How AI Engines Choose Which Restaurants to Recommend

The mechanics behind the recommendation

When a diner asks an AI engine for restaurant recommendations, the answer is assembled from a surprisingly narrow set of signals. Despite the billions of parameters in large language models, the restaurant that appears in a ChatGPT or Perplexity response is selected based on a small number of measurable factors. Understanding these factors gives restaurant operators a clear path to influencing whether their establishment shows up.

FindingFin analysis of 150 dining queries across ChatGPT, Perplexity, and Gemini reveals the specific signal hierarchy that determines AI restaurant recommendations. The results point to practical steps that independent restaurants in particular can take to improve their visibility.

The study: 150 queries, three engines

Delphium Labs ran 150 dining-related queries across ChatGPT, Perplexity, and Gemini during a two-week window in late January 2026. The queries were designed to represent the full spectrum of how diners actually ask AI engines for help:

• Cuisine-specific queries (40): "best Italian restaurant in Manchester", "authentic Thai food Birmingham", "Japanese omakase London"
• Occasion-based queries (35): "romantic restaurant for anniversary dinner Leeds", "restaurant for group birthday celebration Edinburgh", "business lunch restaurant Liverpool city centre"
• Neighbourhood and location queries (30): "restaurants near Deansgate Manchester", "best places to eat in Leith", "restaurant with parking near Bristol harbour"
• Dietary and preference queries (25): "best vegan restaurant Sheffield", "gluten-free dining options Brighton", "restaurants with good vegetarian menu York"
• Value and price queries (20): "affordable fine dining Birmingham", "best cheap eats in Glasgow", "restaurant tasting menu under 60"

Each query was run across all three engines, producing 450 total responses. We recorded every restaurant mentioned by name, its position in the response, any source citations, and whether the restaurant was part of a chain or group versus independently operated.

The signal hierarchy

Across 450 responses and over 1,900 individual restaurant mentions, four signal categories emerged as the primary drivers of AI restaurant recommendations. Delphium Labs ranked these by the strength of their correlation with appearing in AI answers.

1. Google Business Profile: the foundation

Google Business Profile completeness was the single strongest predictor of restaurant visibility across all three AI engines. Restaurants with fully completed profiles, including accurate hours, up-to-date menus, recent photos, complete attribute fields, and regular posts, appeared 2.4x more frequently than restaurants with minimal profiles.

The specific profile elements that correlated most strongly with visibility were:

• Menu accuracy: Restaurants with a current, detailed menu on their Google Business Profile appeared significantly more often. AI engines pull menu data when answering cuisine-specific queries. If your Google profile lists dishes from two years ago or has no menu at all, you lose this signal.
• Photo recency and volume: Profiles with more than 30 photos, including recent additions, correlated with higher visibility. AI engines appear to use photo metadata and volume as a proxy for business activity and relevance.
• Hours accuracy: Restaurants with verified, recently updated hours appeared more reliably in responses. Engines deprioritise businesses where hours data may be stale, particularly for "open now" or time-specific queries.
• Attribute completion: Google Business Profile attributes for restaurants include fields like "outdoor seating", "good for groups", "takes reservations", "serves alcohol", and many others. Restaurants with 80% or more of applicable attributes completed were 1.7x more likely to appear in relevant filtered queries.

This finding has a straightforward implication: your Google Business Profile is not a secondary listing channel. For AI visibility, it is the primary data source.

2. Review aggregation: volume, recency, and distribution

Review signals were the second strongest predictor. But the relationship is more nuanced than "more reviews equals more visibility". Three specific dimensions of reviews mattered.

Volume matters, but with a threshold effect. Restaurants with more than 200 Google reviews appeared significantly more often than those with fewer than 50. However, the returns diminished sharply above roughly 500 reviews. Going from 50 to 200 reviews had a much larger impact on visibility than going from 500 to 2,000.

Recency matters more than score. A restaurant with a 4.3 rating and 40 reviews in the past three months appeared more frequently than a restaurant with a 4.7 rating and no reviews in the past six months. AI engines appear to weight recent review activity as a signal of current relevance. A high score with stale reviews suggests a restaurant that may have changed or declined.

Platform distribution matters. Restaurants with review presence across Google, TripAdvisor, and at least one specialist platform (such as OpenTable, TheFork, or a cuisine-specific review site) appeared more often than restaurants with reviews concentrated on a single platform. This distribution effect was strongest for occasion-based queries, where AI engines seem to cross-reference across sources for confidence.

3. Website content: menus, story, and specifics

The restaurant's own website was the third most important signal source. Three types of website content correlated with AI visibility.

Menu pages with detail. Restaurants whose websites included full menus with dish descriptions, pricing, and dietary information appeared more frequently in AI answers. A menu page listing "Pan-roasted sea bass, samphire, brown butter, new potatoes - 24" gives an AI engine material to cite. A page that simply says "See our menu" with a PDF download does not. PDF menus are largely invisible to AI engines because many crawlers cannot parse them effectively.

About and story pages. Restaurants with substantive about pages describing the chef's background, the restaurant's philosophy, sourcing practices, and history showed higher visibility, particularly for quality-focused and occasion-based queries. This content gives AI engines context for why a restaurant is worth recommending, beyond simple review scores.

Location and practical details. Clear information about parking, public transport access, private dining capacity, group booking policies, and accessibility was associated with higher visibility for logistical queries. These are the practical details diners need, and AI engines surface restaurants that provide them.

4. Third-party citations: press, blogs, and awards

The fourth signal category was third-party mentions across food blogs, local press, award listings, and curated guides. Restaurants mentioned in food blogs, featured in local newspaper dining sections, or listed in award programmes (Michelin, AA Rosettes, Good Food Guide, Harden's) appeared more frequently in AI responses.

This signal was particularly strong for ChatGPT, which draws heavily from published text content. A single feature in a well-regarded food blog or local publication had a measurable positive effect on visibility. For independent restaurants without large marketing budgets, this is worth noting: a well-written pitch to a local food journalist may deliver more AI visibility than a month of social media activity.

How the engines differ

The three engines share the same broad signal hierarchy, but they weight signals differently in ways that matter for strategy.

ChatGPT: brand recognition and consensus

ChatGPT showed the strongest tendency toward well-known restaurants and brand consensus. Its recommendations leaned toward establishments with broad name recognition, high review volume, and coverage in mainstream publications. In FindingFin analysis, ChatGPT recommended chain or group restaurants at a higher rate (41%) than the other two engines.

For independent restaurants, ChatGPT visibility was most strongly associated with media coverage and review volume. The engine appears to favour restaurants it can describe with confidence, drawing on multiple agreeing sources. If several review platforms, food blogs, and publications all describe your restaurant positively, ChatGPT is more likely to recommend it. If your only digital presence is your own website, ChatGPT has less to work with.

ChatGPT also showed a notable tendency to recommend restaurants it could categorise clearly. A restaurant described consistently as "the best Thai restaurant in Manchester" across multiple sources appeared readily for that query. A restaurant with a harder-to-categorise menu, described differently across different sources, surfaced less reliably.

Perplexity: specificity and independent favouritism

Perplexity was the most favourable engine for independent restaurants. In FindingFin analysis, independents accounted for 44% of Perplexity restaurant recommendations, compared to 33% overall across all three engines. Perplexity also provided the most detailed citations, typically linking to specific review pages, food blog posts, or restaurant website pages.

The key differentiator for Perplexity was content specificity. Restaurants with detailed, recently updated website content performed well. Perplexity was also more likely to recommend restaurants mentioned in niche food blogs and specialist publications, rather than relying primarily on mainstream sources.

For independent restaurants, Perplexity represents the highest-return engine to optimise for. Its reliance on specific, citable content means that restaurants investing in their own website content and building relationships with food writers will see disproportionate results.

Gemini: the Google ecosystem

Gemini's restaurant recommendations were the most heavily influenced by Google's own data. Google Business Profile completeness was an even stronger predictor for Gemini than for the other two engines. Gemini also showed the strongest correlation with Google Maps data, Google Reviews specifically (as opposed to reviews on other platforms), and Google's own restaurant categorisation.

Delphium Labs analysis found that Gemini was the most responsive to Google Business Profile updates. Restaurants that had posted on their Google profile within the past two weeks appeared more frequently in Gemini recommendations than those with older activity. This suggests a recency signal specific to Google's ecosystem that Gemini weights more heavily than the other engines.

The cuisine authority pattern

One of the more striking findings in the data was what Delphium Labs has termed the "cuisine authority" pattern. Restaurants that demonstrated deep expertise in a single cuisine consistently outperformed multi-cuisine establishments in AI recommendations.

When FindingFin analysed visibility for cuisine-specific queries, restaurants with a clear single-cuisine identity appeared 2.8x more often than restaurants offering broad menus spanning multiple cuisines. A restaurant consistently described as "authentic Sichuan" or "traditional Neapolitan pizza" or "modern Indian fine dining" had a clear advantage over a restaurant offering "Italian, Mediterranean, and British classics".

This pattern makes sense from the perspective of AI engine mechanics. When a user queries "best Indian restaurant in Birmingham", the engine is looking for signals that confirm a restaurant's authority in that specific cuisine. A restaurant whose website, reviews, and third-party mentions all consistently reinforce "Indian cuisine" provides a strong, unambiguous signal. A restaurant whose digital footprint contains mixed signals about its cuisine identity is less likely to be selected for any single cuisine query.

The implication for multi-cuisine restaurants is not necessarily to narrow their offering, but to strengthen the consistency of their primary identity across their digital presence. If your restaurant serves primarily modern British food with some Mediterranean influences, ensure that "modern British" is the dominant descriptor across your website, Google profile, and the language you use in press materials.

Occasion-based versus cuisine queries

FindingFin analysis revealed that occasion-based queries and cuisine queries draw on different signal weightings.

Cuisine queries ("best Thai restaurant Manchester") are driven primarily by cuisine consistency, review sentiment specific to food quality, and menu detail. The AI engine is answering "what is the best example of this cuisine in this area?" and prioritises signals of authenticity and food quality.

Occasion queries ("romantic restaurant for anniversary Leeds") are driven more by ambiance cues in reviews, practical logistics (booking availability, private dining options, location details), and third-party coverage that describes the dining experience rather than just the food. The engine is answering "what is the right setting for this occasion?" and looks for different signals accordingly.

Restaurants that perform well across both query types tend to have rich, specific content addressing both food and experience. Their websites describe the cuisine in detail and also describe the atmosphere, the service style, and the practical details that matter for special occasions.

What independent restaurants should do

Based on FindingFin analysis across 150 queries, here are the most impactful actions for independent restaurant operators.

Treat your Google Business Profile as your primary digital channel. Update your menu, add recent photos monthly, complete every attribute field, verify your hours weekly, and post updates about seasonal menus, events, or changes. This is the single most controllable and highest-impact action for AI visibility.

Build review volume with a focus on recency. A consistent flow of recent reviews matters more than a high cumulative score. Implement a simple post-visit mechanism that encourages guests to leave reviews. Even a printed card with a QR code linking to your Google review page can shift the recency signal over time.

Put your full menu on your website in HTML format. Not a PDF. Not an image. Text-based menu content with dish names, descriptions, prices, and dietary indicators. This is one of the most common gaps FindingFin identifies in restaurant audits, and one of the simplest to fix.

Write a substantive about page. Tell your story. Describe your chef's background, your sourcing philosophy, why you chose your location, and what makes your approach distinctive. This content feeds directly into how AI engines describe and recommend your restaurant.

Pursue local food press and blog coverage. A feature in a food blog or local newspaper dining column provides a third-party citation signal that is particularly valuable for ChatGPT visibility. Reach out to local food writers with a specific angle, not a generic press release.

Strengthen your cuisine identity. Ensure that your website, Google profile, and social channels consistently reinforce your primary cuisine category. If you serve modern British food, say so clearly and repeatedly. Ambiguity in cuisine identity costs AI visibility.

The Delphium Labs methodology

The research behind this analysis follows the methodology Delphium Labs applies across all hospitality AI visibility studies. Queries are structured to represent real user behaviour, tested across multiple AI engines in controlled time windows, and analysed for correlation between specific digital signals and recommendation outcomes. FindingFin operationalises these research findings into ongoing visibility monitoring, allowing individual restaurants to track their own AI visibility and identify the specific gaps holding them back.

The restaurant sector is at an inflection point in how diners discover where to eat. The operators who understand the mechanics behind AI recommendations, and act on that understanding, will capture a growing share of high-intent discovery traffic. The signals are measurable, the actions are specific, and the advantage goes to those who move early.