/conversion-track

/seo:conversion-track — does the traffic actually convert?

Requires: a GSC export (Performance CSV / Bulk Export) or the toprank plugin's GSC integration, plus a conversion data source (an analytics/CSV export mapping pages → signups/sales). With no data, this command prints the export guide and stops. Human gate: never auto-publishes. Ends by presenting flagged pages + recommended actions for your approval. No vanity metrics: the KPI is conversion, never impressions or raw clicks.

The skeptic correction (szymon-slowik-seo): impressions and clicks are not the goal — conversion is. A page can win clicks and still convert nobody. This command closes the "traffic ≠ value" gap by joining search traffic to conversion per page and surfacing where the clicks are wasted.

GSC data contract (shared across the suite)

$ARGUMENTS accepts one of:

• gsc-export-path — path to a GSC Performance export (CSV/XLSX: Queries/Pages/Dates with Clicks, Impressions, CTR, Position) or a Bulk Data Export (searchdata_url_impression / searchdata_site_impression rows). This is the canonical contract; the full column spec lives in ${CLAUDE_PLUGIN_ROOT}/reference/gsc-data-contract.md.
• --since N — analyze the last N days (used with toprank's GSC integration, if present).
• --site URL — the property to analyze.

If no data source resolves, do not fabricate data. Detect toprank's GSC integration first; if absent, print the 3-step export guide: (1) Search Console → Performance → Export, or set up Bulk Data Export to BigQuery; (2) include Queries + Pages + Dates with Clicks/Impressions/CTR/Position; (3) save the file and re-run with its path. Then stop.

Flow

1. Resolve both data sources. Detect toprank (look for its seo-analysis / content-planner skills); if present, prefer delegating the GSC pull to it, else read the export at gsc-export-path. If neither resolves → print the 3-step guide and stop. Then resolve the conversion source — an analytics export or CSV mapping pages/paths to conversions/signups/sales. If conversion data is absent, say so plainly and stop: do not invent conversion numbers. Error clearly on a missing/empty/malformed file — never silently pass.
2. Join traffic ↔ conversion. Delegate to the seo-strategist agent via the Agent tool (fall back to general-purpose if the agent file isn't installed — same precedent as finance/organizze) to join GSC pages to the conversion source by page/path (normalize trailing slashes, query strings, and protocol), compute conversion rate per page (conversions ÷ clicks), and rank pages by traffic.
3. Flag the gaps. Surface high-traffic / low-conversion pages (clicks coming in, nothing converting — the wasted-traffic pages) and high-conversion winners worth doubling down on. Every flag must trace to a real row in both exports.
4. HUMAN GATE. Present the ranked per-page table + flags + recommended actions (improve the CTA, match search intent to the page, or stop chasing a query that never converts) and ask via AskUserQuestion what to action — top recommendation first. Never change anything automatically.

Output artifact

A per-page table — traffic (clicks), conversion rate, flag (high-traffic/low-CVR vs winner) — with a recommended action per flagged row, plus one line on data confidence. All for review; nothing is published or changed.

Guardrails

• Conversion is the KPI: rank and headline on conversion, never on impressions or raw clicks.
• No fake conversion data: if no conversion source is provided, say so clearly and stop — don't estimate it into existence.
• No vanity metrics: a high-traffic page with zero conversions is a problem to flag, not a win.
• toprank optional: detect-and-delegate per skill; degrade to native tools (Read/Bash/WebFetch); never hard-fail on a missing plugin.
• Human gate: every path ends at a human decision — improve, re-target, or abandon. The command never publishes or edits.