{"@context":"https://schema.org","@type":"Blog","name":"Nicolas Sitter","url":"https://nicolassitter.com","author":{"@type":"Person","name":"Nicolas Sitter","url":"https://nicolassitter.com/about","sameAs":["https://www.linkedin.com/in/nicolassitternolleau/","https://github.com/Nicositter88","https://hotelrank.ai"]},"blogPosts":[{"@type":"BlogPosting","headline":"Gemini Answers Hotel Questions From a Booking Tool, Not the Web (2026)","description":"Across 450 questions in September 2026, every recommendation-shaped hotel question sent to Gemini was answered by the Google Hotels tool and returned zero cited publishers; explanatory questions about the same hotels in the same cities returned 5.94 each. Six accommodation nouns behave alike, so it is not lexical. Six non-accommodation controls never reached the Hotels tool - they reached Google Maps and lost their citations the same way. The phrasing decides whether a tool fires; the category decides which one.","datePublished":"2026-09-08T00:00:00.000Z","dateModified":"2026-09-08T00:00:00.000Z","url":"https://nicolassitter.com/research/gemini-hotels-tool-study-2026","category":"research"},{"@type":"BlogPosting","headline":"Pick a Number: AI Engines Swept Across Five Ranges (2026) — 1,798 Draws","description":"The same question fired at five ceilings — pick a number between 1 and 10, 20, 30, 50 and 100 — through Bright Data, with nothing changed between rows but the number after “between 1 and”. Four engines (ChatGPT, Gemini, Microsoft Copilot, Google AI Mode) × three wordings × 30 draws per cell, one number per capture with a fresh context every time. 1,798 analysed draws: the 360 published for 1–100 on 2026-09-07 plus 1,438 of 1,440 captures fired at the other four ceilings on 2026-09-09. The most frequent answer runs 7 (97.5%), 7 (48.3%), 17 (85.3%), 37 (55.9%) and then 42 (46.7%) — four of five ceilings won by a number ending in a seven, and four different numbers, so what scales is the kind of number rather than the token. Below a ceiling of 100, 64.4%–97.5% of draws end in a seven and 83.9%–98.3% are odd, against a uniform 10% and 50% that hold at every ceiling in the sweep. Six draws in 1,798 were a multiple of five (0%–1.1% per row) where matched controls returned 373 (18.1%–22.8%). 1–100 is the outlier on every axis: ends-in-seven 41.4% and odd 52.2%, both the lowest of the five, because 42 takes 46.7% of that row and is even. The 1–50 row reproduces the published 1–50 attractor 27 at 27.1% and is won by 37 at 55.9%; 42 is still in range there at 12.6%.","datePublished":"2026-09-07T00:00:00.000Z","dateModified":"2026-09-10T00:00:00.000Z","url":"https://nicolassitter.com/research/ai-random-number-2026","category":"research"},{"@type":"BlogPosting","headline":"AI Search for Tango Schools in Buenos Aires (2026): The Canon Is Bilingual","description":"The series' first South American city, and its first classes-and-lessons vertical since the Paris and Berlin yoga studies: 24 prompt templates × EN/ES × US/AR proxies × 5 engines (ChatGPT, Perplexity, Gemini, Copilot, Google AI Mode), 392 of 480 captures, captured 2026-09-02 (Perplexity 8/96 after two failed batches and a targeted refire — disclosed, not imputed). The headline breaks the series' tourist-city pattern: English and Spanish answers to the control prompt share 43% of their top-5 where Istanbul and Mexico City both measured 0%, and the best-milongas prompt returns the identical five venues in both languages. La Viruta is named in 240 of 392 answers (61.2%, all five engines), while citation counting crowns Escuela Mundial de Tango (score 131). ChatGPT's own-website share is 25.3% — the fifth city measured on the density curve, filling its empty middle and keeping all five in perfect order — it cites the Buenos Aires city government (94) more than all venue websites combined (93), and its Reddit share is zero. Gemini routes 24.2% of its citations to dinner-show ticketing agencies, led by showdetango.com at 81. 2,025 citations, 2,281 extracted mentions (83.8% resolved) against a 441-row registry with 257 real venues.","datePublished":"2026-09-07T00:00:00.000Z","dateModified":"2026-09-07T00:00:00.000Z","url":"https://nicolassitter.com/research/tango-schools-buenos-aires-ai-search-2026","category":"research"},{"@type":"BlogPosting","headline":"What Selects Google AI Mode's Entity Wrapper (2026): 44 Places Types × 10 Query Forms","description":"A crossed factorial of 44 Google Places Table-A primary types, 10 query forms and 4 cities (New York, Berlin, Seoul, Helsinki): 1,868 Google AI Mode captures replicated two hours later at 1,496. Holding the noun and the city constant, the pooled wrap rate of citations carried in google.com/searchviewer falls from 0.876 on a \"best X in Y\" list to 0.198 on \"why are there so many X in Y?\" and 0.000 on an informational question, while removing only the city still leaves 0.663. Geographic areas are the one exempt family, falling from 0.273 to 0.086 between waves. Overall AI Mode wraps 76.2% of citations in wave one, which scopes the previously published 99.96% to the hotel-list corpus it was measured on.","datePublished":"2026-09-06T00:00:00.000Z","dateModified":"2026-09-06T00:00:00.000Z","url":"https://nicolassitter.com/research/google-ai-mode-places-category-census-2026","category":"research"},{"@type":"BlogPosting","headline":"AI Traffic Is Undercounted (2026): Four Measurements of One Hotel","description":"Of 35 hotel visitors who said an AI assistant sent them, 24 carried no AI signal at all — neither a referrer nor a campaign tag. ChatGPT tags its outbound links, so 11 were traceable; the 12 who arrived from google.com were not. Over the same 62 days, 96.4% of the hotel's Google clicks were people typing its name, and unbranded discovery queries converted at 0.26%.","datePublished":"2026-09-02T00:00:00.000Z","dateModified":"2026-09-02T00:00:00.000Z","url":"https://nicolassitter.com/research/ai-traffic-undercounted-hotel-2026","category":"research"},{"@type":"BlogPosting","headline":"Google AI Mode Stopped Linking Publishers (2026): Decoding the searchviewer svid","description":"Google AI Mode replaced publisher citation links with google.com/searchviewer wrappers — 29,556 of 29,568 hotel citations (99.96%) in the run of 31 August 2026. The svid parameter is base64url protobuf wrapping a second base64url protobuf carrying a Google Knowledge Graph machine ID, decoded for 100.000% of wrapped citations with zero failures, recovering 6,792 distinct entities. The rollout was gradual: 37.7% on 18 May, 61-72% through the summer, 100.0% on 31 August.","datePublished":"2026-09-01T00:00:00.000Z","dateModified":"2026-09-01T00:00:00.000Z","url":"https://nicolassitter.com/research/google-ai-mode-searchviewer-svid-2026","category":"research"},{"@type":"BlogPosting","headline":"Best Western France: Does ChatGPT Know the Network? (2026)","description":"A brand AI-visibility audit of Best Western France's 322-hotel/242-city network (unaffiliated with Best Western): 8,370 unbranded prompts, 23,220 captures, French/FR-proxy as the primary pass plus an English/US-proxy add-on for Paris+PACA. Best Western surfaces organically 64.8% of the time when the query never names the brand — with a 28-point single-vs-multi-hotel-city gap, a near-disappearance on budget queries (33.8%), a business-traveler skew (80.7% vs 56.2% for couples), and two cities (Honfleur, Lyon) with zero organic mentions. Chain co-occurrence added 2026-09-03: every lift falls between 0.18 and 1.14, so answer-sharing is almost entirely base-rate driven, with French mid-market chains just above chance and upscale/international brands below it.","datePublished":"2026-08-31T00:00:00.000Z","dateModified":"2026-09-03T00:00:00.000Z","url":"https://nicolassitter.com/research/best-western-france-ai-visibility-2026","category":"research"},{"@type":"BlogPosting","headline":"AI Search for Saunas in Helsinki (2026): The First City Where Every AI Engine Agrees","description":"The series' first Nordic city and first Finnish-language study: 24 prompt templates × EN/FI × US/FI proxies × 5 engines (ChatGPT, Perplexity, Gemini, Copilot, Google AI Mode), 475 of 480 captures, captured 2026-08-26. The headline is a series first: English and Finnish answers to the control prompt share the identical top-5 saunas — 100% overlap where the previous two studies measured 0% — and the language→TLD coupling reads a neutral 0.92×. Löyly is named in 379 of 475 answers (79.8%), the strongest single-venue consensus in ten studies, and holds a clean dual-metric double (cite score 162). ChatGPT cites sauna-owned websites 45.5% of the time — a series high completing the density arc (CDMX 12.5% density → 0% share, Seoul 13% → 2.3%, Istanbul 22.1% → 21.3%, Helsinki 78.4% → 45.5%) — while its Reddit share collapses from a 17–20% band to 0.4%. myhelsinki.fi, the city's own tourism guide, is the most-cited non-Google domain (246 citations, all five engines). 3,493 citations, 3,017 extracted mentions (94.5% resolved — series high) against a 395-row registry with 125 real sauna venues.","datePublished":"2026-08-26T00:00:00.000Z","dateModified":"2026-08-26T00:00:00.000Z","url":"https://nicolassitter.com/research/saunas-helsinki-ai-search-2026","category":"research"},{"@type":"BlogPosting","headline":"Apart-Hotels in Paris on ChatGPT (2026): ChatGPT Doesn't Have an Aparthotel Category — It Has a Duration Heuristic","description":"The standing weekly Paris hotel panel almost never surfaced aparthotel chains — Adagio, the city's largest operator, appeared zero times across 1,305 map entities over three months. A dedicated 68-prompt panel (EN+FR, 5 repeats, FR proxy, 340 captures) found the chains were never invisible: tracked brands appear in 98.7% of mapped answers once asked directly. But 56.1% of what surfaces is still typed plain 'Hôtel', and the split runs along brand lines inherited from Google (Adagio 97.8% 'Hôtel', Appart'City 100% 'Résidence hôtelière') rather than property type. Length of stay is the strongest lever: three-month prompts return 70.5% apart-typed venues vs a 39.7% baseline; a gym-and-pool amenity request returns zero. The category taxonomy is locale-locked to the proxy country, not the prompt language, and repeat consistency across 5 identical asks is a mean Jaccard of just 0.270.","datePublished":"2026-08-25T00:00:00.000Z","dateModified":"2026-08-25T00:00:00.000Z","url":"https://nicolassitter.com/research/aparthotels-paris-ai-search-2026","category":"research"},{"@type":"BlogPosting","headline":"Is LinkedIn Useful for AI Visibility in Hotel Search?","description":"A full backfill of a new linkedin_count column across 105,377 AI hotel-search captures (all engines, all-time). LinkedIn appears in 197 captures (0.19%). ChatGPT cites LinkedIn roughly 60% of the times it retrieves it, mostly company pages. Grok retrieves LinkedIn social posts far more often but converts only about 16% to citations. Google AI Mode, Copilot and Perplexity barely touch it; Gemini never does in this dataset.","datePublished":"2026-08-24T00:00:00.000Z","dateModified":"2026-08-24T00:00:00.000Z","url":"https://nicolassitter.com/research/linkedin-hotel-ai-citations-2026","category":"research"},{"@type":"BlogPosting","headline":"Did GPT-5.6 Really Kill Listicles? Not in Hotel Search","description":"A publicly shared analysis (Tomek Rudzki / Peec AI, 1M prompts, ~180M sources) reported that after GPT-5.6, ChatGPT fan-outs per chat rose ~154%, site: search adoption jumped to 18.4% of chats, and listicle/comparison citation share fell 50.5%/32.1%. We re-ran the same before/after comparison on our own 616-prompt AI Hotel Landscape corpus, split at our own observed model transition (GPT-5.5 through Aug 3, GPT-5.6 from Aug 10 — bracketing August 6, ChatGPT.com's consumer-default rollout date; July 9 was the developer/API release). None of the reported effects replicated in this vertical.","datePublished":"2026-08-24T00:00:00.000Z","dateModified":"2026-08-24T00:00:00.000Z","url":"https://nicolassitter.com/research/gpt-5-6-hotel-search-changes-2026","category":"research"},{"@type":"BlogPosting","headline":"AI Search for Tacos in Mexico City (2026): Michelin Beats Every Taquería on the Internet","description":"The series' first Americas city, first Spanish-language study, and first complete five-engine grid since Berlin tattoo: 25 prompt templates × EN/ES × US/MX proxies × 5 engines (ChatGPT, Perplexity, Gemini, Copilot, Google AI Mode), 500 of 500 captures, captured 2026-08-05. Two zeros anchor the findings: ChatGPT cited taquería-owned websites 0 times in 356 citations (the first absolute zero in nine studies), and the English vs Spanish top-5 lists on the control prompt share zero venues. The vacuum belongs to professional guides: guide.michelin.com is the most-cited non-Google domain (200 citations, reaching all five engines), Mexican critic Marco Beteta's mbmarcobeteta.com takes 147, and Gemini gives the two of them 16.8% of its 1,187 citations. Copilot's Seoul entity-share collapse replicates at full n: 38.8% on a complete 100-capture batch vs its 74–97% series band. Consensus winner: El Vilsito, the Narvarte auto shop turned al pastor taquería, named in 174 of 500 answers on all five engines — while domain-matched cite-counting crowns Taquería Orinoco (51) and hands Michelin-starred El Califa de León a zero. 5,867 citations, 3,026 extracted mentions (86.6% resolved) against an 828-row registry.","datePublished":"2026-08-19T00:00:00.000Z","dateModified":"2026-08-19T00:00:00.000Z","url":"https://nicolassitter.com/research/tacos-mexico-city-ai-search-2026","category":"research"},{"@type":"BlogPosting","headline":"Raw-Stream Hidden Signals: What ChatGPT Actually Sends","description":"A read-only re-test of three claimed undocumented fields inside the raw ChatGPT server-sent-event stream, mined against our own AI Hotel Landscape captures (ai-scrapers’ public.fanout_captures table), not a third-party dataset. Partially confirmed: ChatGPT’s winner-vs-runner-up supporting_websites citation field (141 groups, 77% of captures, but every snippet is empty so the match is title-only). Refuted as clean structural zeros in this pipeline: ChatGPT narrating brand priors in “thoughts” (0/146), and ChatGPT model escalation via default_model_slug (0/146 qualify — always “auto”). A companion Perplexity re-test (trust tiers, intent-classifier scorecard, mode/model escalation) is scoped out of this article — ai-scrapers’ own storage layer strips Perplexity’s raw SSE field from every stored capture, confounding that result until a cleaner re-test lands.","datePublished":"2026-08-13T00:00:00.000Z","dateModified":"2026-08-24T00:00:00.000Z","url":"https://nicolassitter.com/research/raw-stream-hidden-signals-2026","category":"research"},{"@type":"BlogPosting","headline":"AI Search for Barbershops in Istanbul (2026): Two Cities, One Map","description":"First grooming vertical and first Turkish-language run of the cross-vertical AI-search series: 23 prompts × 4 AI engines (Gemini not fired this run — disclosed, not imputed), EN + TR, matched to a 374-row Istanbul registry (367 real shops); all 368 planned captures landed. The English and Turkish top-5 lists on the control prompt share zero shops — the series floor for control-prompt overlap, matched by Mexico City tacos a week later. ChatGPT cites shop websites 21.3% (tracking the 22.1% own-site density of the registry), skips web search on 40% of answers, and cites Instagram zero times for the sixth straight city. Booking platforms take 18.9% of Perplexity's citations, with armut.com cited by all four engines. Rimedzo Barber Shop is a clean dual-metric #1.","datePublished":"2026-08-12T00:00:00.000Z","dateModified":"2026-08-12T00:00:00.000Z","url":"https://nicolassitter.com/research/barbershops-istanbul-ai-search-2026","category":"research"},{"@type":"BlogPosting","headline":"How many apps are there in Claude and ChatGPT? (2026)","description":"The app layer under AI assistants, counted from the platforms' own directory endpoints on August 3, 2026: 1,735 apps visible in ChatGPT's directory from France, 1,859 from the US — the catalog is country-shaped, with Expedia, Skyscanner and Uber among the US-only tier — and 1,375 Claude connectors, 584 added in July alone. The travel deep-dive probed every published endpoint read-only: registries and stores overlap by 5%, two-thirds of listed travel servers are unreachable, tool manifests carry publicly readable competitive positioning (Agentorist's tells the model to ALWAYS book through it), and Expedia's manifest reached v7 while its checkout actions remain unannounced.","datePublished":"2026-08-04T00:00:00.000Z","dateModified":"2026-08-04T00:00:00.000Z","url":"https://nicolassitter.com/research/mcp-apps-census-2026","category":"research"},{"@type":"BlogPosting","headline":"AI Cross-Platform Consensus 2026: Five Engines, Five Lists","description":"A cross-engine comparison of AI hotel recommendations in one pinned week (2026-08-03): the per-city top-10 lists of ChatGPT, Gemini, Perplexity, Copilot and Google AI Mode intersected across 56 destinations (1,490 distinct city-hotel recommendations). Findings: 57.6% of recommendations exist on exactly one engine, 18.9% on two, 11.2% on three, 7.7% on four and 4.6% (69 hotels) on all five. Most similar pair: Copilot × Google AI Mode (34.8 avg Jaccard / 51.2 containment); least similar: ChatGPT × Perplexity (16.4). Perplexity is the contrarian with 45.8% unique picks. Chain-branded hotels average 2.06 backing engines vs 1.65 for independents and reach all-five consensus at 6.6% vs 3.2%. City consensus spans Auckland (49.6 avg pairwise Jaccard) to Bali (3.8); 17 of 56 cities, including Paris, London, Rome, Tokyo, Bangkok and Bali, have no all-five hotel. Exactly one hotel is ranked #1 on all five engines: The Taj Mahal Palace, Mumbai (runner-up: Hotel Monteleone, New Orleans). Grok excluded — data stale since week 2026-05-18.","datePublished":"2026-08-04T00:00:00.000Z","dateModified":"2026-08-04T00:00:00.000Z","url":"https://nicolassitter.com/research/ai-cross-platform-consensus-2026","category":"research"},{"@type":"BlogPosting","headline":"AI Search for Specialty Coffee in Seoul (2026): The City Where the Social Layer Splits Three Ways","description":"Second coffee city in the cross-vertical series: 23 prompts × 4 AI engines, EN + KO, US + KR proxies, matched to a 494-café Seoul registry. ChatGPT cites café websites just 2.3% (Marseille's 10% was a vertical trait, not a city artifact). With almost no own-web layer, each engine reroutes differently: ChatGPT to Reddit and tourism guides, Copilot to Instagram (53%, partial batch), Perplexity to Naver blogs. Fritz Coffee Company is the consensus #1 with a citation score of zero.","datePublished":"2026-07-29T00:00:00.000Z","dateModified":"2026-07-29T00:00:00.000Z","url":"https://nicolassitter.com/research/specialty-coffee-seoul-ai-search-2026","category":"research"},{"@type":"BlogPosting","headline":"The Cost of AI Crawling (2026): What AI Bots Cost, Measured From First-Party Logs","description":"A first-party measurement of AI-bot crawling: 119,609 AI-bot requests across a real hotel and a marketing landing page (Dec 2025–Jul 2026). An 81% OpenAI/Claude duopoly on the hotel, ByteDance #2 on the landing page, on-demand fetches (ChatGPT-User/Claude-User) as a server-side impression counter that rose 7 → 389/mo, a crawl-to-referral picture too thin to state, and a search-crawler worked example where a Vercel→Supabase log drain grew one table to 5.6 GB before triage cut it to 952 MB.","datePublished":"2026-07-24T00:00:00.000Z","dateModified":"2026-07-24T00:00:00.000Z","url":"https://nicolassitter.com/research/cost-of-ai-crawling","category":"research"},{"@type":"BlogPosting","headline":"AI Citation Throughput in Hotel Search (2026, Live): Retrieved vs. Cited","description":"A live weekly measurement of citation throughput (selection rate) — cited sources ÷ retrieved sources per AI engine, computed from the AI Hotel Landscape corpus (616 prompts × 56 destinations weekly, 1.1M+ retrieved sources). ChatGPT ~26%, Google AI Mode 9–10%, Grok 0.3% historically; Reddit tops every major domain at 59% throughput within hotel intent.","datePublished":"2026-07-22T00:00:00.000Z","dateModified":"2026-07-22T00:00:00.000Z","url":"https://nicolassitter.com/research/ai-citation-throughput-2026","category":"research"},{"@type":"BlogPosting","headline":"AI Hotel Volatility 2026: The Top 10 Rewrites Itself Weekly","description":"An 11-week time series (2026-05-11 to 2026-07-20) of AI hotel recommendations: the same 616-prompt library (56 destinations × 11 templates) fired weekly at ChatGPT, Gemini, Perplexity, Copilot and Google AI Mode, hotels entity-resolved to canonical IDs and ranked per engine per week by mention count. Findings: only 29.2% of a weekly top-10 survives to the next Monday (avg across engines); top-100 lists keep 49.7%. Churn personalities: Gemini is noise around a stable core (60% of its week-0 top-100 alive at week 10, 18 hotels never left), ChatGPT is drift (51.8% weekly retention but the cohort erodes to 24%), Perplexity is near-lottery (19% cohort survival, zero anchors, 10.5% weekly newcomer rate). Pool membership is calm: top-10 hotels remain in the 500-deep list 94–99% of the time on four of five engines. 29 distinct hotels held a permanent top-100 seat; 28 of the 41 seats are chain-branded. Grok excluded (absent from the data since 2026-05-25, scraper suspended), disclosed not imputed.","datePublished":"2026-07-22T00:00:00.000Z","dateModified":"2026-07-22T00:00:00.000Z","url":"https://nicolassitter.com/research/ai-hotel-volatility-2026","category":"research"},{"@type":"BlogPosting","headline":"ChatGPT's result_source for Flights (2026): A Different Retrieval Stack","description":"A flights replication of the hidden result_source study: 2,007 ChatGPT captures from US and UK vantage points. The licensed labrador tier drops from 99.85% (hotels) to 74.6%, bright carries 23.1%, serp appears for the first time, only 8.6% of flight questions skip the web, half of cited documents bypass the tagged retrieval layer, and Reddit is cited on 83% of its fetches.","datePublished":"2026-07-21T00:00:00.000Z","dateModified":"2026-07-21T00:00:00.000Z","url":"https://nicolassitter.com/research/chatgpt-flights-retrieval-tiers-2026","category":"research"},{"@type":"BlogPosting","headline":"Where the AIs Sent L'Étape du Tour Riders to Sleep (2026)","description":"An event-anchored lodging study fired in the days before L'Étape du Tour 2026 (Le Bourg-d'Oisans → Alpe d'Huez). 5 AI engines (ChatGPT, Perplexity, Gemini, Copilot, Google AI Mode), EN + FR, 14 lodging prompts that name the event but never the towns — the engines had to know the route. 252 captures, 901 lodging mentions matched to a 74-place Oisans registry (55 real hotels/B&Bs). Consensus pick: Hôtel Oberland in Le Bourg-d'Oisans (29 mentions), then the Alpe d'Huez block (Royal Ours Blanc 24, Le Castillan 23, Grandes Rousses 22). Only 29% of named lodging verifies as real route-area properties; 8% are whole towns, 4% cycling tour operators. ChatGPT alone recommended Albertville/La Plagne hotels — the 2025 edition's geography, 100 km away — 26 times; the four live-search engines never did. Airbnb/camping/gîtes: ~3% of mentions despite being how much of the field actually stays.","datePublished":"2026-07-21T00:00:00.000Z","dateModified":"2026-07-21T00:00:00.000Z","url":"https://nicolassitter.com/research/letape-du-tour-hotels-ai-search-2026","category":"research"},{"@type":"BlogPosting","headline":"Argentina vs Spain in Paris: Who Wins the AI Restaurant War? (2026)","description":"A World-Cup-final head-to-head, fired with France knocked out. We asked 5 AI engines (ChatGPT, Perplexity, Gemini, Copilot, Google AI Mode), EN + FR, for the best Argentinian vs the best Spanish restaurant in Paris — 20 balanced prompts, 360 captures, 2,006 mentions matched to 81 real venues (42 Argentinian, 39 Spanish, seeded from Google Maps using only cuisine-specific categories). Argentina wins decisively: a front-three consensus (LOCO 129, Santa Carne 116, Les Grillades de Buenos Aires 105, each beating Spain's best pick Bodega Potxolo at 93) and tighter venue concentration on every engine. Two asides: ChatGPT's structured map widget crowns Bistro Caminito (#1 there, only #10 in prose), and \"Rosario\" — an Argentine city — gatecrashes the Spanish arm. AI Mode × FR proxy rejected by Bright Data, absent.","datePublished":"2026-07-17T00:00:00.000Z","dateModified":"2026-07-17T00:00:00.000Z","url":"https://nicolassitter.com/research/argentina-vs-spain-paris-ai-search-2026","category":"research"},{"@type":"BlogPosting","headline":"AI Search for Tattoo Studios in Berlin (2026): Same City, New Vertical — the Baseline Snaps Back","description":"The series' first same-city control: Berlin, already measured for yoga, re-run for tattoo studios with identical languages and proxies. 26 prompts × 5 AI engines, EN + DE, against a 551-studio registry. Perplexity's booking-platform share collapses from 31% (yoga) to 4%, ChatGPT returns to 35% studio-website citations, EN/DE top-5 overlap hits 67% — the highest in the seven cases of this series — and famous private artists are recommended while their mentions never resolve to a Google Maps place.","datePublished":"2026-07-15T00:00:00.000Z","dateModified":"2026-07-15T00:00:00.000Z","url":"https://nicolassitter.com/research/tattoo-studios-berlin-ai-search-2026","category":"research"},{"@type":"BlogPosting","headline":"AI Search for Bistros in Paris (2026): The Guidebook City Where Restaurant Websites Disappear","description":"A cross-vertical field test: 26 prompts × 5 AI engines (ChatGPT, Perplexity, Gemini, Copilot, Google AI Mode), EN + FR, matched to 579 Paris bistro venues. ChatGPT cites bistro websites just 0.9% — the lowest in the six cases of this series — while the Paris guide layer (parisjetaime.com, Time Out, the Michelin Guide) absorbs the citations and the predicted TheFork takeover never happens (~2% booking share). 94% arrondissement accuracy overall; Perplexity the outlier at 47%.","datePublished":"2026-07-08T00:00:00.000Z","dateModified":"2026-07-08T00:00:00.000Z","url":"https://nicolassitter.com/research/bistros-paris-ai-search-2026","category":"research"},{"@type":"BlogPosting","headline":"How ChatGPT Pulls Hotel Prices (2026): It Scrapes the OTAs, Then Cites Reddit","description":"A network-source forensic study of how ChatGPT sources hotel prices. Across 240 captures and 3,092 fetched documents: price questions hit the live web 98.8% of the time, OTAs are 46% of everything fetched but cited only 11%, Reddit is cited 100% of the time it is fetched, and the cited price rides one licensed retrieval tier (labrador) not the Bright Data scraper tier.","datePublished":"2026-06-26T00:00:00.000Z","dateModified":"2026-06-26T00:00:00.000Z","url":"https://nicolassitter.com/research/chatgpt-hotel-price-sources-2026","category":"research"},{"@type":"BlogPosting","headline":"How to Measure AI Hotel Traffic and Bookings (2026)","description":"A practical framework for measuring AI-driven hotel traffic and bookings when attribution is broken: GA4 referrers as a floor, branded-query growth, pre-booking forms, and ChatGPT Apps. Includes the attribution-gap benchmark — analytics likely undercounts AI influence by ~2-5x for hotels.","datePublished":"2026-06-25T00:00:00.000Z","dateModified":"2026-06-25T00:00:00.000Z","url":"https://nicolassitter.com/research/how-to-measure-ai-hotel-traffic-2026","category":"research"},{"@type":"BlogPosting","headline":"ChatGPT's Hidden result_source (2026): How It Sources Hotel Answers","description":"An undocumented field, result_source, tags every page ChatGPT retrieves with the pipeline that fetched it. Across 30,002 hotel citations, 99.85% come from one licensed tier (labrador), serp never appears, and 37.3% of questions never search the web. Within the tier, brand sites are cited while aggregator listicles are retrieved then discarded.","datePublished":"2026-06-25T00:00:00.000Z","dateModified":"2026-06-25T00:00:00.000Z","url":"https://nicolassitter.com/research/chatgpt-result-source-retrieval-tiers-2026","category":"research"},{"@type":"BlogPosting","headline":"AI Hotel Memory 2026: What Chatbots Remember About Hotels Without Searching","description":"A parametric-recall study with web search OFF. Three cheap models (GPT-5.4-nano, GPT-5.4-mini, Gemini 3.1 Flash-Lite) named hotels in JSON, returning each hotel's website so it could be verified by DNS. ~1,400 generations across global chains and Paris/Dubai/London/New York. Findings: hotel chains are known cold (~99% correct websites everywhere); individual-hotel websites resolve only 47%–97% of the time depending on model and city; the failure mode is the model knowing a real hotel but inventing its web address (Le Bristol Paris → dead bristolparis.com vs real oetkercollection.com); Paris is the hardest city (independent palaces on collection domains); and the cheapest model, Gemini 3.1 Flash-Lite, had the most accurate hotel memory.","datePublished":"2026-06-16T00:00:00.000Z","dateModified":"2026-06-16T00:00:00.000Z","url":"https://nicolassitter.com/research/ai-hotel-memory-2026","category":"research"},{"@type":"BlogPosting","headline":"AI Search for Specialty Coffee in Marseille (2026)","description":"Marseille breaks the entity-engine pattern. Across Paris yoga, Berlin yoga and Amsterdam bikes, ChatGPT cited shop/studio websites ~32% of the time. For Marseille specialty coffee it's only 10% — instead 31% social (Reddit leads that bucket) + 32% review-aggregators + 14% French local blogs = 77% third-party. 23 prompt templates × 5 AI engines × EN/FR (9/10 platform-proxy batches; AI Mode × FR rejected at the Bright Data trigger), 413 captures, 3,442 citations, 786 map entities against 280 specialty cafés. Three Marseille-only findings: Instagram at 237 cites (the highest social signal we've measured in any city/vertical), Gemini swinging to global specialty press (baristamagazine.com = 34% of its citations) when local trade press is absent, and a two-metric leaderboard split — Deep is the text-mention consensus winner (200 mentions across all five engines, 59.3% of ChatGPT) while Nua tops the cite-counted score (356) only because its brand_key is instagram.com and every Instagram cite attributes to it. EN vs FR control prompt = 11% overlap, the most language-divergent result so far.","datePublished":"2026-06-15T00:00:00.000Z","dateModified":"2026-06-15T00:00:00.000Z","url":"https://nicolassitter.com/research/specialty-coffee-marseille-ai-search-2026","category":"research"},{"@type":"BlogPosting","headline":"Are Hotels in Common Crawl? 39% Are Missing From AI Training Data (2026)","description":"108,109 hotel websites checked against the May 2026 Common Crawl snapshot: 60.6% are in it, 39.4% absent. Independents (61%) beat chains (45.9%); local-market TLDs are present but shallow; .es lags at 37%. Includes an interactive coverage map and a free checker.","datePublished":"2026-06-09T00:00:00.000Z","dateModified":"2026-06-09T00:00:00.000Z","url":"https://nicolassitter.com/research/hotels-in-common-crawl-2026","category":"research"},{"@type":"BlogPosting","headline":"AI Search for Yoga Studios in Berlin (2026)","description":"Direct replication of the Paris yoga study in Berlin: 27 prompts × 5 AI engines × EN/DE, 540 captures, 5,293 citations matched to 631 studios. The three engine personalities replicate almost to the percentage point (Copilot 95% entity, ChatGPT 32% studios + 19% Reddit, AI Mode 59% google.com) — strong evidence they're structural, not city-specific. The Berlin twist: booking platforms (Urban Sports Club, Eversports, ClassPass) climb to 31% of Perplexity citations, with blog.urbansportsclub.com and classpass.com the two most-cited non-Google domains overall.","datePublished":"2026-05-27T00:00:00.000Z","dateModified":"2026-05-27T00:00:00.000Z","url":"https://nicolassitter.com/research/yoga-studios-berlin-ai-search-2026","category":"research"},{"@type":"BlogPosting","headline":"AI Search for Bike Shops in Amsterdam (2026)","description":"27 prompts × 5 AI engines × EN/NL against 228 Amsterdam bike shops, 378 captures, 3,010 citations. The cleanest cross-engine entity consensus in the study: Copilot 97% shop websites, Reddit the single most-cited external domain anywhere (198 cites, 4 platforms), AI Mode 82% google.com self-citation. Perplexity's exposed search query (fanout_count=1, prompt language preserved) is the mechanism behind 0% EN/NL overlap on repair and commuter queries.","datePublished":"2026-05-26T00:00:00.000Z","dateModified":"2026-05-26T00:00:00.000Z","url":"https://nicolassitter.com/research/bike-shops-amsterdam-ai-search-2026","category":"research"},{"@type":"BlogPosting","headline":"AI Search for Bookstores in Tokyo (2026)","description":"22 prompts × 4 AI engines (Perplexity returned no usable Tokyo data), EN + JA, US/JP proxies, matched to 584 Tokyo bookstores. Tokyo is where AI search stops looking Western: Gemini cites store sites just 5% of the time and runs on a 58% local-guide web (whenin.tokyo, Tokyo Weekender, GaijinPot), Japanese prompts cite .jp domains 5× more than English (the sharpest language→TLD coupling in the study), and DAIKANYAMA T-SITE / Kinokuniya tie at the top (122 each).","datePublished":"2026-05-26T00:00:00.000Z","dateModified":"2026-05-26T00:00:00.000Z","url":"https://nicolassitter.com/research/bookstores-tokyo-ai-search-2026","category":"research"},{"@type":"BlogPosting","headline":"AI Search for Yoga Studios in Paris (2026)","description":"First non-hotel field test of the AI-search methodology. 27 prompts across 5 AI engines (ChatGPT, Perplexity, Gemini, Copilot, Google AI Mode), EN + FR, matched to 369 Paris yoga studios. Copilot cites studio sites 96% of the time, ChatGPT leans on Reddit (its #1 source at 17%), Google AI Mode cites its own SERP back 52%. The per-engine citation split from the hotel studies generalises intact.","datePublished":"2026-05-24T00:00:00.000Z","dateModified":"2026-05-24T00:00:00.000Z","url":"https://nicolassitter.com/research/yoga-studios-paris-ai-search-2026","category":"research"},{"@type":"BlogPosting","headline":"The ChatGPT Direct-Traffic Explosion for Hotels (May 2026)","description":"On May 7, 2026, ChatGPT started embedding hotel-brand URLs inline in answers. Across The Hotels Network's panel of 17,000+ hotels, daily AI referrer sessions jumped +62% (31,688 → 51,282/day) and held through May 25. A skewed tail: 286 hotels now draw 5%+ of new sessions from AI, 43 get 10%+. Perplexity and Claude lost share — it's a ChatGPT-only story.","datePublished":"2026-05-21T00:00:00.000Z","dateModified":"2026-05-29T00:00:00.000Z","url":"https://nicolassitter.com/research/chatgpt-hotel-direct-traffic-explosion-2026","category":"research"},{"@type":"BlogPosting","headline":"The Schema.org Debate (2026): Why It Still Matters for Hotels","description":"AEO oversold schema as an LLM unlock. The pushback that transformers read tokens, not JSON-LD, is right for the general case. For hotels, every major AI still grounds against Places / KG / OTA aggregators — surfaces that sit downstream of schema. The four fields that actually move the needle: Hotel, sameAs, starRating, alternateName.","datePublished":"2026-05-13T00:00:00.000Z","dateModified":"2026-05-13T00:00:00.000Z","url":"https://nicolassitter.com/research/schema-org-grounding-loop-2026","category":"research"},{"@type":"BlogPosting","headline":"How Mistral Searches Hotels","description":"Captured Le Chat event streams. One Brave web_search call per entity (parallelised for brand-vs-brand prompts), prompt-language preserved with per-term rewrites, the current year injected as a freshness anchor, snippet paraphrase. Niche queries surface real specialists; generic queries surface SEO-spam aggregators. Authority-laundering patterns turn single reviews and self-marketing into asserted features — and one hallucinated a hotel that doesn’t exist.","datePublished":"2026-05-05T00:00:00.000Z","dateModified":"2026-05-05T00:00:00.000Z","url":"https://nicolassitter.com/research/how-mistral-searches-hotels-2026","category":"research"},{"@type":"BlogPosting","headline":"How Claude Searches Hotels","description":"Captured event streams across several Claude hotel conversations. With Connector Discovery off (the default), almost everything goes through one Google Places call. Turn it on and Claude branches into a small curated OTA-connector picker (Booking.com / Tripadvisor / Trivago and a few others) — no ads by design, so the curation logic itself becomes the product.","datePublished":"2026-05-01T00:00:00.000Z","dateModified":"2026-05-04T00:00:00.000Z","url":"https://nicolassitter.com/research/how-claude-searches-hotels-2026","category":"research"},{"@type":"BlogPosting","headline":"ChatGPT Hotel Ads Are Live — CPC Pivot, Ads Manager, $50K Entry","description":"Sponsored ads in 20-35% of hotel queries. Booking.com at 43.5%. April 29 update: OpenAI launched a self-serve ads manager, moved from CPM to CPC, dropped entry to $50K, ~$100M annualised revenue six weeks in.","datePublished":"2026-04-29T00:00:00.000Z","url":"https://nicolassitter.com/research/chatgpt-hotel-ads-live-2026","category":"research"},{"@type":"BlogPosting","headline":"ChatGPT 5.3 Halved Its Hotel Sources — March 5, 2026 Cutover","description":"Daily ChatGPT UI runs of 140 world hotel prompts from 4 locales. On Mar 5, URLs per answer fell 49% (24→12). Booking −82%, Expedia −76%, Reddit −93%.","datePublished":"2026-04-17T00:00:00.000Z","url":"https://nicolassitter.com/research/chatgpt-hotel-source-shift-2026","category":"research"},{"@type":"BlogPosting","headline":"Hotel YouTube Channels — Activity Study 2026","description":"We analyzed YouTube channels linked from 9,889 hotel websites. 43.7% are ghost channels. Only 11.3% post monthly.","datePublished":"2026-04-16T00:00:00.000Z","url":"https://nicolassitter.com/research/youtube-hotel-visibility-2026","category":"research"},{"@type":"BlogPosting","headline":"ChatGPT Hotel Index vs Live Web — What Changes When Search Goes Offline","description":"400 hotel queries, 2 models, 2 search modes. 83% of cited domains differ between live and cached.","datePublished":"2026-04-09T00:00:00.000Z","url":"https://nicolassitter.com/research/chatgpt-hotel-index-vs-live-web-2026","category":"research"},{"@type":"BlogPosting","headline":"How Dirty Is Google Maps Hotel Data? 179K Study","description":"17% of Google Maps hotel listings fail QA. 8,167 OYO vacation rentals. Belgium loses 54% after cleaning.","datePublished":"2026-04-01T00:00:00.000Z","url":"https://nicolassitter.com/research/google-maps-hotel-data-quality-2026","category":"research"},{"@type":"BlogPosting","headline":"ChatGPT Hotel Data Sources: 100K Entity Study","description":"Google dropped from 100% to 70.3% in 90 days. TripAdvisor descriptions are 8.8x longer.","datePublished":"2026-03-24T00:00:00.000Z","url":"https://nicolassitter.com/research/tripadvisor-chatgpt-hotels-study-2026","category":"research"},{"@type":"BlogPosting","headline":"What Hotel Footers Reveal — 98K Study","description":"Instagram is in 40.8% of hotel footers. 24% of copyright years are 3+ years stale. 10% link to OTAs.","datePublished":"2026-03-23T00:00:00.000Z","url":"https://nicolassitter.com/research/hotel-footer-analysis-study-2026","category":"research"},{"@type":"BlogPosting","headline":"Hotel llms.txt Adoption Study 2026","description":"105,002 hotel websites scanned for llms.txt. Only 6.3% have one. US leads at 12.4%.","datePublished":"2026-03-21T00:00:00.000Z","url":"https://nicolassitter.com/research/hotel-llms-txt-adoption-study-2026","category":"research"},{"@type":"BlogPosting","headline":"Hotel robots.txt & AI Blocking Study 2026","description":"105,002 hotel robots.txt files parsed. Only 3.3% block any AI crawler. France leads at 7.5%.","datePublished":"2026-03-20T00:00:00.000Z","url":"https://nicolassitter.com/research/hotel-robots-ai-blocking-study-2026","category":"research"},{"@type":"BlogPosting","headline":"What Hotels Are Actually Called: A Naming Study","description":"Analysis of naming conventions across 121,425 hotels in 7 countries","datePublished":"2026-03-10T00:00:00.000Z","url":"https://nicolassitter.com/research/hotel-naming-study-2026","category":"research"},{"@type":"BlogPosting","headline":"Hotel Schema.org Adoption Study 2026","description":"121,425 hotels scanned — 36.3% have no schema at all","datePublished":"2026-03-05T00:00:00.000Z","url":"https://nicolassitter.com/research/hotel-schema-adoption-study-2026","category":"research"},{"@type":"BlogPosting","headline":"Anatomy of a ChatGPT Hotel Search","description":"12 systems, 7 providers, 424 A/B tests — a technical teardown","datePublished":"2026-03-01T00:00:00.000Z","url":"https://nicolassitter.com/research/anatomy-chatgpt-hotel-search-2026","category":"research"},{"@type":"BlogPosting","headline":"Yelp in ChatGPT: Hotel Data Study","description":"33% Yelp integration rate in US hotel queries across 14 destinations","datePublished":"2026-02-20T00:00:00.000Z","url":"https://nicolassitter.com/research/yelp-chatgpt-hotels-study-2026","category":"research"},{"@type":"BlogPosting","headline":"How Consistent Are AI Hotel Rankings?","description":"Only 50.5% position stability across query reruns","datePublished":"2026-02-15T00:00:00.000Z","url":"https://nicolassitter.com/research/ai-hotel-rankings-consistency-study-2026","category":"research"},{"@type":"BlogPosting","headline":"Google AI Mode: Where Do Hotel Clicks Actually Go?","description":"79% of hotel clicks in Google AI Mode go to Google Business Profiles","datePublished":"2026-02-10T00:00:00.000Z","url":"https://nicolassitter.com/research/google-ai-mode-hotel-study-2026","category":"research"},{"@type":"BlogPosting","headline":"Do French Hotels Blog? A 15,000-Hotel Study","description":"49.3% have blogs but only 1 in 4 are active","datePublished":"2026-01-20T00:00:00.000Z","url":"https://nicolassitter.com/research/french-hotel-blog-study-2026","category":"research"},{"@type":"BlogPosting","headline":"The AI Hotel Landscape 2026","description":"How 6 AI Models Rank 12,500+ Hotels Across 1.2 Million Citations","datePublished":"2026-01-15T00:00:00.000Z","url":"https://nicolassitter.com/research/ai-hotel-landscape-2026","category":"research"}]}