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.<\/p>\n
That assumption gets the value backward.<\/p>\n
EmDash\u2019s 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 \u2014 or lets it through without a second thought. In practice, site owners running EmDash<\/a> approve more<\/em> comments than they did on WordPress, not fewer. The AI handles the volume. You handle the judgment calls that actually matter.<\/p>\n 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.<\/p>\n 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<\/strong> shows every submission in a queue.<\/p>\n What you get out of the box:<\/p>\n The AI moderation runs on Workers AI<\/a> \u2014 Cloudflare\u2019s 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.<\/p>\n Open Settings \u2192 Comments<\/strong> in the EmDash admin panel. You will see a moderation slider labeled AI Sensitivity<\/strong>. The default is set to Strict<\/strong>.<\/p>\n Strict means the model flags anything remotely suspicious. A comment that says \u201cGreat post! Check my bio\u201d 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.<\/p>\n This is safe. It is also wasteful.<\/p>\n 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.<\/p>\n Most of those readers never return.<\/p>\n Small blog with 50-100 comments per month? Keep Strict. The manual review overhead is trivial.<\/p>\n Growing publication with 500+ monthly comments? Drop to Moderate<\/strong>. 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.<\/p>\n High-traffic site with thousands of comments? Use Lenient<\/strong> 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.<\/p>\n Under AI Moderation Rules<\/strong>, you can set three actions per classification:<\/p>\n The default sends everything except \u201cClean\u201d to Pending. Change the Spam (high confidence)<\/strong> action to Trash. This removes the obvious garbage from your queue entirely. You never see it.<\/p>\n EmDash can email you when comments land in Pending. Under Notifications<\/strong>, toggle New comment awaiting review<\/strong>. Without this, you have to check the dashboard manually. Most site owners forget. Comments sit for days.<\/p>\n 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<\/strong>.<\/p>\n 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.<\/p>\n EmDash uses a fine-tuned version of Cloudflare\u2019s text classification model. It evaluates three dimensions:<\/p>\n The relevance check is the part most site owners miss. A comment that says \u201cNice site, check mine\u201d 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.<\/p>\n This means an opinionated, slightly aggressive reply that is on topic<\/em> passes moderation more easily than a polite but irrelevant one. That is by design. EmDash\u2019s AI was trained to surface engagement, not sanitize it.<\/p>\n 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<\/a>,000 monthly comments, the cost is zero.<\/p>\n 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.<\/p>\nWhere EmDash\u2019s Comment System Lives<\/h2>\n
\n
The Counterintuitive Default: Why the AI Is Set to \u201cStrict\u201d<\/h2>\n
How to Configure EmDash\u2019s AI Moderation for Your Actual Traffic<\/h2>\n
Step 1: Assess your comment volume<\/h3>\n
Step 2: Configure the automated actions<\/h3>\n
\n
Step 3: Set up moderator notifications<\/h3>\n
Step 4: Whitelist trusted commenters<\/h3>\n
What the AI Model Actually Checks<\/h2>\n
\n
The Cost Question<\/h2>\n
When the Default Fails<\/h2>\n