By Chikei Yung, Founder chiwai — Published June 2026
Case Study: How China Restaurant Yung Became Visible in 39 Countries — Without an Ad Budget
A family restaurant in Frankfurt. 191 dishes. A PDF menu that existed for decades. And a question that changed everything: why does ChatGPT recommend our competitors but not us?
Author note: I am Chikei Yung, co-founder of chiwai. My family has operated China Restaurant Yung in Frankfurt since 1988. My sister Wai Wah and brother-in-law Li Xiejie run the kitchen. When I noticed that AI systems could not find the restaurant for specific queries — despite 38 years of operation and a loyal customer base — I built chiwai to solve the problem. China Restaurant Yung was the first restaurant on the platform. This case study documents what we did and what the data shows.
The Situation in 2023: Invisible by Default
China Restaurant Yung had operated in Frankfurt’s Sachsenhausen neighborhood since 1988. The restaurant had a loyal local following, a reputation for authentic Cantonese and Sichuan cuisine, and a kitchen run by Li Xiejie — a chef with decades of experience in Chinese culinary technique.
Online, the situation was different. The website existed primarily in German. The menu was a PDF file — a scanned document with 191 dishes organized in traditional menu categories. No individual dish pages. No structured data. No multilingual content.
For most of the restaurant’s existence, this was fine. Google rankings for „Chinese restaurant Frankfurt“ were built on proximity and review volume — signals that favored established local businesses. China Restaurant Yung had both.
Then the restaurant discovery landscape changed.
ChatGPT launched to the public in late 2022. By 2023, guests were using it to find restaurants with specific requirements: „Chinese restaurant Frankfurt peanut-free,“ „vegan dim sum Frankfurt,“ „gluten-free Chinese food near me.“ For these queries, a restaurant with no structured data simply did not exist in the AI’s answer — regardless of how good the food was.
China Restaurant Yung was invisible. Not because of quality. Because of data architecture.
The Five Steps We Took
Step 1: PDF Menu to Structured Dish Pages — One URL Per Dish
The first and most fundamental change: transforming a static PDF into a structured database of individual dish entities.
Each of the 191 dishes received its own page with a permanent URL. The dish name, description, price, category, and photo were stored as discrete data points — not as text in a document, but as structured fields in a database that could be rendered as machine-readable markup.
Names and descriptions were created in three primary languages (German, English, Chinese) and three additional languages (Spanish, Turkish, Arabic) — six languages total, covering the primary spoken languages of Frankfurt’s international population and tourist base.
The practical result: 191 individual pages, each with a unique URL, each containing a specific dish as a structured entity. Not a PDF. Not a menu page. Individual web entities that could be indexed, crawled, and cited.
Step 2: Schema.org Restaurant and MenuItem Markup on Every Page
Once the dish data existed as structured entities, the next step was making them machine-readable in the format that AI systems understand: Schema.org JSON-LD markup.
The restaurant homepage received a complete @type: Restaurant block: name, address, telephone, openingHoursSpecification, servesCuisine („Chinese“), priceRange, and a hasMenu link connecting to the menu entity.
Each individual dish page received a complete @type: MenuItem block: dish name in all six languages, description, price in EUR, image URL, menu section, and the full allergen and dietary data. The JSON-LD was embedded directly in each page — readable by Google Googlebot, GPTBot, and Gemini’s crawler in a single HTTP request.
This is the technical foundation that makes specific AI queries answerable. When someone asks „Which Frankfurt Chinese restaurants have peanut-free dishes?“ — the AI can now retrieve and parse the containsAllergen data for each of our 191 dishes and provide a factual answer. Without this markup, the question returns no result for us.
Step 3: Machine-Readable Allergen Labeling for 191 Dishes
Allergen data was already required by EU law (EU Food Information to Consumers Regulation, LMIV in Germany). But compliance — a PDF allergen table or a footnote on the menu — does not equal machine-readability.
The 14 EU-regulated allergens were mapped to every dish individually using the Schema.org containsAllergen property. Li Xiejie reviewed each dish against his ingredient knowledge and the German nutrient and allergen database BLS. The data went through an operator confirmation step — a manual verification by the kitchen team — before being marked as confirmed.
Dietary suitability flags (VeganDiet, VegetarianDiet) were added using the Schema.org suitableForDiet enumeration, based on the same ingredient-level review.
The result: 191 dishes with verified, machine-readable allergen and dietary data. A restaurant that can factually answer AI queries like „Which dishes are suitable for guests with gluten intolerance?“ — not with a general statement, but with a structured list of specific dishes.
Step 4: llms.txt on the Domain
The llms.txt file is a relatively new but increasingly important element of AI-first web presence. Analogous to robots.txt for search engine crawlers, llms.txt provides AI systems with a direct, curated overview of what a website contains and what data is available.
For China Restaurant Yung, the llms.txt file at chinayung.de/llms.txt includes: the restaurant’s founding story (1988, family-operated, Frankfurt), the cuisine type and specialties, a link index to all dish pages organized by category, allergen and dietary coverage information, and language availability.
AI systems that respect llms.txt — including several major crawlers — read this file first. It reduces hallucination risk (AI systems making up information about the restaurant) and increases the probability of accurate, specific recommendations.
→ How llms.txt works for restaurants — the full guide
Step 5: Google Business Profile Sync via API
Google Business Profile is the most direct channel into Google Gemini’s local restaurant knowledge. The Food Menus API allows structured menu data to be transmitted directly into GBP — meaning the same allergen-complete, multilingual dish data that lives on the website is also present in Google’s own structured data layer.
For China Restaurant Yung, 171 dishes were transmitted to GBP via the Food Menus API — with multilingual names and descriptions, allergen data, and dietary flags. The sync runs automatically: when a dish changes in the chiwai system, the GBP entry updates without manual intervention.
This creates a closed loop: the restaurant’s data is consistent across the website, Schema.org markup, and Google’s own database. AI systems that cross-reference these sources find consistent, verifiable information — which is exactly what increases the probability of recommendation for specific queries.
→ Google Business Profile and AI: GBP as bridge to Gemini
What the Data Shows
Within weeks of the five steps being completed, the dish pages at chinayung.de began generating search impressions from outside Germany. Google Search Console data shows impressions from 39 countries — including queries from English-speaking markets, Chinese-speaking markets, and others where the restaurant has no direct marketing presence.
The queries generating these impressions are not generic („Chinese restaurant Frankfurt“). They are specific: dish names in different languages, allergen-specific searches, cuisine-specific searches. These are exactly the queries that structured data makes answerable — and that were invisible to us before.
No advertising spend was used. The visibility increase came entirely from structured data making the restaurant’s information machine-readable and retrievable by AI systems across languages and geographies.
The live site: chinayung.de — all 191 dishes, DE/EN/ZH/ES/TR/AR and more languages on request, full Schema.org markup, live.
What This Means for Other Restaurants
China Restaurant Yung is not an exceptional case in terms of scale or resources. It is a single independent restaurant that had operated for 35 years with a traditional online presence. The changes we made — structured data, individual dish pages, allergen markup, multilingual content, GBP sync — are replicable for any restaurant.
The barrier is not technology. It is the time and expertise required to implement and maintain structured restaurant data correctly. That is the problem chiwai solves: automating the entire data layer so any restaurant — a family-operated Cantonese kitchen or a single-location bistro — can have the same AI-visible data infrastructure as a major chain.
The restaurant discovery landscape has fundamentally changed. The restaurants that understand this early will have a structural advantage that compounds over time — because structured data, once in place, works continuously without additional cost.
Your restaurant can do this too.
chiwai is the platform that makes this infrastructure accessible for independent restaurants. Free plan available — no credit card required.
Become a Pioneer Partner →FAQ
How long did the implementation take?
The five steps for China Restaurant Yung were completed over several weeks. The most time-intensive part was the allergen verification — having Li Xiejie confirm the allergen data for each of the 191 dishes against his kitchen knowledge. The technical steps (Schema.org generation, GBP sync, llms.txt) were automated by chiwai and required no technical expertise from the restaurant team. For new restaurants onboarding through chiwai today, the initial setup takes significantly less time because the automation pipeline is more mature.
Is this only relevant for restaurants with a large menu?
No. A restaurant with 20 dishes that has complete, verified, multilingual Schema.org markup is more visible to AI systems for specific queries than a restaurant with 200 dishes and no structured data. The size of the menu matters less than the completeness and accuracy of the structured data. A small focused menu with complete allergen and dietary data can actually have an advantage — simpler data is easier for AI systems to match to specific queries with high confidence.
Does this replace traditional marketing?
No — it complements it. AI visibility is a data infrastructure layer, not a marketing channel. It does not replace a strong social media presence, Google review management, or local SEO. What it does is create a foundation that makes all other marketing more effective: when an AI recommendation leads a potential guest to your website, the structured data they find there — complete dish information, allergen data, multilingual content — converts more effectively than an unstructured page. The two layers work together.