SEO sprint

Phased organic-search roadmap for codevetter.com — demand-mapped page coverage, competitor comparisons, and evidence-corpus scale.

SEO sprint — CodeVetter

Initialized 2026-09-22 from the Fleet scoreboard (seo-scoreboard.mjs --project codevetter): 403 impressions, 2 clicks, mean position ~37 over 28 days. Google is already serving us on the right terms — the gap is coverage depth and position, not topic selection.

Demand evidence (Search Console, 28d ending 2026-09-18)

Cluster	Queries seen	Impressions	Position
Validation	`ai code validation`, `validating agent generated code`, `coding agent validation`	~57	66–95
Verification	`ai code verification`, `verify code in agent loop`, `coding verification`, `deterministic code verification`	~29	47–98
Benchmarks	`ai code review benchmark`, `code review benchmark`, `codexglue benchmark`, `coding agent refactoring benchmark`	~6	50–79
Detection	`check if code is ai`, `is this ai generated code`	~2	72–77

Rules that keep this honest:

One canonical page per intent — no near-duplicate keyword variants that would cannibalize /verify-ai-generated-code or /coding-agent-verification.
Competitor pages quote the vendor’s own docs with a checked-on date and publish no invented head-to-head numbers (same standard as /codevetter-vs-coderabbit).
Evidence pages (/xray/*, /benchmark) are generated from the benchmarks/public-catch-rate/ corpus, never hand-invented.

Phases

Phase 1 — demand-mapped coverage + corpus scale
- xray-examples.json regenerated from the full 27-case catch-rate corpus (was 3 hand-written) via scripts/sync-xray-examples.mjs → 27 evidence pages + Markdown mirrors.
- New /ai-code-validation page targeting the largest open cluster (validation phrasing; disambiguates review vs validation vs verification rather than duplicating existing pages).
Phase 2 — competitor comparison matrix. codevetter-vs-* pages for the tools developers actually evaluate against: Qodo Merge, Sourcery, Cursor Bugbot, Copilot code review, Korbit, Bito, Ellipsis, cubic. Each entry needs its own docs-checked claims — batch draft, then verify quotes before publishing.
Phase 3 — benchmark cluster depth. Per-language case pages (the corpus spans ts/py/go/rust), a code review benchmark explainer comparing recognition fixtures vs repository-task benchmarks (SWE-bench lineage), and the codexglue-style “coding agent benchmark” explainer.
Phase 4 — Repo Unpack evidence pages. /unpack/<repo> publishes deterministic structural scans of well-known OSS repositories (express, flask, requests, axios, lodash, fastapi, redis, sqlite, ripgrep, httpie, curl, jsonwebtoken) — targeting “ codebase explained / architecture” demand. Corpus lives in benchmarks/repo-unpacks/; each page prints the scanned commit, upstream last-commit date, and codevetter version. Collect: node scripts/collect-unpack-repo.mjs <org/repo>; regenerate: node scripts/sync-unpack-pages.mjs. The claim-cited analysis layer is live on all 12: node scripts/analyze-unpack-repo.mjs <slug> --clone <path> drives claude -p (read-only tools) against the pinned clone — ~$2-4 and 5-10 min per repo, ~120 claims and ~250 verified file citations each.
Phase 5 — internal-link lattice. Once phase 2 pages exist, cross-link every comparison page to the nearest evidence page (/xray/<case>) and the benchmark methodology.

Measurement

Site Health search family: per-term position for the four clusters above; expect validation cluster pos 83→<40 within ~6 weeks of indexing.
Site Health github family: referrer entries from codevetter.com.
Indexing: submit new routes via pnpm --dir site-health indexing submit --project codevetter --sitemap after deploy.