Project

Most sites are invisible to AI. This tells you why.

GEO Audit is an open-source audit system that runs inside Claude Code. Six agents analyse your site in parallel and produce a scored report with prioritised fixes — in markdown and PDF.

Agents

Parallel

Execution

Free

Open source

The problem

Search is changing. When someone asks ChatGPT, Perplexity, or Claude to recommend a consultant, an agency, or a tool, the answer doesn't come from your homepage. It comes from what AI systems have found, understood, and chosen to trust about you across the entire web.

Most sites fail this test. Not because they're badly built. Because they haven't been optimised for how AI systems read, evaluate, and cite content.

That discipline is called Generative Engine Optimisation — GEO.

I audited my own site using this tool. Technical execution scored 88/100. External authority scored 10/100. The gap told me exactly where the work needed to happen, and it wasn't the code.

What it audits

Six agents run in parallel, each looking at a different layer.

AI Citability

Can your content be extracted, understood, and quoted by AI systems? Checks content structure, quotability, and whether your key ideas are distinct enough to cite.

External Authority

How your brand shows up across the web — directories, publications, podcasts, third-party mentions, reviews. The signal AI systems use to decide whether others recognise you.

Content E-E-A-T

Experience, Expertise, Authoritativeness, Trustworthiness. How well your content demonstrates genuine depth and earns trust from both readers and AI evaluators.

Technical GEO

AI crawler access, llms.txt, robots.txt, rendering, canonicals, and page speed. The infrastructure layer that determines whether AI systems can reach your content at all.

Schema & Structured Data

Schema.org markup quality and completeness. Structured data is how you tell AI systems what your content means, not just what it says.

Platform Readiness

How well your site is positioned for AI-powered search platforms — Google AI Overviews, ChatGPT, Perplexity, Gemini, and Bing.

What you get

Two files. Every finding cited.

Output 01

GEO-AUDIT-REPORT.md

A full scored report with issues ranked by severity, category deep dives, quick wins, and a 30-day action plan. Every finding cites the exact source it was drawn from.

Output 02

GEO-AUDIT-REPORT.pdf

The same report in a formatted PDF with selectable text, a branded cover page, and a score summary.

Every technical finding is verified with curl before it's written. The tool does not hallucinate schema findings or infer crawler access — it checks.

Why I built it

GEO audits from agencies cost thousands. I wanted something I could run myself, and I wanted the results to be reliable — not a list of assumptions dressed up as findings.

I built this inside Claude Code, drawing on an existing open-source GEO project and adding verification logic, parallel agents, and PDF generation. The evidence rule is simple: if a subagent finds something but curl disagrees, curl wins.

The tool is open source. You can run it yourself with a Claude subscription — no API key required.

Read: What happens when B2B buyers search →

Run it

Requirements and installation.

Requirements

Claude Code with an active Claude subscription
Python 3.9+
Git

Install

git clone https://github.com/Apostle36/geo-audit.git
cd geo-audit

mkdir -p ~/.claude/skills/geo-audit ~/.claude/agents
cp .claude/skills/geo-audit/SKILL.md ~/.claude/skills/geo-audit/SKILL.md
cp .claude/agents/*.md ~/.claude/agents/

python3 -m venv .venv && source .venv/bin/activate
pip install -r scripts/requirements.txt
playwright install chromium

Run

/geo-audit https://yoursite.com

View on GitHub →

Scoring model

Weights reflect where the gap between technically sound sites and AI-visible sites tends to be largest. Most sites fail on authority, not on code.