So Outside Technology Security CMS
AnQiCMS
Online Demonstration

How to arrange the order of sensitive word filtering and manual review before the collected content goes live

📅 October 26, 2026 👁️ 0

How should the order of sensitive word filtering and manual review be arranged before the collected content goes live? The order should be to first draft, then run sensitive word filtering and keyword replacement, then hand it over for manual review, and finally cut the formal document. The rule layer handles deterministic issues, while the manual layer handles judgmental issues. Let the machine filter first, and people only look at the uncertain parts, so that the amount of rework in the entire chain will be reduced. AnQiCMS collects the source content into a draft state to ensure that this order can be executed conditionally.

Why can't it be published directly?

There are three inherent risks in the content collected: the quality of the source itself is uneven, it may contain outdated or invalid information, and there may be non-compliant statements. If it is released as an official document at one step, these three risks will enter the public page simultaneously, and they will also be included in the sitemap's crawling list. It will take a long time to delete them after the search engine updates the cache, and the cost of rolling back is much higher than that of reviewing first and then releasing.

The draft status is a necessary intermediate layer here. The document status of AnQiCMS is divided into four types: formal document, draft, pending release, and recycle bin. The draft can be used with the preview parameter to generate a preview link for the reviewer to view, without having to expose the content to the public collection first. The output of the collection defaults to this layer, and the risk surface is reduced to the backend.

Which events are executed first in the rule layer?

Content security settings, sensitive word filtering and keyword replacement belong to the rule layer. It is good at dealing with issues with clear lists: prohibited expressions, residual contact information, incorrect writing of competitor names, and promotional language left on the source page. The output of the rule layer is reproducible, and the same word list yields consistent results today and tomorrow, which determines that it can be executed in bulk.

There are two advantages to placing the rule layer in front of manual work in sequence. The first is to save labor: Clean up the dirty words that the machine can judge first, and the editor only needs to look at the remaining part. The second is to provide a stable input for manual work: If the replacement is run first after the person reads it, the replacement action will change the text that has already been confirmed by the person, which means the person is not looking at the final version that will be published.

After the rule layer runs, a hit record should be kept. The hit rate, which words frequently trigger, is the basis for the next round of word library adjustment; without a record, the word library can only rely on feeling to add or delete.

The manual layer is looking at the three things that the rule layer cannot see.

  • Is the fact established: Are the data, version number, price, and time in the source still valid, and should the description be updated?
  • Is the口径 uniform: Is the statement about the same event on this site consistent with the existing articles?
  • Is the structure usable: Can the text paragraph be cited separately, and is the title the sentence that the user might ask?

These three things all depend on judgment, the rule layer cannot do it. The人工 review should give a clear conclusion rather than 'looking at it and changing it', pass and go to the next state, if not pass, specify whether it should be deleted, rewritten, or returned to the source without using it.

The fallback for status flow and accidental deletion

The reviewed manuscripts are scheduled for release at a fixed time, the time factor allows the same batch of collected content to be released in a rhythm, rather than being pushed out all at once on the same day, to avoid a large number of contents flooding into the capture queue at the same time.

The delete operation should be designed in a recoverable way. In AnQiCMS, deleting official documents means moving them to the recycle bin instead of deleting them physically, deleted items can be found in the recycle bin; the value of this step is to let the reviewer dare to make the decision to delete. Before actually emptying the recycle bin, make sure that the backup contains the data and static files of this session.

Common questions

Can we skip the rule layer and view it manually?Small batches are okay, but large batches can trap people in mechanical work; it's more time-saving to run the rules first and then do it manually when the vocabulary already exists.

Can the content imported in bulk be used in the same order?Yes, both compressed packages and spreadsheet imports should first be drafted, then checked according to the vocabulary and manual two paths; the difference is just that fields and images need to be verified first.

What should be done if there is a mistake in replacing the rule layer?Keep the hit records and the original text before replacement, and use the full-site replacement tool for pinpoint corrections, which is faster than re-collecting; every time the vocabulary is changed, a version should be kept for easy rollback.

Related articles

What are the recovery methods available in the background if an article is deleted or modified

First, determine whether it was a deletion error or a modification error. In AnQiCMS, the deletion action is to move the formal document to the recycle bin, not a physical deletion, and it can be restored from the recycle bin; for modification errors, check if there are draft and historical versions of the pending release status. If there are no rollback versions available, there is only the last line of defense, which is the data backup, and the backup range should include static files, and it should be verified regularly whether it can really be restored.

2026-10-25

Before importing tens of thousands of documents in bulk, you need to confirm which few things

Before importing tens of thousands of documents in bulk, you need to confirm the field mapping, duplicate check, batch rhythm, and rollback path. This article takes the AnQiCMS ZIP and Excel batch import, and document content import interface as an example, explaining the inspection methods before import, and how to recover after import errors using the recycle bin and backup.

2026-10-25

How do the fields and tags correspond between the Chinese and English stations?

When the Chinese and English stations share a set of content models, the field correspondence should follow the principle of 'define structure first, then language': use a custom content model to define bilingual common fields, maintain Chinese-English matching words with a keyword library, the whole page HTML translation is responsible for the final draft, and multi-site management is responsible for switching and isolating. This article provides common errors in field mapping and the implementation methods of label correspondence.

2026-10-25

After AI training crawling and AI search crawling are divided into two categories, how to write the robots file without affecting indexing?

The crawling control has separated the training and search class crawlers: it is possible to block training crawling while retaining search crawling. This article provides the order of judgment for writing robots configuration, common misjudgment scenarios and self-inspection methods, and explains which part of the on-site robots configuration, llms.txt and link push modules are responsible for, in order to avoid one-size-fits-all blocking of the entire station, leading to a drop in both indexing and citation.

2026-10-25

The revision needs to batch replace keywords and links, how to replace the entire site without damaging the text.

When replacing keywords or links in the whole site, first lock the word table and target links in the keyword library, and then use the AnQiCMS site-wide replacement tool to execute in batches by range; entrust the keyword links in the text to anchor text management for unified density control, and coordinate with 301 redirect to retain the existing weight of the changed URL part. Backup before execution and run a small sample test to have a rollback path.

2026-10-25

Can deleted documents be recovered, and will the draft preview link be seized?

Deleted documents are usually recoverable: Deleting formal documents in AnQiCMS moves them to the recycle bin instead of deleting them physically, and they can be restored from there; if the recycle bin is cleared, data backups are the only way to recover.

2026-10-25

Moving thousands of historical posts, compressed package import and table import each handle what kind of data

When moving historical content, articles with text and images are suitable for creating ZIP compressed packages for batch import, and field-based directory content is suitable for Excel table import; before importing, align the column names with the custom fields of the custom content model, and use the document content import interface with title-based duplicate checking when dealing with large amounts of data or when resuming at a breakpoint, to avoid writing the same topic twice.

2026-10-25

Import in bulk using both zip packages and tables. When images and fields do not match, check which side first

When importing in bulk, check the table side first and then the image side: if fields do not match, it is mostly because the column names and custom fields are not aligned, and if images are missing, it is mostly because the file paths in the zip package do not match the names referenced in the table. AnQiCMS supports ZIP package and Excel imports, the import interface can be queried by title to check if it already exists, and it should be used to judge before repeating the execution

2026-10-26