- 83% of restaurants and the majority of hotels are entirely invisible in the recommendations of ChatGPT, Perplexity and Gemini (Uberall 2026 study).
- AIs recommend establishments with 3.4 times more reviews than the non-recommended ones. Thresholds observed: ChatGPT ≥ 4.3 stars, Perplexity ≥ 4.1, Gemini ≥ 3.9.
- The correct diagram: Hotel (not LocalBusiness) for a hotel, Restaurant for a restaurant. 41.1% of hotels are using the wrong schema.org type and going unnoticed.
- Winning combo: Hotel/Restaurant schema + FAQPage + reviews ≥ 4.3 + consistent NAP everywhere + complete Google Business Profile. First results within 4 to 8 weeks.
A client is looking for a hotel in Bordeaux for a weekend with his family. In 2024, he typed “family hotel Bordeaux center” into Google. In 2026, he opens ChatGPT and asks: “Find me a quiet boutique hotel in Bordeaux for two nights, with breakfast included, around €200.” The AI responds with two or three names.
Is yours on the list?
For 83% of restaurants and a good half of French independent hotels, the answer is no. They are indexed in Google, present on Booking and TripAdvisor, they have their own site, and yet they do not exist in the AI conversation which now decides on a growing portion of reservations. The worst: their competitors lower ranked on Google, but better structured for AI, win the recommendations.
Here is the game-changing guide. Why AIs choose who to recommend, the exact diagram to put on your establishment profile, the citation ecosystem to master (NAP, Google Business Profile, TripAdvisor, TheFork), the critical role of reviews, JSON-LD copy and paste examples for a hotel and for a restaurant, and an action plan in 30-60-90 days.
Why 83% of restaurants are invisible in AI
The figure comes from the Uberall 2026 study, and it is massive: out of all the American and European restaurants analyzed, Only 17% ever appear in a ChatGPT response when a consumer asks “where to eat near me tonight”. On the hotel side, the Hotelrank Research study arrives at a similar result: the majority of independent establishments do not generate any AI mentions, even with a good classic Google presence.
The most telling example comes from a boutique hotel in Charleston, United States. This hotel appears on page one of Google for “historic downtown Charleston hotels”. Page one. And yet, on the same request in conversation with ChatGPT or Perplexity, it generates zero mention. Meanwhile, lower-ranked competitors on Google capture 100% of AI recommendations, because their structured data and entity signals are better done.
The same logic applies to restaurants. The Hospitality Technology study from February 2026 compared restaurants recommended by AI to those not recommended on the same query: the recommended ones displayed on average 3,424 Google reviewsnon-recommended only 955. A postman 3.6. Stars alone are not enough: it is the cumulative volume of reviews that weighs, because the AIs combine volume and rating to decide who to mention.
The raw conclusion: a hotel or restaurant that does not explicitly work on its GEOGenerative Engine Optimization, the optimization of content to be cited by AI response engines (ChatGPT, Perplexity, Gemini). now leaves dozens, sometimes hundreds of reservations on the table each month.
The three major engines do not use the same sources. This is the nuance that most guides miss, and it’s what dictates strategy.
- ChatGPT is mainly based on third-party directories : 48% of its citations in travel and dining responses come from TripAdvisor, Booking, TheFork, Yelp and Google Maps. Practical consequence: being present and well rated on these platforms weighs more than having a beautiful site.
- Gemini favors branded sites : 52% of quotes come from the establishment’s official website. Consequence: your own site, its schema, its structured establishment file count more for Gemini than for ChatGPT.
- Perplexity is the most demanding: he cites on average 22 sources per answerby cross-referencing several directories, the official website and the local press. You have to be everywhere to exist at home.
Beyond the sources, the three engines seek the same signals:
- NAP consistency (Name, Address, Phone): exactly the same name, same address, same phone everywhere. A variation (“Hotel Le Tilleul” vs “Le Tilleul Hôtel”) and the AI hesitates, so cites the more stable competitor.
- Credible review volume : the thresholds measured are ChatGPT ≥ 4.3 stars, Perplexity ≥ 4.1, Gemini ≥ 3.9. Below these thresholds you are not recommended.
- The most specific diagram possible : Hotel or Restaurant (not Organization, not generic LocalBusiness), with aggregateRating and amenityFeature.
- Freshness of information : up-to-date schedules, recent photos, updated menu. AI penalizes files that appear asleep.
The citation ecosystem: what AI really crosses
An AI does not read your site and conclude. She compares at least five sources before recommending you, because she needs to verify that she isn’t making you up. For a French hotel or restaurant in 2026, the ecosystem to control looks like this.
The 7 sources that an AI crosses before recommending you
-
Google Business Profile
The pivot plug
Address, hours, photos, reviews, FAQ. This is the source that all AI consults first. Incomplete or obsolete, it is unacceptable.
-
TripAdvisor
The hotel and restaurant authority
Source cited by ChatGPT and Perplexity in priority on travel recommendations. Recent volume and rating count as much as the average.
-
Booking, Expedia, Hotels.com
Hotel OTAs
For hotels only. ChatGPT massively cites OTAs. A complete profile, professional photos, recent reviews, is non-negotiable.
-
TheFork
The FR restaurant network
For restaurants in France. Allows direct booking, aggregates reviews. Essential for AI recommendations on French restaurants.
-
Your official website
The entity reference
Source most heavily weighted by Gemini (52% of citations). With a complete Hotel or Restaurant schema, it defines your brand for AI.
-
Tourist office and local press
Third-party authority
A mention in the regional tourist office or an article from South-West, Le Progrès, La Provence weighs heavily in the authority score.
-
Social media and video reviews
The signal of freshness
Active Instagram and TikTok, recent YouTube mentions: signal that the establishment is alive. Weighs less individually but cumulatively, it is decisive.
The golden rule of this ecosystem: absolute consistency. The same business name, the same mailing address, the same phone number, exactly, on all seven sources. Any discrepancy, even minor, breaks your identity as an entity in the eyes of the AIs.
The trap of the wrong @type: Hotel vs LodgingBusiness vs LocalBusiness
This is the most common mistake, and probably the most costly. According to 2026 analyses, 41.1% of hotels using JSON-LD are using the wrong schema type : they declare Organization Or LocalBusiness instead of Hotel Or LodgingBusiness. The same goes for restaurants: many say LocalBusiness instead of Restaurant.
Why is it serious? The schema.org vocabulary is hierarchical: Thing > Organization > LocalBusiness > LodgingBusiness > Hotel. The lower you go in the hierarchy, the more you give contextThe context window: the maximum number of tokens that a model can process in a single request (eg: 200k, 1M). AI-accurate semantics. A diagram Hotel unlocks specific properties like starRating, checkinTime, petsAllowed, numberOfRoomswhich are exactly what the AI is looking for to respond to a travel request.
Hierarchy to remember:
- Classic hotel:
@type: "Hotel" - Guesthouse, B&B, rooms:
@type: "BedAndBreakfast" - Hotel-residence or aparthotel:
@type: "ApartmentComplex"OrHotelwithaccomodationCategory - Classic restaurant:
@type: "Restaurant" - Wine bar, brasserie:
@type: "BarOrPub" - Bakery, tea room:
@type: "Bakery"OrCafeOrCoffeeShop - Pizzeria, fast food:
@type: "FastFoodRestaurant"
The reflex is simple: find the most specific type that corresponds to your establishment. For the majority of cases, Hotel Or Restaurant are enough. To go further on the choice of schemas, read our complete schema.org guide for GEO.
Hotel: the complete schema.org file
Here is a complete example for a fictitious boutique hotel in Bordeaux. To be adapted with your real data.
{
"@context": "https://schema.org",
"@type": "Hotel",
"@id": "https://le-tilleul-bordeaux.fr/#hotel",
"name": "Hôtel Le Tilleul",
"url": "https://le-tilleul-bordeaux.fr",
"telephone": "+33556001234",
"email": "contact@le-tilleul-bordeaux.fr",
"image": ["https://le-tilleul-bordeaux.fr/cover.jpg"],
"description": "Hôtel boutique 4 étoiles au cœur du centre historique de Bordeaux, 18 chambres avec parquet d'époque, jardin intérieur et petit-déjeuner local inclus.",
"starRating": { "@type": "Rating", "ratingValue": "4" },
"priceRange": "120-280 €",
"numberOfRooms": "18",
"petsAllowed": true,
"checkinTime": "15:00",
"checkoutTime": "11:00",
"address": {
"@type": "PostalAddress",
"streetAddress": "7 rue Vital Carles",
"addressLocality": "Bordeaux",
"postalCode": "33000",
"addressCountry": "FR"
},
"geo": {
"@type": "GeoCoordinates",
"latitude": 44.8404,
"longitude": -0.5805
},
"aggregateRating": {
"@type": "AggregateRating",
"ratingValue": "4.6",
"reviewCount": "847",
"bestRating": "5"
},
"amenityFeature": [
{ "@type": "LocationFeatureSpecification", "name": "Wi-Fi gratuit", "value": true },
{ "@type": "LocationFeatureSpecification", "name": "Petit-déjeuner inclus", "value": true },
{ "@type": "LocationFeatureSpecification", "name": "Climatisation", "value": true },
{ "@type": "LocationFeatureSpecification", "name": "Parking privé", "value": false }
],
"sameAs": [
"https://www.tripadvisor.fr/Hotel_Review-g187079",
"https://www.booking.com/hotel/fr/le-tilleul-bordeaux.html",
"https://www.google.com/maps/place/?cid=12345678901234567890"
]
}
The properties that make the difference for AI:
@idanchored on the domain: allows you to link your hotel to the same entity everywhere (home page, room profile, contact page).aggregateRatingwithratingValueAndreviewCount: this is what the AI reads first to decide if you pass the threshold.amenityFeature: helps match specific queries like “pet-friendly hotel” or “hotel with breakfast included”.priceRange: structures your pricing positioning for requests filtered by budget.sameAs: confirms your entity identity by pointing to your TripAdvisor, Booking, Google Maps records.
Restaurant: the complete schema.org file
For a fictitious gourmet restaurant in Lyon:
{
"@context": "https://schema.org",
"@type": "Restaurant",
"@id": "https://atelier7-lyon.fr/#restaurant",
"name": "Atelier 7",
"url": "https://atelier7-lyon.fr",
"telephone": "+33478001234",
"image": ["https://atelier7-lyon.fr/salle.jpg"],
"description": "Restaurant gastronomique 1 étoile Michelin à Lyon, cuisine de saison signée Chef Alice Martin, 25 couverts, menu unique en 5 services.",
"servesCuisine": ["Française", "Bistronomique", "Saisonnière"],
"priceRange": "75-120 €",
"menu": "https://atelier7-lyon.fr/menu",
"acceptsReservations": true,
"hasMenu": "https://atelier7-lyon.fr/menu",
"address": {
"@type": "PostalAddress",
"streetAddress": "7 rue Mercière",
"addressLocality": "Lyon",
"postalCode": "69002",
"addressCountry": "FR"
},
"geo": {
"@type": "GeoCoordinates",
"latitude": 45.7619,
"longitude": 4.8334
},
"openingHoursSpecification": [
{
"@type": "OpeningHoursSpecification",
"dayOfWeek": ["Tuesday", "Wednesday", "Thursday", "Friday", "Saturday"],
"opens": "19:30",
"closes": "22:30"
}
],
"aggregateRating": {
"@type": "AggregateRating",
"ratingValue": "4.7",
"reviewCount": "1342"
},
"sameAs": [
"https://www.thefork.fr/restaurant/atelier-7",
"https://www.tripadvisor.fr/Restaurant_Review-g187265",
"https://www.google.com/maps/place/?cid=98765432109876543210"
]
}
The properties that weigh on the restaurant side:
servesCuisine: structures the type of cuisine for filtered queries (“Italian restaurant Lyon”, “bistronomic Marais”).priceRange: helps AI match with “good restaurant for less than €50” or “gourmet for birthday”.openingHoursSpecification: essential for contextual queries (“where to dine tonight”, “brunch Sunday lunch”).acceptsReservations: strong practical signal, especially for high-end restaurants.hasMenu: link to the map. AIs can access it to refine their recommendations.
NAP, Google Business Profile, TripAdvisor: cross-source consistency
The NAP (Name, Address, Phone) is the triad that AIs check to confirm that you are the same entity everywhere. Any inconsistency, no matter how small, causes you to lose the quote to a more disciplined competitor.
Three rules for a clean NAP:
1. One exact business name. Choose the official form (“Hôtel Le Tilleul” for example) and use it strictly identical on the site, Google Business Profile, TripAdvisor, Booking, TheFork, the tourist office, your invoices. No “Le Tilleul” on Booking and “Hotel Le Tilleul ★★★★” on TripAdvisor.
2. A standardized postal address. Full French format: number, route, postal code, city, without abbreviation. “7 rue Vital Carles, 33000 Bordeaux”. Not “7 r. V. Carles” elsewhere.
3. A number in international format. +33 5 56 00 12 34 everywhere, never 05.56.00.12.34 on certain platforms and +33 556 001234 on others. AIs look for the exact numeric ID, formatting matters.
Google Business Profile deserves special attention. This is the pivot sheet that all AI consults first. To be completed in full: name, main category (Hotel or Restaurant, not generic LocalBusiness), address, telephone, full hours, website, up to 10 recent professional photos, FAQ with 5 to 8 questions and answers, attributes (Wi-Fi, disabled access, payments accepted).
TripAdvisor and OTAs. For a hotel, mandatory presence on Booking, Expedia, Hotels.com in addition to TripAdvisor. For a restaurant, TheFork and TripAdvisor minimum. Professional photos, careful descriptions, systematic responses to reviews. A sparsely completed form sends a negative signal to AI, just like it would to a human.
The critical role of reviews (and their volume)
The Hospitality Technology 2026 study established a painful observation: on the query “best restaurant + city”, restaurants recommended by AI have an average of 3,424 Google reviews compared to 955 for non-recommended ones. A factor of 3.6 on the cumulative volume.
The surprising detail: the average note weighs less than volume. A 4.8-star restaurant with 200 reviews will be beaten by a 4.4-star competitor with 1,800 reviews. For what ? Because AIs are looking for a signal of statistical reliability : 1,800 convergent opinions on 4.4 are more credible than 200 opinions on 4.8.
Thresholds per engine (measured on local 2026 queries):
| Engine | Average threshold | Involvement |
|---|---|---|
| ChatGPT | ≥ 4.3 stars | Below, rare recommendation |
| Perplexity | ≥ 4.1 stars | More tolerant but cites more sources |
| Gemini | ≥ 3.9 stars | The most permissive, but crosses the volume |
Concrete strategy: aim Minimum 1,000 Google reviews for a restaurant that wants to be mentioned, 300 minimum for a boutique hotel, 500+ for a 3-4 star hotel. How to get there without cheating:
- Systematically ask satisfied customers to leave a review at the time of check-out or payment (QR code on the ticket, automatic email on D+2).
- Respond to 100% of reviews, positive and negative, within 48 hours. AI penalizes listings whose reviews remain unanswered.
- Distribute between Google and TripAdvisor (both weightThe numerical parameters of a neural network, adjusted during training to learn patterns in the data. heavy for AI), plus TheFork for restaurants.
Concrete case: what separates a cited hotel from an ignored hotel
To make it tangible, let’s compare two fictitious but realistic establishments, in the same category and the same city.
Hotel A, Le Tilleul Bordeaux. 4 stars, 18 rooms, 4.6/5 on Google (847 reviews), 4.5 on TripAdvisor (612 reviews), 8.9 on Booking (1,040 reviews). Site with diagram Hotel complete (JSON-LD like the example above), Google Business Profile filled to 100%, professional photos updated 2 months ago, response to 100% of reviews, FAQ on the site with 7 questions, active Instagram presence. Result: cited by ChatGPT, Perplexity and Gemini on the queries “boutique hotel Bordeaux center”, “quiet hotel Bordeaux”.
Hotel B, Bristol Bordeaux. 4 stars too, 22 rooms, 4.2/5 on Google (210 reviews only), 4.0 on TripAdvisor (180 reviews), 8.3 on Booking. Site with diagram LocalBusiness (bad type), Google Business Profile 60% filled, last photos in 2024, response to reviews sporadic, no FAQ, Instagram dormant. Result: page 2 of Google on “Bordeaux center hotel”, zero mention in AI responses.
Both hotels have the same category, the same city, and even the same approximate rating. What separates them: volume of reviews, precise diagram, complete GBP profile, recent photos, active social presence. It is this accumulation of signals which decides the AI citation today.
The iceberg: what the customer sees vs. what the AI exploits
The sheet you present
- Beautiful professional photos of the bedroom or hall
- An engaging description of the place and its atmosphere
- Average rating and some recent reviews
- Contact details and a booking link
The entity signal bundle
- The exact @type schema.org (Hotel/Restaurant, not LocalBusiness)
- NAP consistency across 5 to 7 third-party sources
- The cumulative volume of Google + TripAdvisor + Booking + TheFork reviews
- The response rate to reviews and the freshness of photos
- Specific attributes (amenityFeature, servesCuisine, openingHours)
- External mentions (tourist office, local press, travel blogs)
The visible layer is what sells once the customer is on your listing. The hidden layer is what decides whether the customer lands on your listing in the first place. In 2026, it is the second that matters most for conversion.
The 7 mistakes that make an establishment invisible
Seven recurring pitfalls observed in GEO audits of French hotels and restaurants.
-
Bad @type schema.
LocalBusinessinstead ofHotelOrRestaurant. You lose the specific properties (starRating,servesCuisine) that the AIs are looking for. -
Inconsistent NAP. Slightly different name between site, Google and Booking. Abbreviated address on one side, full address on the other. The AI isn’t sure it’s you, it cites a less ambiguous competitor.
-
Google Business Profile 60% completed. No FAQ, photos over a year old, no special hours, no attributes. This is the pivot plug, it must be at 100%.
-
Unanswered reviews. AI penalizes listings with a review response rate below 60%. Responding to all reviews within 48 hours is the basis.
-
Outdated photos. The latest pro photos are from 2024 or earlier. For AI, this is the signal that an establishment “no longer lives”. A recent photo session every 12 to 18 months is mandatory.
-
Absence on TheFork or Booking. For a French restaurant that is not on TheFork, that’s 30 to 40% fewer citations on ChatGPT and Perplexity. For a hotel without full booking, it’s even more brutal.
-
No structured FAQ. An FAQPage on your site with 5 to 8 questions and answers (40-60 words each) increases the probability of an AI citation by approximately 28% (Waymarker 2026 study). And yet, the majority of HORECA sites do not have one.
30-60-90 day action plan
D 0 to 30
Audit + Google Business Profile + NAP
Audit the 7 sources (site, GBP, TripAdvisor, Booking or TheFork, tourist office, OTAs, networks). Standardize NAP everywhere (same name, same address, same phone). Complete Google Business Profile 100%: recent professional photos, hours, FAQ, attributes, correct category.
Foundations~10 cumulative hours
D 0 to 60
Schema.org Hotel or Full Restaurant
Add the correct @type (Hotel or Restaurant), with aggregateRating, address, geo, amenityFeature or servesCuisine, openingHours, priceRange, sameAs to your third-party files. Add an FAQPage with 5-8 stand-alone answers (40-60 words). Validate on the Schema Markup Validator.
Technical~1 day of development
D 0 to 90
Boost volume of reviews and quality of response
Set up the post-stay or post-meal QR code to request a Google review. Aim for 50 to 100 new reviews over 90 days. Respond to 100% of reviews within 48 hours, both positive and negative. Update photos on GBP, TripAdvisor and OTAs.
Reputationcontinuous
D 30 to 60
Social presence and freshness
Reactivate Instagram (3 to 5 posts per week, daily stories in season) and TikTok for the young target. If possible, a mention in the local press or a travel blog is the most powerful accelerator. Ask the regional tourist office for a link.
Fame~5 hrs/week
D 60 to 90
Measure and iterate
Take the monthly test of the three AIs with 10 typical queries for your target (“quiet hotel Bordeaux center”, “gastronomic Lyon 2nd”, “Sunday brunch Marais”). Note if you are cited. Compare with competitors who are. Adjust according to what they are missing or what you are missing.
Measure1 hour/month
Measure without Search Console: the monthly testing protocol
There is no Search Console (yet) for ChatGPT, Perplexity or Gemini. Manual testing remains the most reliable measure.
4-step protocol, to be reproduced each month:
-
List 10 typical queries what a target customer would ask. Vary the angles: generic (“boutique hotel Bordeaux”), thematic (“romantic hotel Bordeaux”), occasion (“Bordeaux family weekend”), practical (“hotel near Saint-Jean train station”).
-
Ask the 10 requests to ChatGPT, Perplexity, Gemini and Claude Search, in incognito mode (without history which biases recommendations). Note if you are cited, where, and who appears in your place.
-
Analyze the competitors mentioned. Why them? More reviews? Best cover photo? More complete diagram? Recent article in the local press? This analysis is your list of projects for the following month.
-
Iterate and re-test on D+30. The goal is not to correct everything at once, it is to close one gap per month until it appears in 7 or 8 out of 10 queries.
To understand how each engine cites its sources and adapt your strategy, see our comparison of Perplexity vs Gemini Search vs ChatGPT Search.
What to remember
83% of restaurants and half of independent hotels are invisible in AI responses in 2026. This is not inevitable, it is an optimization delay that the first to correct transform into a sustainable competitive advantage.
The method consists of three stages. The foundations : strictly consistent NAP, 100% Google Business Profile, recent photos, complete Hotel or Restaurant diagram (never generic LocalBusiness). The authority layer : volume of reviews (1,000+ for a restaurant, 500+ for a hotel), systematic response to all reviews within 48 hours, active presence on TripAdvisor, Booking or TheFork. The accelerator : external mentions (tourist office, local press, travel blogs), active social networks, permanent freshness of information.
The first results are seen between 4 and 8 weeks. Stable AI quotes are established between 3 and 6 months. This is exactly the timing of a classic SEO audit, except that this time, the return is not measured in clicks but in direct reservations.
To go further on the overall GEO strategy, read our complete guide: how to be visible in ChatGPT in 2026. For technical details of schemas and the entity graph, see our complete schema.org guide for GEO. And to prepare your site for AI agents, read our practical guide LLMs.txt 2026.

