blog-performance-monitor

Blog quality monitor for recurring SEO health checks and single-URL visibility incidents. Use when monitoring blog-only performance from Google Search Console, investigating a page whose impressions/c

SKILL.md

GitHub

Blog Performance Monitor

Overview

Monitor blog-only GSC performance, explain drops, correlate them with recent blog backlog/change files, and hand off fixes to blog-edit. This skill is the monitoring and diagnosis layer, not the long-form editing layer.

The monitor must be operational, not merely descriptive. If the same page/query problem appears in repeated runs, escalate to a concrete action instead of writing another "monitor one more run" recommendation.

For incident-style prompts like "this page got fewer impressions than usual" or "impressions dried up out of nowhere", treat the requested URL as the primary subject. First determine whether the URL was redirected, canonicalized, unpublished, noindexed, consolidated, or recently edited before recommending content changes.

Compose existing skills:

  • google-search-console-analysis: impressions, clicks, CTR, position, page/query deltas, cannibalization.
  • seo-content-3-kings-technique: title/H1/first paragraph refresh for pages ranking around positions 5-15.
  • seo-audit: canonical, indexability, internal links, schema, and page quality checks.
  • pagespeed: Core Web Vitals/Lighthouse checks when performance may explain losses.
  • Project-local blog-edit at .claude/skills/blog-edit/SKILL.md: perform blog metadata/content updates after this skill produces the edit brief.

Single-URL Incident Triage

Use this path before the recurring-monitor workflow when the user gives one URL.

  1. Normalize the URL

    • Record the exact URL, slug, canonical destination, and whether production returns 200, 301, 302, 307, 308, 404, or noindex.
    • If the URL redirects, analyze the destination URL as the surviving page and mark the original URL's impression drop as expected unless GSC shows the destination also lost visibility.
  2. Fetch focused GSC evidence

    • Use .claude/skills/gsc-analysis as the data source.
    • Compare the latest complete 14 or 28 days against the prior equal window.
    • Pull page, query, and query+page rows for both the original URL and any redirect/canonical destination.
    • Identify whether loss is from impressions, ranking position, CTR, or URL migration.
  3. Search change clues before editing

    • Read the latest entries in:
      • .claude/skills/blog-changelog.md
      • docs/SEO/maintenance/seo-changes-backlog.md
      • docs/SEO/maintenance/gsc-request-indexing-backlog.md
      • relevant docs/SEO/reports/*.md
    • Search those files for the slug, old URL, destination URL, and top query terms.
    • Inspect recent git history for the slug and blog rendering/redirect code if the markdown backlogs do not explain the change.
  4. Classify the incident

    • intentional-migration: old URL redirects or canonicalizes to a replacement.
    • indexing-regression: URL is noindexed, blocked, non-200, canonicalized incorrectly, or missing from sitemap unexpectedly.
    • content-change-regression: a recent edit changed title, H1, intro, body, schema, internal links, or target intent before the drop.
    • serp-demand-shift: impressions fell while position and destination health are stable.
    • ctr-regression: impressions are stable but clicks/CTR fell.
  5. Pick the next action

    • For intentional migrations, request indexing for the destination if not done, monitor destination queries, and avoid rewriting the old URL.
    • For indexing regressions, fix the technical blocker before content work.
    • For content-change regressions, prepare a narrow blog-edit brief tied to the lost query cluster.
    • For SERP demand shifts, record the finding and avoid unnecessary edits unless the destination has an obvious stale-intent mismatch.

Recurring Cadence

Run twice per week. Default windows:

  • Current: latest complete GSC date minus 13 days through latest complete date.
  • Previous: the immediately preceding 14 days.
  • For explicit incidents, use the user's date, e.g. "after May 2", and normalize before/after windows by days.

Always note data freshness:

  • GSC commonly lags 2-3 days.

Measurement Rules

Be precise about comparison direction. Every table must label metrics as previous -> current; never use current -> previous while calling the row a loss.

When using a helper script or fetcher, verify its window labels before interpreting the results:

  • previous: older comparison window.
  • current: newer comparison window ending on the latest complete GSC date.
  • A visibility loss means current < previous for clicks, impressions, or CTR, or current average position is numerically higher than previous average position.

If the fetcher emits windows in the opposite order, normalize them before writing the report. If the direction cannot be verified, state that the trend confidence is low and do not recommend content changes from that comparison alone.

Daily-normalize unequal windows before ranking losses. Use raw totals only when the windows have the same number of days.

Report both page-level and query-level evidence for major alerts. A page-level position drop without the affected query cluster can be misleading when query mix changes.

Action Thresholds

Use these rules to decide when monitoring is no longer enough:

  • Critical position regression: current average position worsens by 10+ positions and current impressions are down 50%+ on a canonical, indexable 200 blog URL with at least 100 previous-window impressions. Prepare a blog-edit brief now unless the page was edited less than 7 complete GSC days ago.
  • Persistent zero-click CTR leak: a query/page has 300+ impressions, average position 3-10, and zero clicks for two consecutive monitoring runs. Prepare a SERP title/meta/internal-link edit brief; do not keep saying only "monitor."
  • Post-refresh failure: if a page was refreshed recently, wait for 7 complete GSC days before judging severe regressions and 14 complete GSC days before broad rewrites. After 14 complete GSC days, persistent zero clicks, CTR below 0.2% at positions 3-10, or a 30%+ click drop must trigger a narrow edit brief.
  • Indexing backlog blocker: if a report recommends request indexing and the URL is not already checked off in docs/SEO/maintenance/gsc-request-indexing-backlog.md, add or update that URL in the backlog. Do not repeat "request indexing" across reports without changing the backlog state.
  • Indexing backlog alert: count unchecked Request indexing items in docs/SEO/maintenance/gsc-request-indexing-backlog.md during every run. If the backlog has more than 10 unchecked URLs after this run, prominently ping the user in the final response and report Open Actions with the exact count and the oldest pending date.
  • Intentional migration: if an old URL loses impressions because it redirects, do not edit the old URL. Validate the destination page; if the destination also loses impressions or clicks after 14 complete GSC days, escalate on the destination.

Allowed deferrals:

  • The affected canonical page was edited less than 7 complete GSC days ago.
  • GSC data does not yet include the deploy/indexing date.
  • The drop is entirely an intentional redirect/unpublish migration and the destination is healthy.
  • The URL has fewer than 100 previous-window impressions and no click loss.

When deferring, name the exact date when the next run must act and define the trigger, e.g. "On 2026-05-26, if position remains worse than 20 or clicks remain zero, run blog-edit for the sharpener/unblur cluster."

Inputs

Minimum:

  • Site/domain.
  • Blog URL pattern, usually /blog/.
  • Date range or default twice-weekly comparison.

Optional:

  • GSC export/API data.
  • Recent deployment/change log.
  • Blog backlog files.
  • Known business events or publishing bursts.

Workflow

Stateful backlog discipline

Before classifying a page or query, reconstruct its current state from docs/SEO/maintenance/seo-changes-backlog.md, docs/SEO/maintenance/gsc-request-indexing-backlog.md, .claude/skills/blog-changelog.md, the prior two relevant monitor reports, and recent git history. Respect recorded deploy/indexing dates, measurement-lag windows, previous rationale, and exact next-action triggers. Never recommend or apply a repeated edit from stale report context.

After any applied fix or material decision that changes the next-action state, update the canonical backlogs in the same run. Record what was done and why, evidence, affected URLs/files, tests/verification, commit, deploy state, and exact follow-up trigger/date in seo-changes-backlog.md; update indexing rows in place without duplicates; and append blog-changelog.md for blog content/metadata changes. Dated monitor reports are optional supporting artifacts, not the source of truth.

  1. Fetch GSC blog-only data

    • Use page+query+date dimensions where possible.
    • Filter to URLs matching /blog/.
    • Compute daily-normalized before/after metrics by URL and query.
    • Rank by impression loss, click loss, CTR loss, and position loss.
  2. Correlate with recent markdown changes/backlog

    • Check recent markdown backlog/change files before recommending edits.
    • Always read .claude/skills/blog-changelog.md if present; it is the project-local blog edit changelog used by blog-edit.
    • Prefer these locations when present:
      • .claude/skills/blog-changelog.md
      • docs/SEO/maintenance/seo-changes-backlog.md
      • docs/SEO/maintenance/gsc-request-indexing-backlog.md
      • docs/SEO/reports/*.md
      • docs/PRDs/*blog*.md
      • docs/PRDs/*seo*.md
    • Compare publish/update dates, metadata refreshes, content rewrites, redirects, template changes, and indexing requests against the drop window.
    • If git history is available, inspect recent changes for affected blog slugs and blog rendering code.
  3. Classify each loser

    • Impressions down, position stable: query volatility, SERP layout, seasonality, title/query mismatch, or stale indexing.
    • Impressions down, position worse: content quality, cannibalization, indexation, or intent mismatch.
    • Clicks down, impressions stable: CTR/snippet issue.
    • Multiple blog URLs share the same query: cannibalization or unclear target page.
    • Mark each row with an action state: edit-now, indexing-follow-up, technical-fix, migration-monitor, or defer-with-deadline.
  4. Prepare edit brief

    • Prepare an edit brief for every edit-now row. Do not leave these as prose-only recommendations.
    • For each page, specify target query, new SEO title, new meta description, H1/title, first paragraph adjustment, internal links to add, and any content modules needed.
    • Keep SEO titles roughly 30-60 characters and descriptions 120-160 characters.
    • Use query modifiers only when supported by GSC: free, online, 2026, tested, no signup, no watermark, unblur, sharpener, etc.
  5. Apply updates

    • Load .claude/skills/blog-edit/SKILL.md and follow its workflow.
    • Use blog-edit's API-based update path: read the blog changelog, fetch the current post, PATCH only changed fields, verify, then append a changelog entry.
    • Use x-api-key with BLOG_API_KEY from .env.api, as documented in blog-edit.
    • Ensure generated/static pages are revalidated or redeployed after updates.
    • If the task is a scheduled autonomous maintenance run and the edit brief meets an action threshold, apply the narrow update in the same run. If credentials, server, or deploy state block the edit, record the blocker and create a concrete backlog item.
  6. Report and follow-up

    • Save a dated report in the repo's SEO reports directory when available.
    • Include applied fixes, open actions, and the next monitoring date.
    • Add important updated URLs to docs/SEO/maintenance/gsc-request-indexing-backlog.md if they are not already present.
    • Recommend manual Search Console indexing requests for important updated URLs, but do not imply the URL Inspection API can request indexing.
    • Append a concise entry to docs/SEO/maintenance/seo-changes-backlog.md for every monitor run that changes content, metadata, redirects, sitemap state, or indexing backlog state.

GSC Pattern

Use the existing fetcher for broad data:

node ~/.claude/skills/gsc-analysis/scripts/gsc-fetch.cjs --site=myimageupscaler.com --days=28 --row-limit=25000 --output=/tmp/gsc.json

For incident comparisons, query dimensions:

dimensions: ['date', 'page', 'query'];

Calculate:

  • impressions/day before and after
  • clicks/day before and after
  • CTR before and after
  • average position before and after
  • top lost queries per URL
  • competing URLs per query

Edit Brief Format

## Edit Brief: /blog/[slug]

Evidence:

- GSC: [metric change]
- Recent change correlation: [backlog/git/deploy evidence or none found]

Target query:

- [query]

Update:

- SEO title: [title]
- Meta description: [description]
- H1/title: [visible title]
- First paragraph: [one-sentence adjustment]
- Internal links: [source pages and anchor text]
- Content additions: [FAQ/table/section if needed]

Report Format

# Blog Quality Monitor - [date]

Data:

- GSC: [range]
- Backlog/change files checked: [files]

## Blog URLs Losing Visibility

| URL | GSC change | Recent change correlation | Likely cause | Action |
| --- | ---------: | ------------------------- | ------------ | ------ |

## Changes Correlated

[recent edits/publishes/template changes that line up with drops]

## Escalations

| URL | Trigger | Deadline | Required next action |
| --- | ------- | -------- | -------------------- |

## Edit Briefs

[include every edit-now brief; say "None" only when all deferrals are justified by the rules above]

## Fixes Applied

[metadata/content/internal links/revalidation]

## Open Actions

[indexing backlog items, technical blockers, or user/manual actions. If unchecked indexing backlog exceeds 10 URLs, start this section with "User attention required: indexing backlog has [N] unchecked URLs."]

## Next Run

[date and exact triggers that will cause edits rather than another monitor-only report]

Single-URL Incident Report Format

# Blog Visibility Incident - [date]

URL: [original URL]
Destination/canonical: [URL or none]
Production status: [status/canonical/noindex finding]
GSC freshness: [latest complete date]

## Finding

[one-paragraph answer: why impressions changed]

## Evidence

- GSC: [before/after metrics for original URL and destination]
- Backlog/changelog clues: [dated entries]
- Technical check: [status, canonical, sitemap, indexing if checked]

## Classification

[intentional-migration | indexing-regression | content-change-regression | serp-demand-shift | ctr-regression]

## Action

[request indexing, monitor destination, fix blocker, or blog-edit brief]