A Michelin-Starred Restaurant Was Invisible to AI. We Fixed It in 7 Days.

A Michelin-Starred Restaurant Was Invisible to AI. We Fixed It in 7 Days.

When someone asks ChatGPT “best paella in London,” it doesn’t cite the Michelin-starred chef running the best paella restaurant in the city. It cites TripAdvisor.

This is the new search game, and most businesses don’t realize they’re already playing it.

I run Chad, an AI marketing tool. One of our first GEO (Generative Engine Optimization) clients was Arros QD — a Michelin-starred restaurant in Fitzrovia, run by Quique Dacosta, a chef with three Michelin stars under his belt. They serve authentic Valencian paella cooked over a six-metre open flame. The credentials are real. The food is exceptional.

And yet when AI systems answered “where should I get paella in London,” arrosqd.com wasn’t in the conversation.

We ran a baseline GEO audit on June 11. By June 25 we’d rebuilt the site. By June 30, Perplexity had started citing them directly. Here’s what we learned.

The Problem: AI Couldn’t See Them

The diagnosis was fast. Arros QD’s site blocked every major AI crawler: GPTBot (OpenAI), ClaudeBot (Anthropic), PerplexityBot.

Their robots.txt said “no” to the systems people were actually using to find restaurants.

But even if the crawlers could access the site, there was nothing structured for them to read. No schema markup. Chef credentials buried in prose. A Michelin star mentioned in a paragraph, untagged, unstructured, invisible to a machine trying to extract facts.

The site scored 27/100 on our GEO audit. Fifth out of five London paella spots we benchmarked against.

Traditional SEO would tell you to build backlinks, wait six months, outspend competitors on ads. AI search doesn’t care about any of that yet. It cares about citability — can the system extract a fact and attribute it with confidence?

Arros QD had the credentials. They just weren’t formatted for machines to read them.

The Fix: 48 Hours, Five Categories

We rebuilt the site in 48 hours with a Slack-based deployment pipeline. Here’s what moved:

AI Citability (+26 points) — Unblocked GPTBot, ClaudeBot, PerplexityBot in robots.txt. Added an llms.txt file signaling what the site is about and where key facts live. Restructured content so credentials appear early, in plain declarative sentences.

Content E-E-A-T (+22 points) — Pulled chef certifications, Michelin stars, awards, and authenticity markers out of narrative text and into clear, citable statements. Made expertise demonstrable, not implied.

Technical (+38 points) — Fast page loads, mobile-first, semantic HTML. Nothing revolutionary — just the basics done right so crawlers don’t give up halfway through a scrape.

Schema Markup (+75 points) — This was the biggest single lever. Restaurant schema, Organization schema (chef credentials), Menu schema (signature dishes, ingredients). Schema went from 10/100 to 85/100. The difference between a paragraph about paella and a machine-readable fact: “Arros QD serves Paella Valenciana, prepared by Michelin-starred chef Quique Dacosta.”

Platform Optimization (+23 points) — Structured content so it works across ChatGPT, Perplexity, Google AI Overviews, and Claude. Different systems parse differently — we made sure all of them could extract clean answers.

The Numbers

Controllable factors only — Brand Authority excluded because it can’t move in a two-week engagement:

Controllable GEO score: 29 → 65. Competitive rank: 5th of 5 → 1st of 5.

Perplexity’s citation behaviour shifted. Before: third-party review sites. After: direct citations to arrosqd.com.

Why This Matters

Traditional search is a war of attrition. You need domain authority, backlinks, ad spend, and time. A boutique operator can’t outspend TimeOut or TripAdvisor.

AI search is different. It rewards demonstrable expertise and structured content. A Michelin-starred chef running a single restaurant has more credible expertise than a review aggregator — the AI just needs to be able to extract and verify that fact.

This is the first time in 20 years that small operators with real credentials can compete in search without a six-figure ad budget or an SEO team.

Arros QD had the expertise. We just made it machine-readable.

What We Built

Chad is the tool that does this. We run six parallel AI agents to audit citability, schema, content quality, technical setup, and platform-specific optimization. The output is a GEO score, a breakdown of what’s broken, and a priority-ranked roadmap for fixing it.

The Arros QD rebuild took two days because we knew exactly what to fix and in what order. Most restaurant sites have the same gaps: blocked AI crawlers, missing schema, credentials buried in prose.

If you’re running a business with genuine expertise and you’re not showing up when people ask AI where to go — this is fixable, and it’s faster than you think.

Try Chad at trychad.ai