Moderation
Comment moderation when it becomes someone’s job
A personal blog moderates comments in the gaps between other work. A newsroom, a multi-author publication or a course community cannot — the volume arrives on a schedule it does not control, and the cost of missing something is public. This is the workflow, the roles, and the written policy that make that survivable.
Short answer
A team moderates comments at volume by putting four things between a submission and the page: an automated classifier that screens every comment before publication, a queue only the borderline cases reach, named roles so each moderator knows which decisions are theirs, and a published comment policy that makes those decisions defensible. Reviewing every comment by hand does not scale, and reviewing none of them is how a comment section becomes a liability.
Three thresholds that turn moderation into work
Comment moderation is not a single problem that scales smoothly. It changes shape at three points, and most teams hit them in this order.
Volume
One post goes wide and the day's comments arrive in an hour. Reading each one in order stops being possible before you notice it has stopped being possible.
More than one author
Writers want to answer their own readers without holding the keys to the whole site, and two people working the same queue need to not undo each other.
Exposure
Once the comments sit under your masthead, what you left up is an editorial decision — and one you may have to explain.
A comment moderation workflow that holds at volume
Five stages, in order. Each one exists to keep work off the stage after it.
1. Screen every comment before it publishes
Score each submission automatically at the moment it is posted. High-confidence spam never reaches the page; everything else continues down the pipeline. This is the only stage that runs on 100% of traffic, so it has to be automatic — and it has to fail closed, holding comments for review rather than publishing them unscreened when scoring is unavailable.
2. Route only the borderline cases to a human
A queue that receives every comment is just a slower comment section. A queue that receives the mid-confidence scores is a review workload proportional to your ambiguity, not your traffic.
3. Give the queue bulk actions and keyboard moves
Moderation at volume is repetitive by nature. Multi-select and single-key approve / reject / spam turn a hundred decisions into a few minutes instead of a few hours, and a deep link to the comment in place settles the ones that need context.
4. Encode the recurring decisions as rules
Anything you decide the same way twice belongs in a rule, not in a person's attention: a restricted-word list, a ban for a commenter who will not stop, trusted status for the regulars who have earned it, and automatic closing of threads old enough that new comments are almost always spam.
5. Keep a record of who decided what
The moment a moderation decision is questioned — by a reader, a writer, or a regulator — the useful artifact is a log showing the action, the moderator, and the time. Retention rules belong here too: decide up front how long removed content is kept.
What changes between a solo blog and a team
| Concern | Solo site | Team |
|---|---|---|
| Who reviews | One person, whenever they get to it. | A rotation, with named roles and no ambiguity about whose call it is. |
| Review volume | Every comment, and that is fine. | Only what an automated screen could not settle — otherwise the queue is the bottleneck. |
| Access | The owner account is the whole story. | Writers moderate their own readers without holding the keys to billing or site settings. |
| Consistency | Whatever the owner decided last time. | A written policy, because two moderators cannot read each other’s mind. |
| Accountability | Rarely questioned. | An audit log, because "who removed this and when" becomes a real question. |
How EchoThread implements each stage
Every comment is scored by our machine-learning classifier at submit time rather than matched against a keyword list. High-confidence spam is filtered out of the public thread, mid-confidence scores route to the queue, and approving something the filter flagged teaches the classifier it was wrong.
The queue is built for sites with thousands of comments: it pages all the way to the end instead of stranding you mid-list, supports multi-select approve / reject / mark-spam, and offers single-key shortcuts on the focused row. "Open comment" deep-links to the exact comment on your page.
Owner and Moderator roles are assigned per seat, and every plan — including the free tier — carries five seats, so adding a second moderator is never the thing that forces an upgrade. Comments posted by you or your moderators publish immediately instead of waiting in your own queue, while your restricted-word rules still apply to staff.
Rules you author yourself cover the recurring cases: a restricted-word list, banning or trusting a commenter per site, and threads that close themselves after a period you choose. An audit log and retention controls are available on the Business tier and above.
One detail that matters more to publications with an international readership than it usually gets credit for: the widget resolves the reader's language per visitor rather than applying one locale you pick for the whole site, so a non-English commenter is not reading an English interface to reach your comment form.
Write the policy before you need it
A moderation strategy that lives in one editor's head produces decisions nobody else can reproduce and nobody can defend. Publish a short comment policy and link it from the comment form. It does not need to be long — it needs to answer, in advance, the questions you will otherwise answer under pressure:
- What gets removed, stated as behavior rather than as topics.
- Whether removal is silent or announced, and whether the commenter is told.
- Who can overturn a decision, and how someone asks.
- How long removed comments are retained before deletion.
- Whether comments close on older posts, and after how long.
Two practical notes. Search engines treat user-generated content as part of your page, so unmoderated spam is a ranking and reputation problem as well as a reader-experience one. And if you operate in the EU, the Digital Services Act sets out statement-of-reasons and complaint-handling duties for content removal by hosting services — worth reading against your own policy rather than after a complaint.
Go deeper on one stage
Each of these covers one part of the workflow above in far more detail than a single page can.
- Building a moderation strategy
The philosophy and guidelines layer that sits above the queue.
- AI-powered comment moderation
How automated screening works, and what to look for in a filter.
- Ethical moderation
Where removal, free expression and consistency actually collide.
- Moderating for legal compliance
Takedowns, liability, and what a defensible process looks like.
- Managing toxic comments
The reactive side: hostility, conflict, and de-escalation.
- Comment policy template
A policy you can copy and adapt instead of drafting from scratch.
Questions teams ask
Should a team pre-moderate every comment or publish first and review after?
Neither on its own. Pre-moderating everything by hand delays the conversation and does not scale past a few dozen comments a day; publishing everything unscreened puts spam and abuse in front of readers. The workable middle is automatic screening on 100% of comments, with only mid-confidence cases held for a human.
How many moderators does a comment section need?
Fewer than most teams expect, once the queue only contains genuine ambiguity. The number to size against is not comments per day but held comments per day — and if that number is close to your total, the screening stage is the thing to fix, not the staffing.
Can writers moderate their own posts without full site access?
Yes. In EchoThread each team seat is assigned an Owner or Moderator role, so a writer can work the queue without reaching billing or site settings. Every plan, including the free tier, includes five seats.
What happens to comments on old posts?
Old threads attract a disproportionate share of spam because they are indexed and unattended. Closing threads automatically after a period you set removes that surface entirely, which is usually a bigger reduction in moderation load than any filter tuning.
Does moderating comments affect SEO?
User-generated content is treated as part of your page, and Google’s spam policies specifically cover comment spam that a site fails to moderate. Screening comments protects the ranking of the article they sit under, not just the reading experience.
What record should we keep of moderation decisions?
At minimum the action, the moderator, and the timestamp, kept long enough to answer a complaint. EchoThread offers an audit log and data-retention controls on the Business tier and above.
Put the workflow behind your comment section
ML screening, a queue built for volume, five seats, and owner-authored rules — on the free tier, without a card.
Create your free account