Skip to content

AI Content Review

Configurable AI-assisted content quality review for Drupal editors. Site builders define Review Rules that map quality criteria to entity types and bundles. Editors trigger reviews from a dedicated Review tab and get scored, explained results with actionable suggestions. Every run is stored as an append-only Review Record.

What it does

  • Review Rules map a list of criteria to an entity type and its bundles. A rule is a config entity, so it ships in your site configuration and moves between environments with it.
  • Criteria are plugins. The one shipped, agent_based, delegates scoring to a Drupal AI Agent, sends it your guidelines and calibration examples, and reads back a structured result through a tool call.
  • Scores and grades. Each criterion returns a score from 0 to 100. Pass and warn thresholds you configure on the criterion turn that score into a pass, warn or fail grade at render time, so changing a threshold reinterprets existing records without re-running anything.
  • Suggestions come back with each result. Text replacement suggestions can be applied to the field directly, accepted or ignored, and the accepted ones build an improvement plan checklist in the edit form sidebar.
  • Review Records are immutable. Each criterion run produces one record holding the score, the explanation, the provider and model used, who triggered it and which revision was reviewed.
  • Bulk and scheduled reviews run in the background through a queue, driven by a bulk operation, a per-rule trigger, or cron.
  • Listings and filters surface review state where editors already work: a summary pill column in the content listing, its own view at /admin/content/ai-review, and filters for outcome and criterion.

How a review runs

flowchart LR
    Editor[Editor or cron] --> Orchestrator[ReviewOrchestrator]
    Orchestrator --> Rule[Review Rule config]
    Orchestrator --> Hash[Input hasher]
    Orchestrator --> Criterion[Review criterion plugin]
    Criterion --> Agent[AI Agent and provider]
    Agent --> Result[CriterionResult]
    Result --> Record[Review Record]
    Result --> Suggestions[Suggestion entities]

The orchestrator is the only code that creates Review Records and Suggestions. A criterion plugin never throws: every error becomes a failed result that is stored like any other.

Requirements

  • Drupal 10.5 or 11.2 and later
  • PHP 8.2 or later
  • AI 1.3 or later
  • AI Agents 1.2 or later
  • Entity API 1.6 or later
  • At least one configured AI provider

See Installation to get started, then Configuration to build your first rule.