Ten Chinese and English articles were published and their direct URLs worked. Five topics should each have contained one article per language. Yet three posts did not retain their intended categories. The backend also contained similarly named categories with additional language suffixes in their slugs.
This is a first-hand observed case. It occurred during GoogleAsk’s bilingual publication on 2 November 2026. The account concerns content relationships and verification, not authentication details. It demonstrates why a successful HTTP response is an incomplete rebuild acceptance test.
The initial contradiction
Titles, bodies, covers and language paths were correct. Publishing returned success. A URL-only review could easily have declared the work complete. However, the topic sections used the original categories while some posts were assigned to newly copied objects with familiar names.
Post language, category language, translation relationships and archive filters are separate layers. A page can display correct English without belonging to the English topic section that visitors use.
First investigation: inspect the actual assignment
We compared the intended category with the category returned by WordPress for each post. Both advertising editions and the Chinese security article failed this comparison.
Reassigning a post to the original category did not initially resolve it: the returned assignment was converted again. That was a useful clue. The problem was no longer adequately explained by choosing the wrong category once; language-related handling during saving needed investigation. Repeated publication would only add uncertainty.
| Evidence | Supported inference | Not yet established |
|---|---|---|
| Direct URL works; status is published | The post exists | It appears in the intended archive |
| Returned category differs | The assignment changed | The template is broken |
| Reassignment converts again | Inspect language relationships | The multilingual plugin has a vulnerability |
| Similar names, different slugs | Distinguish original and copied objects | Every copy is safe to delete |
Second investigation: a name is not a translation relationship
On the original category edit screens, we inspected language ownership and the paired category in the other language. The relevant relationships were not fully aligned. Chinese and English names that look like translations do not themselves establish a software relationship.
This explained why repeatedly changing the post alone did not help. The original section still queried the original category, while the saving workflow handled an incompatible or incomplete category relationship. We needed to repair those relationships before assigning the articles again.
Polylang’s translation documentation treats categories and content as relationships to configure. This incident is not a claim that all Polylang sites duplicate categories; behaviour depends on configuration, existing data and the saving workflow.
The repair: categories before posts
We retained the original categories used by the sections, explicitly set the Chinese and English language ownership for advertising and security, and established their translation pairs. We then reassigned all four related language editions and read the results back. The returned assignments now matched the intended categories.
Only afterwards did we remove empty copies created during this import, checking that they had no current post references. We did not treat unrelated categories or media as cleanup targets. Reversing that order could send posts to a default category and obscure the original cause.
What verified the result
Verification had three layers: data, language and entry points. Each of the ten posts had exactly its intended category. Each edition retained the correct language URL and corresponding translation. Each language’s guide list contained its five expected articles.
All ten language-specific categories ultimately contained one new article, and both guide lists exposed the correct links. That establishes the repaired assignment and discovery path. It does not demonstrate a ranking, indexing or enquiry improvement.
A repeatable diagnostic order
- Check publication status, visibility and the direct URL before assuming a cache problem.
- Identify the category used by the intended section and compare its ID with the post assignment.
- Inspect post language, category language and translation pairing.
- Once relationships are correct, investigate filters, pagination and caches.
- Open the article through the real section, switch language and return to the corresponding list.
A custom topic page may query fixed category IDs. Creating a new category with the same name does not update that query. A rebuild handover should therefore preserve a small relationship register: entry point, original category, language pair and example article.
The wider rebuild lesson
A rebuild moves relationships as well as visible pages. Categories, translations, menus, downloads and enquiry context determine whether visitors can complete a journey. Use the website rebuild delivery scope to turn those relationships into acceptance checks.
After repairing discovery, review whether each article serves a distinct task. The overlapping-content diagnosis case addresses that next problem: correct structure cannot compensate for repetitive content.
Questions after the case
Would renaming the copied category fix it?
Not necessarily. Identical labels can still represent different objects with different language relationships and archive filters.
Is every missing article a category problem?
No. Scheduling, permissions, filtering and caching can also matter. Here the returned category changes narrowed the investigation. Do not skip that evidence.
