Introducing AI Content Review: Your Brand Guide, Applied to Every Drupal Page
AI Content Review is a Drupal module that checks content against criteria your team defines, scores each item from 0 to 100, and tells editors what to change and why. It is in alpha and supports Drupal 10.5 and 11.2 or later.
Most editorial teams already know what good content looks like. It lives in a brand guide, a terminology list, or one senior editor's head. What they lack is a way to check every page against it without that editor reading every page. That is the gap I've been building this module to close.
This article is for content leads and digital managers who own editorial quality on a large Drupal site, and for the site builders who will configure it. One disclosure first: I lead development on this module, so I've tried to be as specific about its limits as about what it does.
Key Takeaways
- AI Content Review scores Drupal content from 0 to 100 against criteria you define, with suggested fixes and the reason for each score.
- Site builders set the rules, thresholds, and calibration examples; editors see scores, pins, and suggestions on the AI Review tab and on the page itself.
- Reviews run one item at a time, in bulk through the cron queue, or on a schedule, and every run stays in the item's review history.
- The module is in alpha with no security coverage yet: pilot it on staging, with your standards written down first.
What Is the AI Content Review Module?
The AI Content Review module checks Drupal content against quality criteria your team defines, then returns a score, a status, and suggested fixes for each criterion. The criteria are yours: tone of voice, grammar, brand terms, reading level, or anything else you can describe to an AI agent.
AI Content Review is not limited to content nodes. A rule can target any entity type, including custom blocks and media, and the agent reviews whatever fields its instructions pass in. It can review file entities but not file data.
What Gap Does AI Content Review Fill in Drupal AI?
AI Content Review gives Drupal AI a dedicated interface for reviewing content the same way every time, across a whole site. The closest tool most teams had for review was AI chat, and chat is the wrong shape for review.
A chatbot's answer depends on what the user types. Two editors asking about the same page in different words get different answers, and an admin can't control what comes back. AI Content Review reverses that: the instructions stay fixed and only the content changes, so results are comparable from one page to the next.
The second gap was Drupal integration. Review results are most useful where editors already work, so AI Content Review builds on the tools Drupal teams use every day: Views, Drupal's built-in list builder, for filtering results by review outcome, plus a review panel on the page itself, revisions, and a review history for every item.
The third gap was time. Content changes after it is published, and standards change too. A review rule can run on a schedule through cron, Drupal's built-in task scheduler. Each pass picks up published items that were edited, or whose criteria changed, since their last review, so nobody has to remember to re-run them.
Who Builds and Maintains AI Content Review?
AI Content Review is a Drupal AI Initiative project built by three organizations. Foster Interactive led the initial design, and Artem Dmitriiev of 1xINTERNET created the project on drupal.org in March 2026. I took it from there, built the backend infrastructure, and now lead its development at Vardot.
The drupal.org project lists all three as supporting organizations, so the module doesn't depend on one agency's roadmap. Vardot committed a full-time contributor when it joined the Drupal AI Initiative as a Gold Sponsor, and that role is mine.
How Do You Set Up AI Content Review?
Setting up AI Content Review is a site builder task, not an editor task. Site builders define the rules and the scoring; editors run reviews and act on the results. The setup has five steps.
- Install the module. Run composer require 'drupal/ai_content_review:^1.0@alpha', then drush en ai_content_review.
It needs AI Agents and a configured AI provider; the project page lists current version requirements.
- Create a review rule. Under Administration > AI > Tools & Automation > AI Content Review, give the rule a label and choose the entity type and bundles it covers, such as a content type. The AI Review tab only appears on entities once at least one rule exists.
- Add criteria to the rule. A rule holds one or more criteria, each with a name and a description, such as "Tone of voice: matches our brand guide." The module ships with one criterion type, agent-based: You could also create a review agent. In AI Agents, give it the "AI Content Review Result" tool, which is what makes it selectable, and instructions that bring in the content with tokens such as [node:title] and [node:body]. One agent can serve every criterion. Each criterion's guidelines tell it what to judge.
- Calibrate the scoring. Set the pass and warning thresholds for each criterion, then add evaluation examples: scored samples, strong and weak, that set the standard before real content is assessed.
- Decide when reviews run. Editors can run a criterion on demand, admins can queue items in bulk, and a rule can repeat on a schedule you enable.
The agent drives most review quality. One agent told to "review everything" is slower and vaguer than several agents with narrow jobs: one for grammar, another for tone, another for terminology.
The Context Control Center (CCC) is a Drupal module for storing brand voice, terminology, and governance rules as reusable context for AI. If your brand guidelines already live there, you don't need to rewrite them.
AI Content Review runs through AI agents, and agents can draw on CCC, so a brand setup you've already done carries into your reviews. Rule-specific instructions sit on top of it.
Each criterion also has an execution mode. Direct mode sends the review as one request; polling mode spreads it across several requests so long reviews don't time out. Polling is driven by the browser: if the reviewer leaves the page, the run pauses.
How Does the 0 to 100 Score Work?
The AI Content Review score is calibrated by your examples, not copied from them. Examples are reference points, not answer keys. Each criterion takes groups of scored examples, at least two per aspect, and the agent sees them as anchors next to the content it is scoring. Content stronger than your best example can still score higher, and content weaker than your worst can score lower.
Examples are optional as well.
Thresholds decide what editors see. The site builder sets the pass mark and the warning band for each criterion, so the same score can mean different things on different sites. Each criterion also returns a severity level and a plain-language explanation.
What Does an Editor See After an AI Content Review Runs?
Editors see AI Content Review results in two places: the AI Review tab on each item and the page itself.
On the AI Review tab, each criterion shows its status, score, grade, and severity, with an Update button to run it again. Every run is kept in the item's review history.
Opening a criterion shows the full explanation, quoting the passages that fell short, followed by recommendations. Where a finding proposes new wording, a side-by-side diff shows the current value against the suggested one.
On the page itself, a score pill next to Edit shows the aggregate score. Selecting it opens a review panel without leaving the page.
The panel lists every suggestion for triage. Each card names the criterion, explains the issue, and shows the suggested change, while a pin marks the exact sentence it targets.
AI Content Review never makes a change on its own; an editor applies, accepts, or ignores each suggestion. If you want that decision enforced by workflow, pair reviews with a content moderation state, as I described in Implementing Human-in-the-Loop AI Review in Drupal Content Moderation.
How Does AI Content Review Handle a Whole Site?
AI Content Review handles a whole site through two things: a site-wide listing for triage and Drupal's bulk operations for running reviews at volume.
The AI content review tab under Content shows one segmented pass, warning, and fail pill per row. It sorts by aggregate score and filters by review outcome, and hovering a pill shows each criterion's score against its thresholds without opening the item.
The same AI review field and outcome filter ship as Views handlers, so you can add them to any content listing your team has already built with Views.
To run reviews at volume, select items, choose the AI content review bulk action, and they join a queue that Drupal's cron scheduler works through at its regular interval.
The queue exists because of timing. Depending not only on the length of the content, but also the model, provider and agent also it could take more or less than 15s and 2mins
That timing points to the run cost, which is easy to miss when evaluating the module. Every review is a set of calls to your AI provider, so the bill scales with items, criteria, and how often scheduled reviews repeat.
Every review is a set of calls to your AI provider, so the cost scales with items, criteria and how much your content changes. As an illustration, bulk-queuing a 300-post blog against four criteria is 1,200 criterion checks. After that, bulk and scheduled passes skip any item whose current revision has already been reviewed against the same criteria, so repeat runs only cost what changed.
Why Does Your Written Standard Matter More Than the AI Model?
My view: the value of AI Content Review comes from consistency, not intelligence. Models keep changing. What makes a review useful is that the same instructions run against every item, every time, and the result is stored where you can compare it.
That's why I don't recommend starting from default rules. Every brand cares about different things. Some teams need a formal tone, and some don't. A default "good content" rule would score everyone against a standard nobody chose.
The consequence is uncomfortable for teams hoping to skip a step. AI Content Review can only be as precise as the standard you write down, and a vague brand guide produces vague scores. The work that decides whether the module helps you happens before you install it.
This matters most where many people publish under one name: a university with departmental editors, a nonprofit with country offices, or a government site with a plain-language mandate. The hard part isn't writing well. It's writing the same way.
Is Your Team Ready to Pilot AI Content Review?
You can decide whether to pilot AI Content Review from five facts you already know about your organization.
- You have a written standard (brand voice, terminology, legal or plain-language rules), ideally already in the Context Control Center (CCC).
- You can name the two to four criteria that matter most, rather than "review everything."
- You have an AI provider your organization has approved for this content.
- You have a staging environment and a site builder who can configure rules and agents.
- You have more content, or more publishers, than one editor can read.
Here is what your answers mean:
- Yes to all five: pilot now on staging with one content type and your top two criteria, and calibrate with real examples from your own site.
- No to 1 or 2: your next step is editorial, not technical. Write and prioritize the standard first.
- No to 3 or 4: resolve the provider approval or the staging setup before installing.
- No to 5: the module adds cost without a matching return for now. Revisit it when publishing volume grows.
What Comes Next for AI Content Review?
AI Content Review will stay in alpha until the maintainers reach a stable release, and there is no fixed date for that. Security advisory coverage applies to stable releases, so that is the release production teams should wait for.
Where Should You Start?
Start on staging, with one content type and the two criteria your team argues about most. Install from the project page and report what breaks in the issue queue; alpha feedback shapes the stable release.
If your team is still deciding which standards to encode and which criteria to start with, Vardot's AI consultancy team can review your readiness with you before anything is configured.