tj-pak — ai-seo-measured % cd ..

% open ai-seo-measured

AI SEO, Measured

Puts a real 0 to 100 score on how citable a site is by ChatGPT and Perplexity, then proves the number moved.

ai live

PROBLEM

Search is quietly moving off the ten blue links. People ask ChatGPT or Perplexity where to eat in Ponsonby, and the answer comes from whatever those models can read and trust about the site. A normal SEO score tells you nothing about that. I had a skill that gave good AI-SEO advice but no way to know if any of it actually worked. It was all vibes.

WHAT I BUILT

A way to put a real number on how citable a site is by AI search engines, then close the gaps and prove the number moved. An advice skill decides what is missing; a measurement engine scores the live site 0 to 100 and re-scores after each fix. I ran it on two live sites: my own portfolio and WATDA, our Korean restaurant in Ponsonby.

HOW IT WORKS

Two halves that pair up. The skill is the strategist: it reads the site and judges what AI engines need in order to understand and quote it. The engine is the proof: it audits the live URL and hands back a hard score plus the exact gaps. The loop is measure, fix, measure again. The fixes are real files a model can read: an llms.txt plain-language summary, JSON-LD schema (Restaurant, FAQ, WebSite) so the facts are machine-readable, and an ai.txt that tells AI crawlers they are welcome. I vetted the engine before adopting it (it runs entirely local, nothing leaves the machine) and bolted it onto the skill I already had.

WHERE IT'S AT

Live on both sites. The portfolio went 45 to 61 and WATDA went 50 to 66, both verified against the real domains rather than a preview build. Every gap the engine could detect is closed.

WHAT I LEARNED

Worth being straight about the limit: this measures whether a site is built to be cited, not whether the AI is quoting it yet. The final step, asking the engines directly whether they actually cite the sites, needs a paid API key I have not wired. So the claim stays honest. The sites are now built to be cited, and the readiness score proves the work landed. The shift it is built for is real though: the question is moving from the search bar to the chat box, and most sites are invisible to it.

STACK

Claude Code skill (ai-seo) · geo-optimizer engine · Astro · JSON-LD · Cloudflare Pages

A real audit-fix-reaudit loop on the live restaurant domain, scored before and after, every detectable gap closed.
$ geo audit --url https://www.watda.co.nz/
[audit]  fetching live page + structured data ...
[score]  before: 50/100   no llms.txt · partial schema · no ai.txt
[fix]    llms.txt · JSON-LD (Restaurant·FAQ·WebSite) · /.well-known/ai.txt
[deploy] cloudflare pages -> watda.co.nz   (live domain, verified)
[audit]  re-scoring live URL ...
[score]  after:  66/100  ✓   every detectable gap closed

% cd ..