seo-for-saas-businesses: a 13-phase SEO and GEO program for SaaS sites under 2026 Google policy

118 files covering technical SEO, INP-first Core Web Vitals at the 200 ms and 500 ms thresholds, AI Overview and AI Mode citation surfaces, schema, hreflang, programmatic templates, and PR-shaped deliverables. Every recommendation ships as a Decision Card with rollback path. Default stack: Next.js 16 + Vercel + Supabase; methodology is stack-agnostic.

• Mode router picks which of the 13 phases get the deepest investment. Modes: greenfield-seo, mature-site-audit, traffic-drop-triage, programmatic-launch-review, migration, ai-visibility-pass, core-update-response, lifecycle-content, maintenance. Diagnose before rewriting.
• Render-diff gate before any keyword work. A crawl + render diff is mandatory in Phase 1 so HTML-limited bots and answer-engine fetchers see the same canonical, schema, and primary content as the rendered DOM. Disagreement is the most common quiet ranking suppressor.
• INP at the official thresholds: good at 200 ms or under, poor above 500 ms. The hidden offender on SaaS marketing pages is usually a heavy dashboard component (chart libraries, CRM widgets, consent banners that block the main thread) leaking into hero or pricing routes.
• Programmatic gates are non-optional: the scaled-content-abuse tripwire is checked before any template family ships. Per-page unique value, real underlying data, and a maintenance contract are required, regardless of whether pages were authored by automation or humans.
• AI Overview and AI Mode count as separate citation surfaces under the same SEO rulebook. Pages still need to be snippet-eligible, expose important content textually, and provide dated, source-backed extractable passages with multiple unique data points.
• 18 operator scripts with output shapes pinned in SCRIPT-IO-CONTRACTS (JSON for most), so dashboards and the verify-prod gate wire against script output: crawl, GSC extract, CWV-by-component, hreflang validator, sitemap audit, schema validator, striking-distance, verify-prod, CrUX.
• 19 phase-mapped subagents: audit-area, cluster-researcher, cwv-collector, three fresh-eyes variants for bughunt, cross-review, and trace, gsc-extractor, ia-mapper, impl-pr, log-analyst, prod-verifier, serp-snapshotter, asset-builder. Each subagent has a single charter.
• Every recommendation ships as a Decision Card: hypothesis, expected impact, tracking plan, rollback path, owner, ship-by, recheck-by. Every claim carries an Evidence Label (confirmed, likely, or hypothesis) plus a recheck-by date so stale assertions do not pile up.
• 16 deliverable templates in assets/, including a 90-day plan, content brief, audit item, KPI target sheet, migration URL map, weekly report, monthly executive summary, decision card, experiment card, source log, seasonality calendar, SERP-feature scan, citation tracking, content inventory.
• Score convention is uniform: 0 worst to 1000 best for every skill-owned score (priority, content quality, citation readiness, authority, experiments, QA). Third-party raw scores like Lighthouse 0..1 or 0..100 are reported alongside their normalized score_0_1000 form.