Case study
Built so the machines cite you, not just rank you.
geo-seo-claude is a GEO/AEO-first SEO toolkit: citability scoring, AI-crawler access analysis, schema markup generation and reports.
Being cited by ChatGPT, Claude and Perplexity is the target; Google rankings are one input. Public on GitHub and running on this site. By Amber Lan, AI marketing engineer. Last updated 20 August 2026.
Back to selected work
The problem.
Buyers ask AI assistants for recommendations, and assistants answer with names.
The gap in classic SEO
Rankings and backlinks say nothing about whether an AI crawler can reach a site, whether the content is quotable in isolation, or whether the structured data says who the business is. Small businesses lose the new front page in that gap.
Checks, not vibes
Crawler access is testable, schema parses or it does not, and a citability score needs stated criteria. It had to run as a practitioner tool inside Claude Code, on a real client site, in minutes.
What was built.
The toolkit audits a site the way an answer engine consumes it, then generates the fixes it recommends.
Crawler access
The nine AI agents that matter, from GPTBot to PerplexityBot, plus llms.txt presence and quality.
Structure
Structured-data coverage and answer-first content structure.
Citability
Passage-level citability scoring, then the fixes: robots rules, JSON-LD schema and a report a client can act on.
Key decisions.
Three calls that change what gets fixed first.
Citability as the metric
A page that ranks but cannot be quoted in isolation loses the AI answer. Scoring quotability changes what you fix first.
Crawler access before content
If GPTBot or PerplexityBot is blocked, nothing else matters, so the audit starts there.
Generate fixes, not findings
A report that ships the corrected robots.txt and schema blocks gets implemented. A list of problems gets filed.
The result.
The proof is recursive: this website runs the toolkit's own playbook.
Live proof
flowai.co.nz allows all nine AI crawlers, serves llms.txt at the root and under .well-known, carries JSON-LD on every page, and structures answers for extraction. The Flow AI retainer applies the same toolkit to client sites.
Evidence
Read the toolkit on GitHub. Inspect the live implementation: robots.txt, llms.txt, and the structured data in this page's source.
What I learned
Answer-engine optimisation is discipline, not tricks: let the crawlers in, say who you are in schema, lead with the answer, keep claims sourced. The toolkit makes that discipline repeatable.
Want a system like this pointed at your business?
Tell me what you do and where the pipeline stalls. You get a written read within the first week.