The One-Session SEO Audit That Finds AND Fixes Issues (with Claude)
Updated September 8, 2026 · Works with Claude Code, Cursor and any MCP client · 6 min read
📊 TL;DR: Most SEO audits die in a document — a €500 agency PDF nobody implements, or a crawler report with 300 errors you don't understand. The diagnosis was never the hard part; implementation is. This workflow makes Claude do both in one session: it reads your live RankWay data through the RankWay MCP, audits your codebase against it, and fixes everything it can reach. You review, it executes.
What you need
- Your website codebase (or CMS access) open in Claude Code, Cursor or VS Code. This is what gets fixed.
- The RankWay MCP — one command connects your live technical audit, rankings, AI visibility and competitors. No CSV exports, no trials that expire: setup takes two minutes.
- Optional: your Google Business Profile info as a text file, if you're a local business — there's no clean API for it, copy-paste is honestly faster.
Never touched a code editor? The setup is three clicks — and if it feels scary, screenshot what you see and ask Claude to walk you through it. That's what it's for.
One honest reality check first
Your CMS decides how automated this gets. A codebase or headless setup: Claude fixes things directly. WordPress, Shopify, Webflow or Wix: RankWay publishes new content there, but existing-page fixes come back to you as a precise, prioritized to-do list — still a fraction of the usual 40 hours.
The prompt
Paste this into Claude Code (adapt the bracketed parts):
You are auditing and fixing the SEO of [yourdomain.com].
Available context: my site's codebase (this folder) and the RankWay MCP
(live audit findings, rankings, AI visibility, competitors).
Step 1 — Audit. Pull audit_findings, keywords and ai_visibility from the
RankWay MCP, then verify against the code:
- On-page: titles, meta descriptions (length!), H1/H2 structure,
internal links, image alts, FAQ blocks
- Technical: everything the RankWay audit flags — crawlability,
sitemap, indexation, structured data
- AI answer readiness: a direct 40-100 word answer under each H2,
FAQ schema, tables, llms.txt, AI crawlers not blocked
- Flag near-duplicate pages (e.g. city pages that only swap the
city name) with a rewrite plan
Step 2 — Fix everything you can reach, directly in the code.
When a fix could break something else, verify and repair first.
Step 3 — Output two files: everything found & fixed, and what
needs manual action from me.
Do not stop until you have re-checked your own work and the only
remaining issues are ones you cannot access.
The part everyone skips: the rerun loop
The first pass will not be perfect — ask anyone to fix 500 things and they'll land 400, and fixing one thing occasionally breaks another. So you rerun: same prompt, "check everything again and fix what remains." Each pass, the count drops.
Crucially, you verify against the crawler, not against Claude's word: re-run the RankWay audit and watch the on-page score climb. A model saying "all fixed" is a claim. A crawler showing zero findings is a fact.
What this won't do
- A clean crawl is necessary, not sufficient — content, backlinks and competition still decide rankings.
- Claude won't build links or write your strategy from this prompt. It will tell you these are out of scope. Believe it.
- A 1,000-page site takes hours. Let it run — every hour it works is a week you didn't spend doing it by hand.
The catch nobody mentions
This workflow is a Saturday afternoon. Then your site changes, Google updates, AI answers shift — and the audit is stale in three weeks. That loop is exactly what RankWay automates: the audit re-runs, rankings and AI citations get checked daily, and the content engine keeps closing new gaps while your assistant reads the results through the MCP. DIY with the prompt works. DIY with the MCP works better. Fully automated is RankWay.
