How Do I Moderate WordPress Comments Automatically With AI?
You can moderate WordPress comments automatically with AI using Krom Automation, a free WordPress plugin. Connect the Comment Submitted trigger to the AI Content Moderation action, set your sensitivity threshold, and every incoming comment gets scored for spam and toxicity before it touches your moderation queue.
The free plugin covers this entirely. No Pro plan required.
WordPress has no built-in AI moderation. Its native comment controls let you hold all comments for manual review, require prior approval, or rely on Akismet’s pattern-matching filter.
None of those use a language model. They catch known spam signatures but miss nuanced toxicity, off-topic attacks and low-quality noise that looks grammatically clean.
Krom Automation takes a different approach. You bring your own API key from OpenAI, Google Gemini or Groq, connect it once, and the plugin calls the model on every comment that arrives. The result determines whether the comment is approved, held for review or moved to trash, automatically, with a full execution log so you can audit every decision.
Browse the full feature list to see everything the free plugin includes before you install it.
Is Akismet Still Good Enough in 2026?
Akismet is a signature-based filter. It compares incoming comments against a database of known spam patterns and blocks matches.
That works well against bot-submitted garbage, and for many small blogs it is sufficient. The gap appears when spam is new, when it looks legitimate on the surface, or when the problem is toxicity rather than spam.
A comment that reads “Great post, but the author clearly has no idea what they’re talking about” is not spam. Akismet passes it.
A language model can flag it as hostile and hold it for human review before it goes live. That is a genuinely different capability, not a better version of the same thing.
The practical answer is that Akismet and AI moderation solve different problems. Running both together is the configuration that covers the most ground. Akismet handles volume bot attacks cheaply.
AI moderation handles nuance. Sites with active comment sections, community forums or product reviews benefit most from adding the AI layer.
Akismet stops bots. AI moderation stops humans behaving badly. You likely need both, and the free tier of each covers most sites.
How the Free AI Content Moderation Action Works
Inside Krom Automation’s visual workflow builder, you connect nodes on a drag-and-drop canvas. The moderation workflow uses three nodes: a trigger, an AI action and a branch.
- Trigger: Comment Submitted. Fires the moment a comment lands, before it is approved or displayed.
- Action: AI Content Moderation. Sends the comment text to your chosen AI provider and returns a verdict: safe, flagged or blocked, along with a confidence score.
- Branch: Yes/No conditional. Routes flagged comments to a hold action, blocked comments to trash, and safe comments to an approve action. You set the thresholds.
The conditions and branching documentation covers how to configure the Yes/No logic for different sensitivity levels. You can run a strict threshold on a children’s site and a looser one on a debate-oriented publication, using the same workflow structure with different values.
Sensitivity is configurable. A score of 0.7 or above holds the comment. A score of 0.9 or above trashes it.
Those numbers are yours to set. The AI does not make binary decisions on its own. It returns a score, and your threshold determines the outcome.
Which AI Providers Can You Use?
Krom Automation supports seven models across three providers in the free version. You connect your own API key in the plugin settings.
wpRigel never charges per AI call and never marks up tokens. You pay your provider directly at their standard rate.
| Provider | Models available (free plugin) | Approximate cost per 1,000 comments |
|---|---|---|
| OpenAI | GPT-4o Mini, GPT-4o | $0.01 to $0.15 depending on model and comment length |
| Google Gemini | Gemini 2.0 Flash, Gemini 2.5 Flash Preview, Gemini 2.5 Pro Preview | $0.01 to $0.10 for most comment volumes |
| Groq | Llama 3.3 70B, Llama 3.1 8B | Very low; Groq’s inference pricing is among the lowest available |
For a blog receiving 500 comments a month, the AI cost using GPT-4o Mini runs well under $1 per month. A high-traffic publisher at 50,000 comments a month should model costs carefully before committing to a premium model. For that volume, Gemini 2.0 Flash or Groq’s Llama 3.1 8B keeps costs in the $5 to $15 range monthly, while GPT-4o at the same volume could reach $75 or more.
The cost modeling matters because no other moderation plugin documentation does it honestly. Choosing the right model for your volume is the most important configuration decision you make.
What Does the Workflow Actually Catch?
Traditional spam filters operate on pattern matching. They look for repeated URLs, known spam domains, or comment text that appeared in their training database.
AI Content Moderation operates on meaning. The model reads the comment as a sentence and scores it against categories you care about.
- Spam detection: Catches low-quality promotional comments, affiliate spam and keyword stuffing that looks grammatically correct and fools signature-based filters.
- Toxicity scoring: Flags hostile, abusive or threatening language even when it uses no known slur and matches no spam pattern.
- Sensitivity detection: Identifies potentially sensitive content appropriate to hold on certain site types, such as health or legal sites where user-generated advice carries risk.
- Off-topic and low-quality signals: Some configurations flag comments that contribute nothing to the discussion, reducing the moderation queue even when a comment is not strictly spam or toxic.
What it does not catch is context it was not given. If a comment uses site-specific jargon that looks hostile out of context, the model may score it incorrectly. The hold outcome exists for exactly that reason.
A held comment goes to your moderation queue rather than being trashed, so a human still reviews edge cases. Only comments scoring above your highest threshold go straight to trash.
Automatic trash should be reserved for the highest-confidence verdicts. Hold everything else and let the queue shrink by 80 percent, not 100 percent.
Setting Up the Workflow: What You Actually Configure
Start with the step-by-step workflow builder guide if this is your first Krom Automation workflow. The moderation setup specifically involves these decisions:
- Choose your AI provider and model. For most sites, GPT-4o Mini or Gemini 2.0 Flash balances accuracy and cost. Connect the API key in Krom Automation settings.
- Set the Comment Submitted trigger. This fires before any comment is displayed. The trigger captures comment content, author name, author email and post ID as merge tags available to every downstream action.
- Configure the AI Content Moderation action. Pass the comment body using the comment content merge tag. Set your sensitivity level: low, medium or high. The action returns a verdict and a numeric score.
- Add a Yes/No branch on the score. Route scores below your hold threshold to an approve action. Route scores between your hold and trash thresholds to an update comment status action set to “pending”. Route scores above your trash threshold to a delete post action targeting the comment.
- Enable run-once per entity enforcement. This prevents a single comment from triggering the workflow twice on sites with caching or retry logic.
- Use the workflow simulator for a dry run. The simulator runs your workflow with zero side effects, so you can confirm the routing logic before it touches live comments.
The merge tags documentation lists every variable the Comment Submitted trigger exposes. Author email is particularly useful if you want to add a second branch that skips moderation for verified registered users with more than five approved comments.
Cost of AI Comment Moderation Versus the Manual Alternative
This is the table nobody publishes. Most plugin pages tell you AI saves time. They rarely show the numbers behind that claim.
| Scenario | Manual moderation cost | AI moderation cost (Krom free + API) | Net saving |
|---|---|---|---|
| Small blog, 200 comments/month | 45 min/month at $50/hr = $37.50 | Plugin free + ~$0.05 API cost | ~$37 per month |
| Active community site, 5,000 comments/month | 18 hrs/month at $50/hr = $900 | Plugin free + ~$1 to $5 API cost | ~$895 to $899 per month |
| High-traffic publisher, 50,000 comments/month | Dedicated moderator or outsourced, $800 to $2,000/month | Plugin free + $5 to $75 API cost depending on model | $725 to $1,995 per month |
The manual cost estimate uses 15 seconds per comment for a trained moderator making a simple approve or trash decision. That is a conservative figure.
Complex community sites where moderators read context and respond to disputes run considerably higher. The AI cost range reflects model choice: Groq Llama 3.1 8B is at the low end, GPT-4o is at the high end.
Even for a small blog, the return is not really about money. It is about whether comment moderation happens at all.
Most site owners on tight schedules let their queue back up for days, which means spam sits visible for hours. Automation eliminates the delay entirely.
Honest Limits: What AI Comment Moderation Gets Wrong
No moderation system is perfect. AI Content Moderation in Krom Automation has specific constraints you should understand before relying on it fully.
- False positives on sarcasm and irony. A comment that uses hostile phrasing in a clearly joking context may score high for toxicity. Set your trash threshold conservatively. Hold is safer than trash for borderline scores.
- Language coverage varies by model. Most models perform best on English. Sites with significant non-English comment traffic should test accuracy on their actual language mix before setting aggressive thresholds.
- WP-Cron timing on low-traffic sites. Krom Automation runs actions via Action Scheduler, which depends on WordPress Cron. On sites that receive fewer than a few visits per hour, WP-Cron can fire late, meaning the moderation action runs minutes after the comment arrives rather than immediately. A real server cron job fixes this. Without it, a flagged comment may sit unapproved rather than visibly published, which is acceptable, but not instant.
- API key dependency. If your API key is invalid, rate-limited or the provider has an outage, the AI action fails. Configure Krom Automation’s failure notification by email and the automatic retry with backoff so you know when this happens.
- Cost at scale without model tuning. If you run GPT-4o on every comment without checking volume first, the monthly API bill can surprise you. Set a budget alert with your provider before enabling on a high-traffic site.
For a broader look at what AI can and cannot do reliably inside WordPress, our post on using AI to automate your WordPress site covers the full picture.
What the Execution Log Gives You
Every comment that passes through the workflow creates a per-step execution record. You can see the comment text that was evaluated, the score returned by the AI, which branch was taken, and the outcome action that ran. This matters for three reasons.
First, it gives you an audit trail if a reader disputes that their comment was removed. You have a timestamped record of the score and the threshold that triggered the outcome. Second, it lets you tune your thresholds over time based on real data rather than guessing.
If you see that 40 percent of your held comments are being manually approved, your hold threshold is probably too aggressive. Third, the analytics dashboard shows a failed execution count and trend chart, so you see immediately if your API key stops working.
The free actions reference documents what each action returns to the execution log, including the AI Content Moderation action’s output fields.
Extending the Workflow Beyond Basic Moderation
Once the core moderation workflow is running, the same trigger and AI action can feed additional steps without creating separate workflows. These extensions all run free.
- Notify the post author when a comment is held. Add a Send Email action on the hold branch so the author knows a comment is waiting for their review, not just sitting in the queue silently.
- Auto-tag approved comments. Use the AI Auto-Tag action after the approve branch to label approved comments by topic, useful for community sites where readers browse by subject.
- Add comment meta on every scored comment. Store the toxicity score as comment meta. This lets you sort or filter comments by score inside WordPress without re-running the AI.
- Send a Slack alert on high-confidence toxic comments. Connect the trash branch to an HTTP Request action pointing to a Slack webhook, or use the Slack and Discord messaging integration if you are on Pro. Your moderation team sees the flagged content in real time without checking the WordPress dashboard.
If your site uses BuddyBoss for community features, the BuddyBoss integration documentation covers how to extend similar moderation logic to activity stream posts and private messages, not just standard WordPress comments.
The real advantage of a workflow tool over a dedicated moderation plugin is that comment moderation becomes one node in a larger response, not the whole system.
Also from wpRigel
Pollify is wpRigel’s poll, survey and quiz plugin, built as native Gutenberg blocks. Creating a poll feels like writing a paragraph in the block editor.
There are no shortcodes and no separate interface to learn. It is the right tool if you want to add structured reader input alongside your comment section.
Commandify is a command palette for the WordPress admin. Press Cmd or Ctrl plus K to jump anywhere, search content, and run admin actions without clicking through menus.
If you moderate comments manually for borderline cases, Commandify cuts the time spent navigating to the comments queue significantly. It is the only command palette with real WooCommerce order, product and customer commands built in.
Verdict: Who Should Set This Up and Who Should Not
Set up AI comment moderation with Krom Automation if your site receives more than 50 comments a month and you currently spend any time on manual moderation. The free plugin plus a Groq or Gemini API key costs almost nothing to run, the workflow takes under an hour to build, and the time saving compounds every month. Community sites, membership sites and WooCommerce stores with active product reviews see the sharpest return.
Do not set this up if your site averages fewer than 10 comments a month. Akismet handles that volume fine, and adding AI infrastructure for a queue you review in two minutes a week adds complexity with no meaningful return.
Also avoid aggressive trash thresholds on sites where community trust matters. Start with hold-only moderation for the first month, review the execution logs, and only move to automatic trash once you have calibrated your thresholds against real comment data.
For most active WordPress sites, this is the automation that pays for itself fastest. Download the free plugin from the WordPress.org plugin directory and build the workflow this week. If you want to review the full plan comparison before going Pro, the free versus Pro feature comparison covers every difference in one place.
See Krom Automation pricing and plan options to find the right plan if you need Pro features across multiple sites.
Frequently Asked Questions
Does WordPress have built-in AI comment moderation?
No. WordPress core includes manual moderation controls and optional Akismet integration, but no AI-powered scoring. A plugin such as Krom Automation is required to add AI moderation via an external language model.
Do I need to bring my own API key for AI comment moderation in Krom Automation?
Yes. You supply an API key from OpenAI, Google Gemini or Groq. Krom Automation calls the model on your behalf and logs the result.
wpRigel does not charge per call and does not mark up token costs. You pay your provider directly at their standard rate.
Will AI moderation block legitimate comments from real readers?
It can produce false positives, particularly on sarcasm, irony or comments using aggressive phrasing in a clearly joking context. The safest configuration uses the hold outcome rather than automatic trash for borderline scores.
Held comments go to your moderation queue rather than disappearing, so a human reviews edge cases. Reserve automatic trash for scores above 0.9 or your equivalent high-confidence threshold.
How much does AI comment moderation cost per month?
Krom Automation is free for this workflow. API costs depend on volume and model choice. A blog receiving 500 comments a month spends under $1 on GPT-4o Mini.
A site at 50,000 comments a month should use Groq Llama or Gemini 2.0 Flash to keep costs in the $5 to $15 range. GPT-4o at that volume can reach $75 or more per month.
Does AI comment moderation work with WooCommerce product reviews?
WooCommerce product reviews use the WordPress comment system internally, so the Comment Submitted trigger fires on them as it does on standard post comments. The AI Content Moderation action evaluates the review text the same way. No additional configuration is required to cover both comment types from one workflow.
What happens if my AI provider has an outage during high comment volume?
The AI action fails and Krom Automation logs it as a failed execution. The plugin sends a failure notification by email and retries automatically with configurable backoff.
The comment is typically left in a pending state during the retry window rather than being auto-approved or auto-trashed. Configure your WordPress default comment setting to hold all comments pending review as a fallback, so nothing publishes unreviewed if retries exhaust.