GoBuildWP

Content use case

Moderate WordPress comments with AI, review the queue before you act.

Anti-spam plugins catch the obvious junk. What is left in the queue needs reading: real questions, borderline links, a customer who has waited a week for an answer. GoBuildWP Free lets Claude read that queue, propose moderation decisions and act on what you approve. Review the recorded change and its recovery support; permanent deletion is a separate destructive operation.

GoBuildWP Free, Works in Claude Desktop, Runs under Production Safe, Pricing

Illustrative ability sequence for GoBuildWP Free 1.18.6. IDs, results and recovery outcomes are examples, not a recorded client or WordPress test. Select Read Only explicitly for the initial reads; Production Safe is the default. Connection management requires manage_options on a single site or superadmin on multisite.

The problem

The queue needs judgement, not another filter

A busy blog's comment queue is mostly spam with real people buried in it. The spam is easy. The work is in the rest: a genuine question with a link in it, a complaint that deserves a considered reply, a regular reader caught by an overeager filter.

That is reading work, and an AI client does it quickly when it can see the queue. GoBuildWP Free exposes comments as typed abilities: Claude lists what is held, sorts it with reasons, takes only the actions you agree to, and keeps the previous status of every comment it touches.

What Claude Desktop can do here

  • List comments by status, post, author, type or search term, newest first
  • Approve, hold, mark as spam, unspam, trash and restore, each one reversible
  • Draft replies for you to approve, then post them as the connected account
  • Edit a comment's text or display name when a commenter asks for a correction
  • Summarise the questions that keep coming up on a post

What it cannot do

  • Keep commenter email and IP address away from your AI provider by default: list-comments returns them to accounts that can moderate, which the connected account normally can, so ask for only the fields the review needs
  • Delete a comment permanently without explicit confirmation. Trash and spam are the reversible routes
  • Post a reply on a post whose comments are closed
  • Recall a reply people have already read. Rollback restores the site, not their notifications

The prompt

The prompt to paste into Claude Desktop

Three groups and one rule: nothing is deleted, and no reply posts until you have read it.

  • Spam instead of delete keeps every action reversible, which is the whole point of the ledger.
  • "Needs me" is where the value is: complaints, legal questions and anything personal stay with a person.
  • search-content lets Claude link a reply to a page that exists, not one it imagined.
Paste into Claude DesktopEdit the URLs and brackets
Go through the comments held for moderation on my site.
1. List held comments, newest first, with the post title for each.
2. Sort them into three groups: spam, approve, and needs me. Tell me why for anything in "needs me".
3. Show me the proposed classification with comment IDs and wait for approval before any status write. Return only the personal data needed for that review.
4. Apply only the status changes I approve. Use spam or hold where appropriate; do not delete anything permanently.
5. For approved comments that ask a question, draft a short reply in our voice: friendly, no sales pitch, point to the right page if one exists. Show me the drafts first.
6. Post only the replies I approve.

Under the prompt

The abilities Claude Desktop will call

Registered GoBuildWP abilities, each with its own schema and permission check. The live ability list on your site is the final word; the tool map lists every group.

  • list-comments

    Filters by post, author, type, search and status; hides email and IP from non-moderators

    ReadFree
  • get-comment

    One comment with its context

    ReadFree
  • moderate-comment

    Approve, hold, spam, unspam, trash or restore, recorded with the previous status

    WriteFree
  • create-comment

    Posts a reply as the connected account

    WriteFree
  • update-comment

    Edits a comment's text or author display name

    WriteFree
  • search-content

    Finds the page a reply should point to

    ReadFree
  • delete-comment

    Permanent delete that bypasses the trash

    Needs confirmationFree
  • list-changes

    The session's moderation actions, newest first

    ReadFree

Safety and recovery

An administrator-level account, narrowed by its token

Safety profile
Moderation requires moderate_comments in addition to GoBuildWP's connection requirement: manage_options on a single site or superadmin on multisite. An ordinary Editor or moderator does not meet the connection requirement. Narrow the connection with its token scope and profile ceiling, select Read Only for inspection, and review the proposal before a supported moderation write. That review lives in the prompt; moderation writes are not confirmation-gated, so only the Pro approval queue in its all-writes mode makes the server hold them for a person.
Change ledger
Every moderation action is recorded with the comment ID and its previous status, so the ledger shows exactly what the queue looked like before the session.
Rollback
Moderation is reversible by design: rollback-change or a plain unspam puts a comment back. A posted reply can be trashed, but anyone notified has already seen it.
Change ledgerExample entries
  • #4410Spam, comment 5521Rollback available
  • #4411Approve, comment 5530Rollback available
  • #4412Reply to comment 5530Posted

Worth knowing before the first run

  • Comments are untrusted text, spam links included. An instruction hidden in a comment is data to Claude, but Claude can still be misled by one. The profile, the token's scope and any approval you have switched on decide what can run.
  • create-comment posts immediately. Keep the approval step in the prompt.
  • Whatever Claude reads is sent to Anthropic under your account's terms, including personal details in comments.
  • Keep your anti-spam plugin. This pass is for what gets through or gets held, not a replacement for the first filter.

How profiles, confirmations and rollback fit together: safety profiles and the change ledger.

Setup

Set up Claude Desktop as a moderator

Use a dedicated identity meeting both connection and moderation capability checks. Narrow its credential scope and profile; an ordinary moderator/Editor alone does not meet GoBuildWP's connection requirement.

Free

Everything on this page is in GoBuildWP Free: listing, moderation, replies, edits and the change ledger.

Pro

Pro adds WooCommerce product reviews, with their own list and moderate abilities, plus an approval queue for sites with several moderators.

  1. Install GoBuildWP Free

    Comments are part of the free plugin; no Pro licence is needed.

  2. Use an authorized dedicated identity

    The connected user must meet the connection and moderation capability checks. Review manage_options or multisite superadmin authority; an ordinary Editor is insufficient. Keep its credential separate and revocable.

  3. Add the connector

    Claude Desktop: Settings, Connectors, Add custom connector, then paste the URL from GoBuildWP's Connect screen, or use the .mcpb bundle.

  4. Approve in WordPress

    Sign in as that dedicated administrator-level user when the OAuth screen opens.

  5. First pass in Read Only

    Let Claude sort the queue without acting, compare its groups with your own judgement, then switch to Production Safe.

Rules, backlogs and mistakes

Give the AI your comment policy, not just the queue

To moderate WordPress comments with AI consistently, write down the rules a human moderator would follow and paste them above the prompt. Without them, Claude falls back on general politeness and will approve things your community would not, or hold things it would. A short policy is enough: what counts as self-promotion on your site, whether links to competitors are allowed, how you treat heated but civil disagreement, and which topics always go to a person, such as complaints about a purchase, anything mentioning a lawyer, or a reader in distress.

Ask Claude to quote the rule it applied next to each proposed action. That turns the review into a quick check of reasoning rather than a reread of every comment, and it shows you where the policy is vague. Over a few sessions the policy gets sharper and the "needs me" group gets smaller for the right reasons.

Clearing a long backlog

A site that has ignored its queue for months may hold thousands of comments. Do not try to clear that in one conversation. Ask for the oldest page first, filtered by post if one article attracts most of the traffic, and decide whether very old held comments deserve a reply at all; an answer to a question asked two years ago may be welcome or may look odd, and that is your call.

Each moderation is its own call, recorded separately, so a backlog is worked through in batches across several sittings. That is slower than a bulk filter, and deliberately so: every action keeps its previous status in the ledger. Before starting, read the discussion settings with get-site-settings so you know whether new comments are held by default, which changes how big the queue will be next week.

Undoing a moderation pass that went wrong

Suppose a batch marked a regular reader's comments as spam because they included a link. Each status change keeps the comment's previous status, so it can be reversed on its own with rollback-change, or by asking Claude to unspam the listed IDs. If a whole session misread the policy, list-sessions shows that session's actions and undo-session reverses them newest first. It refuses the run if a person has moderated any of those comments since, so your own corrections are not overwritten; undo those items one at a time instead.

Two things sit outside that safety net. A reply posted with create-comment goes live at once and may already have notified the commenter; trashing it removes it from the page, not from their inbox. delete-comment bypasses the trash and cannot be undone, which is why the prompt never uses it. It is destructive, so the call needs a confirmation flag, and with "A person approves" turned on a person must approve that exact call.

FAQ

Questions about how to moderate WordPress comments with AI

Can AI moderate WordPress comments automatically?

It can take every action a moderator can. This workflow asks it to stop for your decision before acting, but that stop is a prompt instruction; the Pro approval queue is what makes the server hold each write for a person. Supported status changes can be reversed and inspected in the ledger; posting replies may send notifications that cannot be recalled.

Is this a replacement for Akismet?

No. Keep your anti-spam plugin as the first filter. The AI pass handles what gets held or slips through: borderline comments, real questions and replies.

Does it work with WooCommerce product reviews?

Product reviews have their own Pro abilities, woocommerce-list-reviews and woocommerce-moderate-review. Ordinary post comments are covered in Free.

Will Claude see commenters' email addresses?

Usually, yes. list-comments includes email and IP for a user who can moderate comments and leaves them out otherwise, and the administrator-level account a GoBuildWP connection needs can normally moderate. Whatever Claude reads goes to Anthropic, so decide whether you need those fields and say so in the prompt.

Try it on staging first.

Install the free plugin, connect Claude Desktop, and run the read-only half of this prompt before anything on the site changes.