SEO & GEO audit — the internal tool that analyses Google and AI engines

Context

On SEO engagements, gluing Screaming Frog, Search Console, a spreadsheet and a SERP tool no longer holds once you also need to know what AI engines say about the brand. The studio therefore built its own audit engine — two Django apps, seo_audit and geo_audit — to analyse client sites and review generative answers.

Problem

A classic crawler sees tags, not meaning. It also does not see whether ChatGPT cites a competitor instead of the client, or whether Gemini hallucinates an offer. We needed a semantic audit pipeline (BERTopic, spaCy) and visibility tracking across several LLMs, async, with a dashboard — without making it a public product.

Approach

Same family of stack as RankyDocky (RankyDocky is in fact a fork of it): Django, GraphQL, React, Celery for long jobs (30 to 120 s), SSE for progress. Two distinct domains, one backend.

Tools

• Django · GraphQL
• BERTopic · spaCy
• Celery · SSE
• React

Where AI comes in

In the craft, not only in the code: topic modeling, semantic scoring, questions to LLMs for GEO, reports. GSD agents also speed up the making of the tool itself — I automate my audit work, I do not sell a chatbot.

Results

In production on client analyses. Not a commercial SaaS, so no usage metric to display — the deliverable is the audit, not a subscriber count.

• SEO

  crawl, keywords, SERP, links, reports

• GEO

  5 AI engines tracked · mentions · sentiment

• Fork

  stack reused by RankyDocky

Takeaway

1. 1

   Instrument your own craft: an SEO consultant who ships an audit engine no longer waits for the market tool to see what LLMs say about a client.

2. 2

   GEO is not glued-on SEO: you have to query the engines, detect hallucinations, follow a trend — another data model.

3. 3

   An analysis backend and a CRM can share the stack. RankyDocky does not have to rewrite the audit: it just has to reattach the apps.

Visuals