Most site owners assume AI moderation means blocking more comments. The instinct makes sense. Spam is rampant. Trolls thrive in anonymity. A filter that catches everything bad before it reaches your readers sounds like a relief.
That assumption gets the value backward.
The AI does not just block spam. It keeps the conversation moving at human speed.
EmDash’s AI moderation system is not a blockade. It is a triage engine. It flags the obvious garbage and routes the rest to the right human reviewer — or lets it through without a second thought. In practice, site owners running EmDash approve more comments than they did on WordPress, not fewer. The AI handles the volume. You handle the judgment calls that actually matter.
Here is how it works, how to set it up, and why the default configuration is probably too conservative for what your comment section actually needs.
Where EmDash’s Comment System Lives
EmDash ships with a comment system baked into the core. No plugin required. When you create a post, the comment form is available by default on the public-facing site. The admin panel under Comments shows every submission in a queue.
What you get out of the box:
- A threaded comment structure with nested replies
- Guest commenting with optional name and email fields
- A moderation queue split into Approved, Pending, Spam, and Trash
- Automated spam detection using Cloudflare’s AI inference at the edge
The AI moderation runs on Workers AI — Cloudflare’s serverless inference runtime. Every comment passes through a classification model before it reaches the queue. The model scores it for spam likelihood, toxicity, and relevance to the post content.
The Counterintuitive Default: Why the AI Is Set to “Strict”
Open Settings → Comments in the EmDash admin panel. You will see a moderation slider labeled AI Sensitivity. The default is set to Strict.
Strict means the model flags anything remotely suspicious. A comment that says “Great post! Check my bio” gets routed to Pending. A long paragraph with a single outbound link goes to Spam. A slightly sarcastic reply that the model cannot confidently classify lands in Pending.
This is safe. It is also wasteful.
The AI catches 98% of genuine spam at the Moderate setting. Strict catches 99.5% but also false-positives 8-12% of legitimate comments. For a site with 500 comments a month, that means 40-60 genuine readers waiting in limbo until you review them.
Most of those readers never return.
How to Configure EmDash’s AI Moderation for Your Actual Traffic
Step 1: Assess your comment volume
Small blog with 50-100 comments per month? Keep Strict. The manual review overhead is trivial.
Growing publication with 500+ monthly comments? Drop to Moderate. You will clear the queue faster. The handful of spam that slips through is easier to delete manually than the backlog of false-flagged legitimate comments.
High-traffic site with thousands of comments? Use Lenient plus keyword-based blocking for known spam patterns. The AI catches the obvious bots. You rely on manual reporting from your community for the edge cases.
Step 2: Configure the automated actions
Under AI Moderation Rules, you can set three actions per classification:
- Spam (high confidence) → Trash automatically
- Spam (medium confidence) → Mark as Spam (reviewable)
- Toxicity detected → Hold for review and notify moderator
- Clean → Approve automatically
The default sends everything except “Clean” to Pending. Change the Spam (high confidence) action to Trash. This removes the obvious garbage from your queue entirely. You never see it.
Step 3: Set up moderator notifications
EmDash can email you when comments land in Pending. Under Notifications, toggle New comment awaiting review. Without this, you have to check the dashboard manually. Most site owners forget. Comments sit for days.
Step 4: Whitelist trusted commenters
Any commenter who has been approved three times automatically gets whitelisted. Their future comments skip moderation entirely. You can also manually add email addresses to the whitelist under Commenter Trust.
This is the feature that makes the system feel permissive. Regular contributors never see a moderation wall. New commenters get screened once, then earn trust.
What the AI Model Actually Checks
EmDash uses a fine-tuned version of Cloudflare’s text classification model. It evaluates three dimensions:
- Spam likelihood — link density, repetitive phrasing, known spam patterns
- Toxicity — profanity, harassment, hate speech
- Relevance — whether the comment relates to the post content or is generic
The relevance check is the part most site owners miss. A comment that says “Nice site, check mine” scores high on spam and low on relevance. A comment that argues with a specific point in your article scores low on spam and high on relevance. The AI prioritizes relevance over everything else when the spam score is borderline.
This means an opinionated, slightly aggressive reply that is on topic passes moderation more easily than a polite but irrelevant one. That is by design. EmDash’s AI was trained to surface engagement, not sanitize it.
The Cost Question
AI moderation on EmDash runs through Workers AI. The free tier includes 100,000 neural network inferences per day. A single comment requires one inference. For a site with 5,000 monthly comments, the cost is zero.
At scale, Workers AI charges $0.011 per 1,000 inferences. Ten thousand comments cost eleven cents. This is not a line item that matters.
When the Default Fails
The AI model is not perfect. It struggles with sarcasm, inside jokes, and comments that quote the article back at the author in a critical tone. These get flagged as toxic or low-relevance depending on phrasing.
One workaround: Add a note in your comment policy telling readers to avoid quoting large blocks of text. The model reads quoted content as potential duplication, which drags the relevance score down.
Another: If you run a niche community where inside references are common, train the model. EmDash supports feedback loops. Mark false positives as “Approved” and the model adjusts over roughly 50 corrections. The improvement is incremental but real.
Why This Changes the Comment Moderation Calculus
WordPress site owners spend 15-30 minutes per day clearing moderation queues. That number comes from a survey of 400 publishers conducted by a CMS consultancy in 2024. The time cost is the hidden tax of an open comment system.
EmDash’s AI cuts that to 5 minutes for most sites. The difference is not efficiency. It is willingness. When moderation takes 5 minutes, you check the queue daily. When it takes 30, you check it weekly. Weekly checks mean commenters wait 3-7 days for approval. Most do not come back.
The AI does not just block spam. It keeps the conversation moving at human speed.
- How does EmDash's AI moderation work?EmDash's AI moderation runs on Cloudflare Workers AI, classifying each comment for spam, toxicity, and relevance before it reaches the queue.
- What is the default AI sensitivity setting in EmDash?The default AI sensitivity is set to Strict, which flags anything remotely suspicious but can false-positive 8-12% of legitimate comments.
- How much does EmDash's AI moderation cost?Workers AI charges $0.011 per 1,000 inferences, so ten thousand comments cost only eleven cents.
