We Tested 200 Restaurant Queries Across Three AI Engines
By Delphium Labs
The top finding
When Delphium Labs tested 200 real restaurant queries across ChatGPT, Perplexity, and Gemini in January 2026, one signal stood above everything else: Google Business Profile completeness was the single strongest predictor of whether a restaurant appeared in AI-generated recommendations. Properties with a fully optimised profile were 3.1x more likely to be cited than those with minimal Google presence.
That finding matters because it is actionable. Unlike hotel AI visibility, where chain scale and OTA distribution create structural advantages, restaurant AI visibility is driven by factors that independent operators can control directly. This post covers the full methodology, findings, and practical implications.
How we ran the study
Delphium Labs conducted this analysis using 200 distinct restaurant queries run across ChatGPT, Perplexity, and Gemini within a 72-hour window in late January 2026. The query set was designed to mirror how real diners search for restaurants using AI, based on Delphium Labs analysis of common prompt patterns.
We structured queries across four categories:
- Cuisine-specific (60 queries): "best Italian restaurant Birmingham", "authentic Thai food Manchester", "Japanese omakase London"
- Occasion-based (50 queries): "romantic dinner restaurant Leeds", "restaurant for business lunch Bristol", "birthday dinner venue Edinburgh"
- Dietary and preference (40 queries): "vegan restaurant Brighton", "gluten-free dining options Oxford", "best vegetarian tasting menu UK"
- Neighbourhood and discovery (50 queries): "best restaurants Digbeth Birmingham", "where to eat in Ancoats Manchester", "restaurants near Southbank London"
Each query was run across all three engines (600 total responses). We recorded every restaurant mentioned, its position in the response, whether sources were cited, and whether the restaurant was part of a chain or group.
Google Business Profile is the dominant signal
Delphium Labs analysis found that Google Business Profile completeness correlated more strongly with restaurant AI visibility than any other single factor. We scored GBP completeness on a 0-100 scale based on: photos (quantity and recency), menu upload, hours accuracy, attribute completion, owner responses to reviews, regular Google Posts, and category accuracy.
Restaurants scoring above 85 on GBP completeness appeared in AI results 3.1x more often than those scoring below 40. The effect was strongest on Gemini (3.8x difference) but significant across all three engines.
This is a different dynamic from hotels. In the hotel study, structured data and website content were the leading signals. For restaurants, GBP acts as the primary data layer because most independent restaurants have minimal websites. AI engines pull from whatever data source is most complete, and for restaurants, that source is overwhelmingly Google.
Menu content changes everything
The second strongest signal Delphium Labs identified was the presence of detailed menu content on a restaurant's own website. Restaurants with full menu content published on their site appeared 2.8x more often in AI recommendations than those without it.
But the detail level mattered enormously. Delphium Labs found three tiers of menu content and their relative impact:
Basic menu (dish names and prices only): Minimal impact. This gave AI engines a list but no descriptive material to work with.
Descriptive menu (dish names, prices, and ingredient lists): Moderate impact. Restaurants at this level appeared 1.6x more than those with no menu content.
Rich menu (dish names, prices, ingredients, and narrative descriptions): Strong impact. Restaurants at this level appeared 2.8x more. These are menus where a dish is described as "slow-braised Herefordshire beef cheek with celeriac puree, charred leek, and bone marrow jus" rather than "beef cheek - 18".
The difference between descriptive and rich menus was the largest single gap in the dataset. AI engines do not just match restaurants to queries. They generate recommendations that include reasons for the recommendation. A rich menu gives the engine specific, compelling material to cite. Delphium Labs consistently observed AI responses quoting or paraphrasing dish descriptions from restaurant websites when those descriptions were detailed enough to be useful.
Specific dishes outperform generic menus
Delphium Labs found a related but distinct pattern: restaurants that highlighted specific signature dishes on their websites performed better than those presenting only a standard menu page.
When a restaurant's website featured a section like "Our signature dishes" or had individual pages for notable menu items, those restaurants appeared 2.2x more often for cuisine-specific queries. The AI engines treated these highlighted dishes as evidence of specialisation and quality.
This was especially pronounced for occasion-based queries. When a diner asked "best tasting menu Birmingham", AI engines strongly favoured restaurants that had a dedicated tasting menu page with course-by-course descriptions over restaurants that mentioned a tasting menu only as a line item in their general menu.
Dietary and allergen information drives filtered queries
Dietary queries made up 20% of the Delphium Labs test set, and the results here were striking. Restaurants with clear, specific dietary and allergen information on their websites dominated the results for these queries.
Delphium Labs analysis found that for queries like "vegan restaurant [city]" or "gluten-free dining [city]", the top-recommended restaurants almost always had one of the following: a dedicated dietary information page, menu items clearly marked with dietary labels, or a specific allergen policy page.
Restaurants without this information were largely invisible for dietary queries, regardless of whether they actually offered suitable options. A restaurant might serve excellent vegan dishes but if that information exists only in the heads of its staff and not on its website or GBP, AI engines cannot recommend it.
For dietary-specific queries, the gap was the largest in the study: restaurants with dedicated dietary content appeared 4.1x more often than those without. As dietary and allergen queries represent a growing share of how diners search, this is a visibility gap with direct revenue implications.
Review recency matters more for restaurants
In the hotel study, Delphium Labs found that review volume was the key metric. For restaurants, recency proved more important.
Restaurants with a steady flow of recent reviews (at least 10 in the past 30 days) appeared 1.9x more often than restaurants with higher total review counts but few recent additions. A restaurant with 200 total reviews but only 3 in the last month performed worse than one with 90 total reviews and 15 in the last month.
Delphium Labs attributes this to how AI engines assess restaurant relevance. Restaurants change more frequently than hotels. Menus rotate, chefs move, quality shifts. Recent reviews signal current relevance in a way that older reviews do not. AI engines, particularly Perplexity and Gemini, appear to weight review recency as a freshness signal for restaurant recommendations.
Photos and image alt text correlate with Gemini citations
Delphium Labs found a notable correlation between image content and Gemini restaurant recommendations specifically. Restaurants with more than 30 photos on their Google Business Profile and descriptive alt text on website images appeared 1.7x more often in Gemini responses.
This correlation was weaker for ChatGPT and Perplexity, but for Gemini it was consistent across query types. Delphium Labs analysis suggests this reflects Gemini's integration with Google's visual data layer. Well-photographed dishes with descriptive file names and alt text (for example, "wood-fired-margherita-pizza-sourdough-base.jpg" rather than "IMG_4392.jpg") provide additional signals that Gemini can use to match restaurants to queries.
Restaurants vs hotels: different AI visibility dynamics
Delphium Labs found several important structural differences between restaurant and hotel AI visibility.
The chain gap is smaller. In hotel queries, chains accounted for 72% of AI mentions. For restaurants, the split was closer to 58% chain / 42% independent. Independent restaurants have a more level playing field in AI results than independent hotels do.
GBP matters more, website matters less. For hotels, website quality and schema markup were the leading signals. For restaurants, GBP is dominant. This reflects the reality that most independent restaurants invest less in websites than hotels do, so AI engines rely more heavily on third-party data sources.
Locality is stronger. Restaurant queries are inherently more local than hotel queries. A diner asking "best Thai restaurant near me" expects results within a few miles. This locality bias helps independents because their GBP presence is concentrated in a specific area, while chain restaurant content is spread across hundreds of locations.
Freshness matters more. Hotels are relatively stable. Restaurants change constantly. AI engines appear to account for this by weighting recent data more heavily in restaurant recommendations.
The discovery query problem
Despite the more favourable overall split, independent restaurants face a specific challenge with broad discovery queries. When a diner asks "best restaurants in Birmingham" or "where to eat in Manchester", chain and well-known establishments still dominate.
Delphium Labs analysis found that for generic discovery queries, the top three recommended restaurants were established, high-profile venues in 84% of responses. New restaurants, recently opened venues, and neighbourhood spots rarely appeared for these broad queries regardless of their quality.
The opportunity for independents lies in specific, ownable queries. Delphium Labs found that independents performed strongly for queries like:
- "Tasting menu Birmingham" - 61% independent mentions
- "Vegan brunch Digbeth" - 72% independent mentions
- "Natural wine bar Manchester" - 68% independent mentions
- "Sunday roast with a view Leeds" - 54% independent mentions
- "Chef's counter restaurant London" - 59% independent mentions
These are queries where a specific offering or experience matches a specific diner intent. Independents who clearly articulate their distinctive features online are well positioned to own these searches.
How FindingFin restaurant audits revealed these patterns
The restaurant findings in this study emerged partly from patterns Delphium Labs observed while building FindingFin's restaurant audit capability. As we tested the tool across dozens of UK restaurants, consistent gaps appeared: missing menu content, incomplete GBP profiles, absent dietary information, and generic image filenames.
FindingFin's restaurant audit now checks for each of the visibility signals identified in this study. A property can see exactly which factors are present, which are missing, and what the expected impact of fixing each gap would be. The audit is specific to each restaurant's actual content and competitive context, not a generic checklist.
Practical takeaways
Based on Delphium Labs analysis of 200 restaurant queries across three AI engines, these are the highest-impact actions for restaurant operators:
Maximise your Google Business Profile. Upload at least 30 high-quality photos with descriptive file names. Complete every attribute field. Post updates at least weekly. Upload your menu directly to GBP. Respond to reviews within 48 hours. This is the single most impactful investment for restaurant AI visibility.
Publish your full menu on your website with rich descriptions. Every dish should have ingredients listed and, for signature items, a sentence or two of descriptive context. "Pan-seared sea bass with samphire, brown shrimp butter, and new potatoes" outperforms "sea bass - 22" in every measurable way.
Create dedicated dietary and allergen content. If you cater to vegan, vegetarian, gluten-free, or other dietary needs, make that explicit on your website and GBP. A dedicated page or clearly marked menu section makes you visible for a growing category of AI queries.
Highlight your signature dishes and experiences. If you offer a tasting menu, a chef's counter, a Sunday roast, or any other distinctive experience, give it its own page or prominent section. AI engines use these as matching signals for specific queries.
Maintain review flow. Encourage reviews from recent diners consistently. Ten fresh reviews per month is more valuable than a back catalogue of hundreds of old ones. The recency signal matters more for restaurants than almost any other hospitality category.
Name your images properly. Rename dish photos from camera defaults to descriptive names before uploading. Add alt text to website images. This is a small effort with a measurable impact, particularly on Gemini.
Delphium Labs will continue tracking restaurant AI visibility as engines evolve and as dining search behaviour shifts further toward conversational AI. The data from this study represents a snapshot, but the structural patterns - GBP dominance, menu content depth, dietary specificity - are likely to remain stable signals for the foreseeable future.