How to Audit Your Site with geo-seo-claude (Open-Source, Step by Step)

Updated 2026-09-06 · tool · geo-seo-claude, audit, open-source

Most website checkers tell you what a human sees. But the pages that matter to your traffic are read by crawlers — Google's bots and the AI engines that decide whether to cite you. Those two readers are picky, and they don't care that your site "looks fine in Chrome."

geo-seo-claude is an open-source tool that audits your site the way crawlers actually see it. This guide walks through installing it, running an audit, and — most importantly — what to do with the results.

What this tool does (and doesn't)

It audits your site through the "crawler funnel":

1. Can it be found? — are any pages accidentally blocked by robots rules or access config?
2. Can it be read? — is the page structure and server-rendered content complete?
3. Will it be cited? — is the content structured so AI engines would quote it?
4. Is the content deep enough? — does it compete with what's already ranking?

What it doesn't do: write your content, build your backlinks, or magically rank you. It finds the technical blockers so your good content can actually surface.

The biggest value is the before/after baseline. Run it once before you change anything, fix the top issues, re-run in a month, and let the numbers tell you whether the fix worked.

Step 1 — Install

geo-seo-claude is designed for Claude-style agent environments and installs with a single command on macOS/Linux (Git Bash on Windows):

curl -fsSL https://raw.githubusercontent.com/zubair-trabzada/geo-seo-claude/main/install.sh | bash

The installer sets up the skills and roles the tool uses. You'll then invoke it from your agent by pointing it at your site URL.

Before running anything from the internet, glance at the script. Open the install.sh link, read it, confirm it only sets up local files. Good habit regardless of the tool.

Step 2 — Run a first audit

Give the tool:

1. Your site URL (e.g. https://skillnest.club)
2. A short description of your setup — what your site is, where pages live, any known issues
3. A strict instruction: "only change local copies, never touch production" (worth stating every time)

The tool runs its checks across the funnel and returns a report. It also has specialized roles that run in parallel for a full sweep (technical, content, schema, platform analysis) — you don't have to orchestrate them by hand.

Step 3 — Save your baseline

Baseline metrics make later results credible; Case Studies for AI Products explains how to present before/after evidence honestly. Before fixing anything, save the report as your baseline. Note the specific numbers:

• how many pages are blocked from crawlers (Level 1)
• how many pages fail the "server-rendered content" check (Level 2)
• average content depth vs. competitors (Level 4)

Without a baseline, you'll have no way to prove your fixes worked. This is the step almost everyone skips — and it's the step that makes the tool worth installing.

How to read a finding (before you fix anything)

The report isn't a to-do list — it's a triage. Every finding falls into one of three buckets, and they deserve different responses:

The mistake people make is treating everything as equally urgent. The hard blocker is the one that's costing you months of traffic; the signal requires judgment, not reflex. Sort by bucket, then by effort.

Step 4 — Fix in funnel order

Level 1 first: findability

If pages are blocked, nothing else matters. Typical fixes (a full walkthrough lives in the technical SEO checklist):

• correct an over-broad Disallow rule in robots.txt
• add blocked-but-public pages to the allow list
• ensure important pages are linked via plain <a href> so crawlers can walk to them

This is the layer where the "it works in my browser but Google can't see it" problem hides — the one that silently costs months of traffic.

Level 2 next: readability

Make sure crawlers get the real content:

Then: citation & depth

Step 5 — Re-run and compare

• confirm important pages are server-rendered (content in the initial HTML)
• unique title + description per page
• clean heading hierarchy (one H1, logical H2/H3)
• add structured data (Article/FAQ schema) — feeds both Google and AI engines (see the JSON-LD guide)
• deepen thin pages to genuinely compete (this is where rankings stall if skipped)

Wait 2–4 weeks after deploying fixes, then re-run the audit against the same URL. Compare with your baseline:

• Did the "blocked pages" count drop?
• Did indexed pages rise in Search Console?
• Did impressions for your target pages start from zero and climb?

The delta is the proof. A single audit is a snapshot; the comparison is the story.

Common gotchas

The honest bottom line

• One rule, many readers. Different crawlers can interpret the same robots rule differently. A rule that one tool flags green can be a hard stop for another. Cross-check.
• Don't trust a single check. geo-seo-claude is excellent, but it can miss things — pair it with a second opinion (an independent crawl or a manual spot-check of your most important page).
• "Fixed" isn't "indexed." After fixing a blocking rule, you still need to request re-indexing in Search Console — fixes don't propagate instantly.
• Beware AI-assisted false positives. Audits built on LLM reasoning can hallucinate an issue that isn't there. Verify any surprising finding against the raw source before changing code.
• Don't forget the AI-readable files. By 2026, sitemap.xml, robots.txt and llms.txt are the three files AI crawlers use to map your site. If your audit checks the first two but not llms.txt (or if it's stale), your curated AI shortlist can quietly drift out of date. Regenerate it on every build, and add it to your audit checklist.

A crawler-view audit tool won't make your content great — but it removes the invisible barriers that keep great content from ever being seen. In our experience, most "my AI product gets no traffic" cases are Level 1 and Level 2 problems, not bad writing.

Install it, run it, save the baseline, fix the funnel in order, and re-measure. Boring, repeatable, and it works.

Next: how to interpret the numbers in Google Search Console so you know your fixes actually landed.

FAQ