Author
When indexed page counts fall, do not begin by submitting every URL again. First identify which pages disappeared and whether parameter URLs, redirects, canonical tags, language variants and the sitemap send conflicting signals. Fixing these inconsistencies can remove technical obstacles; it does not guarantee that Google will index every page or restore rankings.
This case covers a bilingual Next.js and WordPress content website. The implementation and acceptance records are dated September 23, 2026. Counts below describe that release snapshot, before publication of this article, rather than a permanently fixed site inventory.
An HTTP 200 response confirms that a resource is available. It does not establish that the page is unique, useful, indexable or Google’s preferred version. Crawlers also encounter redirects, internal links, canonical declarations, alternate-language references and sitemap entries.
If several addresses expose the same content, Google must select a representative URL. A site’s declared canonical is a signal, not an instruction that overrides every other signal. Conflicting destinations make the intended structure harder to interpret.
The investigation identified several interacting code paths: note listings exposed query-string pagination; paginated responses declared their own canonicals; internal navigation continued distributing parameter addresses; some language-neutral content aliases used temporary application-level redirects; and build checks did not cover all alias redirect rules.
Query parameters and self-canonical pagination are not inherently wrong. The issue was inconsistent URL governance and multiple entry patterns. Nor does a declining indexed-page count prove this was the only cause: intentional duplicate consolidation can reduce counts without harming valuable pages. Content quality, accessibility and the affected URL groups must be checked separately.
Classify the affected addresses before applying redirects or indexing restrictions.
| URL type | Indexing decision | Treatment |
|---|---|---|
| Original articles, service and product pages | Intended for search | Return 200, declare the preferred URL and include it in the sitemap |
| Retired article aliases | Not separate search pages | Redirect once to the equivalent permanent destination |
| Tracking-only variants | Not separate content | Consolidate without losing required campaign attribution |
| Internal search, login and registration | Usually excluded | Use suitable noindex rules and omit from the sitemap |
| Paginated content listings | Depends on purpose | Keep accessible navigation; decide indexability independently |
| Nonexistent page numbers | No content to index | Return a genuine 404 |
Do not block every address containing a question mark. A tracking parameter may leave the content unchanged, while a page number can select an entirely different set of articles. Filtering may also produce a useful landing page or a near-duplicate; it needs a separate decision.
This release moved note pagination from query strings to paths:
/zh/notes/?page=2
/en/notes/?page=2&utm_source=campaignThe corresponding destinations became:
/zh/notes/page/2/
/en/notes/page/2/The old listing URLs received single-hop 301 redirects. Requests carrying only tracking parameters returned to the clean listing homepage:
/en/notes/?page=2 -> 301 /en/notes/page/2/
/zh/notes/?page=2&utm_source=x -> 301 /zh/notes/page/2/
/en/notes/?utm_source=x -> 301 /en/notes/Internal pagination links were changed at the same time. Otherwise, each crawl would continue discovering the retired format even though redirects existed.
The benefit is consistency, not an intrinsic ranking advantage of paths over query strings. Valid query-string pagination can remain a sound implementation. Avoid migrating it merely to make URLs look cleaner. Before stripping tracking parameters, confirm how analytics records attribution; a redirect can remove information before client-side tracking sees it.
The site’s selected configuration returns 200 for valid paginated listings, declares a self-referencing canonical and corresponding language alternates, applies noindex, follow, and excludes the listings from the sitemap. Individual articles remain in the sitemap.
This is a site-specific decision, not a rule that every pagination page should be excluded. Pages in a sequence contain different items and should not all canonicalize to page one. Indexable pagination may be appropriate when it supports a useful archive or category experience.
Noindex controls eligibility for search results; it is not a robots.txt crawl block. However, follow does not guarantee that a search engine will keep revisiting an excluded page or discovering its links indefinitely. Important older articles need additional crawlable links from relevant indexable hubs and contextual pages, plus accurate sitemap entries. Do not make a noindexed listing their only discovery route.
Language annotations on an excluded listing also do not turn it into an eligible search landing page. The important language checks concern the indexable article URLs: correct language, corresponding alternatives and consistent canonical destinations.
A canonical tag alone cannot resolve every routing problem. If internal links promote one address, a redirect points elsewhere and the sitemap lists a third, the site is still sending mixed signals.
The acceptance contract should require:
Not every redirect should be permanent. Login or other language-dependent functional entry points may need temporary routing. The permanent-redirect rule in this case applies to content aliases with a stable replacement.
URL inconsistencies often reappear when a template is copied or a new route is introduced. The remediation therefore included a build-time gate rather than relying only on post-release spot checks.
For the static routes covered by the gate, checks inspect canonical metadata, language metadata, sitemap inclusion, functional-page indexing rules and permanent content-alias redirects. A missing requirement makes the production build fail.
Coverage matters. Passing this gate does not prove that every dynamic article or runtime API response is correct. Dynamic content still needs separate API and public-page checks. A test suite should also record the exact routes it covers so a green result is not mistaken for a complete site audit.
The September 23 remediation record contains three distinct evidence groups.
These are scoped results. The full historical test suite was not entirely green: the record lists 155 passing tests, 39 failures and one skipped test. The remaining failures were recorded separately; this article does not describe the entire repository as passing.
Old page-parameter addresses returned a single 301 to their path equivalents. Valid path pagination returned 200 with the expected canonical, language and robots metadata. Page one returned permanently to the listing homepage, invalid query page values resolved to the clean homepage, and nonexistent high page numbers returned a real 404.
The release snapshot contained 1,011 sitemap URLs, all unique. Its article inventory contained 454 English and 454 Chinese notes with unique slugs within each language. Checks covered 48 historical alias mappings and 96 language-specific redirect requests with no failures.
These numbers document technical acceptance at that point in time. They are not the number of pages Google indexed, and they will change when content is published or retired. They also do not establish a ranking increase.
Search Console describes Google’s processed observations, not a live mirror of the current application. Google must revisit affected addresses, process redirects and canonical relationships, and refresh its reporting.
Keep three states separate:
Compare dates before attributing a result to the release. A crawl recorded before deployment cannot confirm the new implementation. Track valuable preferred URLs separately from retired duplicates; successful consolidation may reduce the latter without increasing total indexed counts.
No. Their purpose and implementation matter. A valid page parameter can identify distinct content. Problems arise when variants unnecessarily duplicate content or when navigation, canonicals and sitemap entries disagree.
No. Choose according to the archive’s purpose and discovery architecture. If a listing is excluded, provide other crawlable routes to important articles. Noindex, follow alone is not a permanent discovery guarantee.
Redirect a retired URL to its permanent equivalent. Declare the intended canonical on the destination. When duplicate variants must remain accessible, canonical tags can express a preference, but their signals should agree with internal links and sitemaps.
No. Blocking crawling may prevent a crawler from reading a page’s canonical or noindex directive. Choose redirects, canonical declarations or indexing controls according to what the URL represents rather than treating robots.txt as a duplicate-merging mechanism.
There is no fixed deadline. Recrawling, processing, content assessment and site conditions vary. Technical acceptance can be dated precisely; indexing and ranking outcomes require later evidence and cannot be guaranteed.
The useful outcome of this release is a more consistent URL system and a regression gate, not a promise of instant recovery. Consolidate retired entrances, align the signals for preferred pages and test the public behavior after deployment. Then monitor which affected URLs Google revisits and how its indexing decisions change, without confusing a completed code fix with a completed search recovery.
No comments yet. Be the first!