enabled, the module can also stage draft
revisions on canonical pages for high-value merge/refresh clusters. These are
saved as edit_proposed items with pending status and still require
administrator approval.

From an individual ROT detail page, Suggest AI edit uses the page's
editable source and ROT/lifecycle context to propose a refresh. It does not
merge pages, choose a canonical page, or publish anything; it creates the
same approval-gated edit_proposed revision used by the duplicate workflow.

Reviewing Clusters

Go to ROT Scores to review every page from the newest scan. Filter by
Redundant, Outdated, Trivial, or Review. Use the Recommendation dropdown
to show only outcomes such as Retire, Refresh, Prune, Merge, or Review; use
Sort by separately to order the filtered results. Every row links to a
detail page. Go to
Duplicates to see exact duplicate sets ordered by priority. Click a
duplicate set to see:

• All member nodes with their individual scores.
• The selected working canonical page and the exact duplicate candidates.
• The LLM-generated summary (if configured).
• Recommended action for each node.

Historical runs remain available from Runs, but the default Clusters
view shows only the newest usable scan. A direct link to an older cluster is
marked historical so old labels are not mistaken for current findings.

Approving Actions

Actions proposed by the scan (or by an AI agent) appear in the Action Queue
tab. This includes pending edit_proposed draft revisions, which can be
reviewed through their diff links before approval. Select pending items and
click Approve selected, then click Apply approved when ready. Approved
actions are then executed by the queue processor.

Possible actions per candidate node:

AI Agent Tools

When used alongside the AI Agents module, AI Content Audit exposes the
following tools for agent use:

All write operations (audit_propose_action, audit_propose_edit) are staged
for approval in the queue — the agent cannot apply changes directly.

Issues

Bugs and feature requests should be reported in the
Issue Queue.

Current Maintainer

Justin Keiser

Credits

• Created for Backdrop CMS by Justin Keiser.
• Developed with AI assistance.

License

This project is GPL v2 software. See the LICENSE.txt file in this directory for
complete text.