Claude SEO's falsifiability requirement is what separates it from typical audit tools

Claude SEO is a Claude Code plugin that runs SEO audits by dispatching up to 15 specialist agents in parallel across 25 sub-skills covering technical SEO, Schema.org markup, content quality, local SEO, e-commerce, and what the project calls GEO — optimization for AI-powered search features. The whole thing installs as a Claude Code skill and surfaces through slash commands: /seo audit, /seo schema, /seo geo, and about 30 others.

The architecture is genuinely interesting. Rather than a monolithic crawler, the orchestrator fans out to sub-agents simultaneously, which is why the README claims full site audits complete in 10-15 minutes rather than hours. Each recommendation is supposed to carry four explicit fields: the first-principle observation behind it, its dependency on other recommendations, a failure-mode check ("how would we know this failed?"), and a leading indicator to monitor afterward. That falsifiability commitment is the most distinctive design choice here — most SEO tools produce findings with no mechanism for verifying whether the fix worked.

The AI search angle is handled more carefully than most tools manage. Rather than treating AI Overviews as a separate discipline requiring special keyword strategies, the plugin follows Google's own published position that AEO and GEO are just rebranded SEO. It scores passage citability in the 134-167 word self-contained answer block range, checks question-based heading hierarchy, and explicitly flags three popular myths as unsupported: llms.txt as a citation lever, content chunking for AI, and AI-specific keyword rewriting. The schema deprecation tracking is similarly current — FAQPage lost its rich result benefit on May 7, 2026, and the plugin knows this; HowTo rich results went in September 2023, and a whole cohort of types (SpecialAnnouncement, ClaimReview, VehicleListing, and others) retired in June 2025.

The v2 release added headless rendering via Playwright Chromium with auto-detection of SPA frameworks, which closes the most significant gap in v1. The test suite grew from 39 tests to 410 across the v2 line, with 91 cases specifically targeting SSRF and DNS-rebinding bypass scenarios — a level of security attention unusual for a tool in this category.

Two real limitations are stated plainly. Pages with scroll-bound hydration or post-interaction content fetches still produce noisy findings; the recommended workaround is a manual cross-check with the seo-visual subagent. And without API credentials, Core Web Vitals are lab estimates only — the free tier makes no third-party calls beyond fetching the URLs you point it at, which is honest but means competitive and field data require accounts with Ahrefs, DataForSEO, SE Ranking, or Google's own APIs.

The comparison table against manual audits and commercial tools is self-serving in the way all such tables are, but the data-locality claim holds up: everything runs locally by default, nothing is uploaded to a vendor, and the MIT license means no per-domain pricing or exit friction. For teams already living in Claude Code, the workflow integration is the real argument — the audit runs in the same environment as everything else, and the output is plain markdown files you own.