How ChatGPT Recommends Restaurants — And Why Yours Is Missing

Meta Description: ChatGPT, Gemini and Perplexity recommend restaurants daily. Learn where AI gets its data and how to make your restaurant visible.

The Experiment

We asked ChatGPT, Google Gemini, and Perplexity the same question: „Find me a Chinese restaurant in London with gluten-free options.“ The results revealed a stark pattern.

ChatGPT recommended three restaurants, citing their „extensive gluten-free menu options.“ When we checked their actual websites, two had no gluten-free information at all — pure hallucination. Gemini pulled data from Google Reviews, suggesting restaurants where customers mentioned „accommodating dietary needs“ but provided no specific dishes. Perplexity performed best, citing actual sources, but still couldn’t deliver concrete gluten-free menu items.

The core problem became clear: AI systems excel at finding restaurants with good reviews and basic information, but struggle with specific queries about allergens, nutrition, or particular dishes. Most recommendations rely on Google Reviews sentiment and generic website text. When pressed for specifics — „Which gluten-free noodle dishes do they offer?“ — all three AI systems either hallucinated answers or defaulted to „Please contact the restaurant directly.“

This gap represents a massive opportunity. While AI confidently recommends restaurants for generic queries like „best Italian restaurant near me,“ it fails spectacularly when diners ask the questions that actually drive booking decisions: allergen-safe options, calorie counts, and specific dietary accommodations.

Where AI Gets Its Restaurant Data

AI restaurant recommendations draw from five primary sources, each with distinct limitations:

Google Places and Maps provide the foundation — restaurant name, address, phone number, hours, and review scores. This covers the basics but offers zero insight into actual menu items or dietary accommodations.

Restaurant websites get crawled for text content, but most sites present menus as unstructured text or PDFs. AI systems can read „Pad Thai – delicious stir-fried noodles“ but can’t extract that it contains peanuts or has 650 calories.

Review platforms like Yelp, TripAdvisor, OpenTable, and TheFork provide sentiment and occasional mentions of dietary options. A customer might write „they accommodated my gluten allergy,“ but this doesn’t specify which dishes are actually gluten-free.

Schema.org markup represents structured data gold — machine-readable information that tells AI systems exactly what each dish contains, its price, allergens, and nutritional facts. This is the most reliable source, but fewer than 5% of restaurants implement it properly.

Social media from Instagram, Facebook, and TikTok offers limited value for AI recommendations. While great for marketing, social posts rarely contain the structured information AI needs for specific dietary queries.

The quality gap is enormous. According to BrightLocal’s 2025 study, 87% of consumers use Google to evaluate local businesses, but only a tiny fraction of restaurants provide the structured data that AI systems need to make accurate, specific recommendations.

Why Structured Data Gets Preferred

AI systems face a critical challenge: hallucination. When lacking reliable data, they fabricate plausible-sounding but potentially dangerous „facts“ — especially problematic for food allergies and dietary restrictions.

Structured data using Schema.org JSON-LD markup acts like a verified promise to AI systems. When a restaurant properly marks up a dish with MenuItem schema including allergen information, AI can confidently state „The Kung Pao Chicken is gluten-free“ instead of guessing. With NutritionInformation markup, AI can accurately report „This dish contains 420 calories“ rather than hallucinating a number.

Without structured data, AI must infer from unstructured text. A menu description reading „fresh vegetables and rice“ might prompt AI to assume it’s vegan, potentially missing fish sauce or chicken broth. For someone with severe allergies, this guesswork becomes genuinely dangerous.

Both OpenAI and Google have signaled clear preferences for structured, verifiable data sources in their latest models. OpenAI’s ChatGPT Search explicitly prioritizes websites with proper markup, while Google’s algorithms have long rewarded structured data with enhanced visibility. Perplexity’s citation-focused approach makes structured data even more valuable, as it can point directly to verified information rather than hedging with uncertain language.

This preference isn’t accidental — it’s a quality control measure. Structured data reduces hallucinations and increases AI confidence, leading to more definitive recommendations and higher user satisfaction.

What Your Restaurant Needs to Deliver

Making your restaurant visible to AI systems requires systematic data structuring. Here’s the essential checklist:

✓ Google Business Profile completion with accurate hours, photos, and regular updates. This remains your foundation for local discovery.

✓ Website with Schema.org Restaurant markup including complete business information, cuisine type, price range, and service options.

✓ Individual URLs for each dish rather than a single menu page. AI systems better understand „/menu/kung-pao-chicken“ than „#dish-27“ on a long page.

✓ MenuItem schema per dish with precise descriptions, prices, ingredients, and preparation methods. Avoid generic descriptions like „delicious pasta dish.“

✓ Machine-readable allergen information embedded in each dish’s markup, not buried in footnotes or separate PDF allergen guides that AI can’t easily process.

✓ NutritionInformation per dish including calories, macronutrients, and key dietary flags like vegan, keto-friendly, or low-sodium options.

✓ Multi-language support with hreflang tags, crucial for restaurants in multicultural cities like London, New York, or Toronto where customers search in multiple languages.

✓ Regular content updates as AI systems favor fresh, current information over static content that hasn’t changed in months.

Case Study: WITH vs. WITHOUT Structured Data

Consider two Chinese restaurants in Manhattan with similar quality and service:

Golden Dragon (Restaurant A) operates traditionally with a PDF menu, no structured data, 4.3 Google stars from 500+ reviews, and standard website text describing „authentic Chinese cuisine with fresh ingredients.“

Jade Garden (Restaurant B) implemented comprehensive structured data through ChinaYung, marking up every dish with allergens and nutrition information. Despite having only 3.8 stars from 120 reviews, each menu item includes detailed JSON-LD markup specifying ingredients, calories, and dietary accommodations.

The results split predictably by query type. For generic searches like „best Chinese restaurant NYC,“ ChatGPT and Gemini recommend Golden Dragon based on review volume and rating. But for specific queries — „Chinese restaurant Manhattan with gluten-free options under 500 calories“ — all three AI systems recommend Jade Garden.

They can confidently suggest specific dishes: „Try the Steamed Fish with Vegetables (380 calories, gluten-free, dairy-free)“ rather than the vague „Golden Dragon accommodates dietary restrictions — please call ahead.“

This pattern reflects the future of restaurant discovery. Generic queries („best restaurant near me“) still favor traditional signals like reviews and ratings. But specific, health-conscious queries — the fastest-growing segment according to the National Restaurant Association — increasingly depend on structured data availability.

Jade Garden reports 23% more bookings from health-conscious diners and 31% fewer cancelled reservations due to dietary accommodation issues, simply because customers know exactly what they’re getting before they arrive.

Make your restaurant visible to AI. With ChinaYung you automatically get structured data for every dish — allergens, nutrition facts and Schema.org markup.

Start for Free →

FAQ

Can I influence what ChatGPT says about my restaurant?

You cannot directly control ChatGPT — there is no „ChatGPT Business“ account like Google Business Profile. But you can improve the data foundation that ChatGPT accesses. When your website delivers structured data with Schema.org markup, ChatGPT has verified facts instead of guesses. This reduces hallucinations and increases the chance your restaurant is mentioned correctly and positively for specific queries. The same applies to Gemini and Perplexity — better data in means better answers out. Think of it as indirect influence through data quality rather than direct control through platform settings.

How quickly does AI visibility change for my restaurant?

AI systems update their data at different speeds. Google Gemini accesses the live Google index — changes to your website with Schema.org markup are often visible within days. ChatGPT updates its training data less frequently but increasingly uses live web browsing through ChatGPT Search for current information. Perplexity searches in real-time and can surface your structured data immediately. Generally: the sooner you provide structured data, the sooner you benefit. Restaurants that adopted comprehensive Schema.org markup in early 2025 already report higher visibility in AI-generated recommendations, particularly for specific dietary queries that drive actual bookings.

Are good Google reviews not enough?

Good reviews remain essential for general recommendations like „best restaurant in London.“ But for specific queries — „restaurant with meals under 500 calories“ or „peanut-free Thai restaurant“ — AI needs precise data, not opinions. Reviews say „the food was great.“ Structured data says „Dish X contains 420 calories, is gluten-free and contains no peanuts.“ Both matter, but the future belongs to specific, health-conscious queries. The National Restaurant Association reports that 70% of diners now consider health and dietary factors when choosing where to eat. Reviews build trust and social proof, while structured data enables discovery for the specific needs that actually drive reservation decisions.