SEO Performance Review

SEO Performance Review

You convert each new Google Search Console + Bing Webmaster export into ranked, actionable recommendations for Mini Diarium. This is a performance-data skill, not an on-page audit. For technical/on-page issues use seo-audit. The two are complementary and share the docs/seo/product-marketing-context.md brand context.

Cadence: recommend bi-weekly at the current traffic (~455 Google clicks / quarter). Weekly is noise; GSC also lags 2-3 days. Do not run more often than the data can support.

Step 0 - Read context first

Read, in this order, before touching data:

1. docs/seo/product-marketing-context.md - brand, ICP, owned topic, the password/lock accuracy guardrail.
2. docs/seo/STRATEGY.md - the analysis framework, the keyword & topic-cluster map, the budget model (transfer/freeze/flip), and the measurement regime this skill executes. The cluster map in STRATEGY.md §3 is the scoring rubric for topical coverage.
3. docs/seo/action-plan.md - the live fix queue and the hypothesis log you will update.
4. The previous docs/seo/STATUS_REPORT_*.md (most recent by date) - the prior snapshot you compare against for movement.

Do not re-derive the strategy. Apply it.

Step 1 - Locate the latest exports (by pattern, never hardcoded names)

Exports live under docs/seo/performance/, one dated subfolder per cycle: docs/seo/performance/<YYYY-MM-DD>/, holding every file for that pull under its original downloaded filename. Match by filename pattern within the newest dated folder, never hardcode names — the pattern already changed once (Bing renamed the overview file), and zips are kept unextracted. Expect up to five files per cycle:

• GSC main export — *-Performance-on-Search-<date>.zip (not *Generative-AI*): contains Queries.csv, Pages.csv, Countries.csv, Devices.csv, Chart.csv, Filters.csv, Search appearance.csv. Extract to a temp dir to read; do not commit the extraction — the zip as downloaded is the tracked copy.
• GSC AI Overview export — *-Performance-on-Search-Generative-AI-Features-<date>.zip: same shape minus Queries.csv (Google does not break this report out by query). Impressions-only — no clicks/CTR/position. See Step 3.
• Bing overview CSV — *SearchPerformanceOverview*.csv (daily clicks/impressions/CTR).
• Bing query-level CSV — *KeywordReport*.csv (per-keyword impressions/clicks/CTR/position).
• Bing AI/Copilot citations CSV — *AIPerformanceOverviewStats*.csv (daily Citations / Cited Pages counts). See Step 3.

A folder may also hold stray non-export files (e.g. leftover placeholder text files from manual reorganization) — ignore anything that doesn't match these patterns rather than erroring, but flag unrecognized files to the user instead of silently treating them as data.

Older cycles (pre-2026-08) may still use the legacy layout: GSC CSVs extracted into a nested google-search-console_<date>/ folder, Bing CSVs loose directly under docs/seo/performance/ with no dated wrapper, and no AI-performance exports at all. Treat those as historical — read them for the prior-cycle comparison in Step 2/3, but do not rewrite them into the new layout.

Find the newest dated folder and the immediately prior one (the last STATUS_REPORT names which cycle it came from). If the newest export is older than ~3 weeks, it is stale: guide the user to pull fresh data before analyzing (Step 1a). Do not analyze stale data silently.

Step 1a - Guide a fresh pull when stale

If exports are stale or missing, walk the user through it (this is manual; the skill cannot authenticate). Save everything under one new docs/seo/performance/<today>/ folder, original filenames, zips left unextracted:

• Google Search Console (search.google.com/search-console): Performance > Search results > set date range to Last 3 months > Export > download the zip. Then repeat with the AI Overview filter applied (Search appearance > AI Overview, or the "Generative AI Features" report if offered directly) > Export > download that second zip.
• Bing Webmaster Tools (bing.com/webmasters): Search Performance > export the overview CSV; then the Keyword report > export the query-level CSV; then the AI / Copilot performance view > export the citations CSV.

Step 2 - Run the analysis framework, per engine

Execute the measurement regime from STRATEGY.md §5 for both engines. Respect the GSC caveats: position is an average, data is sampled (low-impression rows dropped), reporting lags 2-3 days. Aggregate low-impression rows before drawing conclusions; single-digit-impression rows are statistically fragile.

Produce these tables (Google from Queries.csv/Pages.csv, Bing from KeywordReport):

1. Branded vs non-branded split. Branded = "mini diarium" and close variants. Report the clicks share and whether generic demand is being captured.
2. Striking distance (position ~8-20 with real impressions). These are the near-term rank wins. Rank them by impressions.
3. CTR gap (high-impression / low-CTR relative to position). These are the title/meta levers. A page at position ~7 with sub-2% CTR is a flag.
4. Topical-coverage score against the cluster map (STRATEGY.md §3). For each cluster, note the owning page, its position, and whether a gap exists (demand with no strong owning page). Prioritize procedural/comparison gaps (highest LLM absorption).
5. Movement vs the previous snapshot. For the tracked queries/pages, note direction of change. This validates the prior cycle's hypotheses (see the hypothesis log).
6. Google-vs-Bing divergence check. Bing is Windows/PC-first and surfaces predecessor ("mini diary" successor) and feature-intent ("does diarium have a password", "offline encrypted diary") demand that Google under-reports. Every recommendation must account for both engines: Google = topic/positioning terms; Bing = platform + successor + feature intent.