P0 SEO: điều tra regression homepage branded/navigational từ 2026-07-18

[tuantmdev]

Problem / opportunity

GSC final data confirms a sustained organic-acquisition regression concentrated on the homepage (https://tak-12.com/) and branded/navigational queries. Do not treat this as non-brand SEO decline and do not roll back content blindly. Freeze further broad homepage SEO/CRO edits while isolating the smallest affected mechanism.

Evidence and baseline

• Source / dates: Google Search Console property sc-domain:tak-12.com, final calendar-date windows 2026-07-18..2026-08-14 vs 2026-06-20..2026-07-17. GSC date-boundary timezone is not exposed by the collector; do not blend this with PostHog’s UTC query window.
• Domain: 62 clicks / 36,219 impressions (weighted CTR 0.171%) vs 595 / 68,482 (0.869%): clicks -89.6%, impressions -47.1%.
• Homepage: 49 / 34,684, CTR 0.141%, average position 5.32 vs 589 / 68,042, CTR 0.866%, position 2.77. It accounts for 540 of the 533 net domain-click decrease (page-query totals can differ because GSC suppresses/anonymizes rows).
• Branded navigational query group: 35 / 30,934, CTR 0.113% vs 506 / 58,268, CTR 0.868%. talk12: 5 / 1,225, position 4.49 vs 280 / 2,846, position 1.06; tak12: 18 / 23,174, position 5.43 vs 178 / 47,633, position 2.63.
• Commercial-branded: 1 / 339 (0.295%) vs 5 / 611 (0.818%); non-brand: 4 / 1,474 (0.271%) vs 38 / 4,420 (0.860%). These query-group rows are directional and may not sum to property totals.
• Relevant changes landed on the homepage after the pre-window: PR feat: clarify homepage qualified intent routing #10 merged 2026-07-23; additional homepage commits landed 2026-08-03. Production currently returns HTTP 200; the latest Pages deployment for c86997a succeeded 2026-08-14. Existing P1: Phục hồi qualified branded CTR và intent routing trên homepage #6 / PR feat: clarify homepage qualified intent routing #10 is closed and therefore is not an active owner for this regression.
• Referral and PostHog are intentionally excluded from any organic conversion rate here: referral is aligned to calendar dates but has no page attribution; the monitor’s PostHog aggregate is a rolling UTC window.

Hypothesis

A loss of SERP rank and click preference for the homepage’s branded/navigational query mix—potentially correlated with changes to title/meta/H1, canonical/robots, structured data, internal routing, or the relevant Google result selection—caused the decline. The mechanism is unproven; correlation with commits is not causation.

Proposed change

Perform a fail-closed root-cause pass before any recovery edit: compare exact pre/post homepage HTML and commit diffs, inspect the canonical URL in GSC, segment the affected query/page mix, and verify current production metadata/robots/canonical/schema. If and only if a specific on-page technical or intent regression is evidenced, implement one smallest-surface recovery PR with regression tests. Do not alter independently improving pages or make a broad content rewrite.

Acceptance criteria

• Record equal-window GSC comparison (clicks, impressions, weighted CTR, impression-weighted position) for domain, homepage, affected queries, commercial-branded and non-brand segments; retain raw exports/queries outside the repo secrets boundary.
• Inspect https://tak-12.com/ in GSC and record verdict, coverage/indexing state, user/Google canonical and last crawl; compare crawl time to affected deployments.
• Diff all relevant homepage commits from 2026-07-18 onward for title/meta/H1, canonical/robots, JSON-LD, internal links and primary CTA routing; verify the same markers on production.
• Identify a falsifiable mechanism and a smallest affected surface, or document a concrete external blocker/no-code finding. No blind rollback.
• If a code cause is found: one focused PR with failing-then-passing regression tests, sitemap/fingerprint updates only when content changes, full suite, git diff --check, independent fail-closed review, and Closes #<issue>.
• Post-merge verification (only if a PR exists): merge SHA deployment success, production HTTP 200 + expected markers, and GSC indexing allowed.

Measurement plan

• Primary: homepage branded/navigational clicks, impressions, weighted CTR and position in 14- and 28-day equal pre/post windows after a verified deployment.
• Guardrails: commercial-branded and non-brand homepage/page performance; canonical/indexability; affiliate_cta_click event schema and destination mix. Keep PostHog legacy/cutover events and calendar-date GSC metrics separate; do not infer a conversion rate without matching scope and timezone.
• Baseline: values above. The immediate task is diagnosis; do not declare recovery from the current 7-day sample (8 vs 9 clicks).

Priority rationale

P0 — sustained 14- and 28-day deterioration with a material, homepage-concentrated acquisition loss. Confidence is high that a regression exists, but only medium on cause; a bounded diagnostic-first scope prevents an unsupported rollback. Effort: medium.

[tuantmdev]

Diagnostic checkpoint — 2026-08-18

I completed the fail-closed Search Console and production/commit checks before making any recovery edit.

Verified GSC evidence

• Property access: sc-domain:tak-12.com as siteRestrictedUser.
• Equal windows confirm the issue baseline: domain 595 / 68,482 impressions / 0.869% CTR / 2.80 weighted position (2026-06-20..07-17) versus 62 / 36,219 / 0.171% / 5.42 (2026-07-18..08-14). Homepage: 589 / 68,042 / 0.866% / 2.77 versus 49 / 34,684 / 0.141% / 5.32.
• URL Inspection for https://tak-12.com/: PASS, Submitted and indexed, INDEXING_ALLOWED, user and Google canonical both https://tak-12.com/, robots allowed, successful fetch; last crawl 2026-08-17T20:28:57Z.

Timing / falsification

Daily homepage data already deteriorated on 2026-07-17 (0 clicks, 1,481 impressions, position 5.87) and remained weak on 2026-07-18 (5 / 1,606 / 4.85). The broad title/meta/H1/hero intent rewrite was committed on 2026-07-23, so it cannot be the initiating mechanism. The later 2026-08-03 recovery restored branded/free-trial wording and referral semantics; production currently serves those markers, but the available evidence does not prove recovery or causality.

The closest preceding source change was the 2026-07-16 campaign carousel / campaign keywords / makesOffer addition. It coincides with the changepoint but did not alter canonical or robots, and correlation alone is insufficient to blame or roll it back. Current production canonical and robots are correct.

Decision

No code change deployed for #13. A blind rollback would violate the issue's smallest-surface requirement. Freeze further broad homepage SEO/CRO rewrites while the next measurement window evaluates the already-live 2026-08-03 recovery. The separately open expired-promotion issue #14 remains the next concrete, independently scoped P0 remediation; it should be handled without conflating it with this ranking diagnosis.