{"@context":"https://schema.org","@type":"BlogPosting","headline":"Hotel recall without search: what AI remembers, and how much survives a web search","description":"616 hotel prompts sent to Muse Spark, GPT-6 Luna, Gemini 3.5 Flash-Lite and Claude Haiku 5.5 with web search off, compared hotel by hotel with searched answers from the same families. Remembered hotels are almost all real (0 to 2.2% estimated invented). Memory predicts search: a hotel a model names in half or more of a city's prompts is named after search 96 to 97% of the time (Muse, ChatGPT), against 41 to 47% for hotels remembered once or twice, and the first searched hotel is in the no-search answer in 67% of prompts. Luxury brands keep their place after search far more than independents (69% vs 41% for ChatGPT); the hotels search adds are 64 to 76% independent.","datePublished":"2026-10-28","dateModified":"2026-10-28","url":"https://nicolassitter.com/research/ai-hotel-memory-vs-search-2026","category":"research","keywords":["AI hotel memory","parametric knowledge hotels","LLM memory vs search","parametric knowledge optimization"],"articleSection":"Research","wordCount":2400,"readTime":"10 min","articleBody":"AI search study · October 2026616 prompts · 4 models · run 27 October 2026\n\n# Hotel recall without search: what AI remembers, and how much survives a web search\n\nI sent the 616 hotel prompts of the AI Hotel Landscape to four current models with web search switched off, checked whether the hotels they named exist, and compared every list with what the same model families recommend after searching.\n\nThe remembered hotels are almost all real. More to the point, memory predicts search: a hotel a model names for many prompts in a city is very likely to be named after searching too, and usually near the top. Luxury brands keep their place after search far more often than independents, while the hotels search adds are mostly independents.\n\nNS\n\nNicolas Sitter\n\nPublished 28 October 2026\n\n67%\n\nof ChatGPT's first-named hotels are in GPT-6 Luna's no-search answer\n\n97% vs 41%\n\nchance search names a strongly vs a weakly remembered hotel (Muse)\n\n58%\n\nof Muse's search queries name a specific hotel\n\n69% vs 41%\n\nof remembered luxury-brand vs independent hotels kept by ChatGPT\n\n[Read the study](#method)\n\n## What I measured\n\nAn AI model can answer a hotel question from what it learned in training, or by searching the web first. The AI Hotel Landscape measures the searched answers. This study measures the remembered ones, on the same prompts, so the two can be compared hotel by hotel.\n\nIn June, [AI Hotel Memory 2026](/research/ai-hotel-memory-2026) asked three small models to list hotels with no other context and checked their websites. Chains were known well, independent hotels less so, and models invented websites more often than hotels. This study asks a different question: when a model does search, how much of what it recommends was already in its memory?\n\nThe question comes from [a study by Graphite](https://graphite.io/five-percent/research/parametric-knowledge-optimization) (October 2026) on what it calls parametric knowledge optimization. On entity-comparison prompts in ChatGPT, it found that entities mentioned more often without search were also more visible with search, even among entities that appear equally often on the pages ChatGPT cites. Hotels are a good test: a destination has hundreds of them, and each is a named entity that can be matched across answers.\n\nI used the cheapest current model of each family:\n\n-   **Muse Spark 1.3** (Meta), also run with web search on the same day\n-   **GPT-6 Luna** (OpenAI), compared with ChatGPT’s searched answers\n-   **Gemini 3.5 Flash-Lite** (Google), compared with the Gemini app and Google AI Mode\n-   **Claude Haiku 5.5** (Anthropic), which has no searched counterpart in the landscape\n\nEach received the 616 prompts word for word, once, with no tools. The prompts are generic (“best hotels in Lisbon”, “family friendly hotels in Dublin”). None names a brand, and only a few landmark prompts name a hotel (“hotels near Bellagio”). Hotel names were extracted from every answer with the step the landscape uses, and grouped per destination so that “Hôtel de Crillon” and “Hôtel de Crillon, A Rosewood Hotel” count as one hotel.\n\nOnly Muse is the same model with search off and on. ChatGPT, the Gemini app and AI Mode run larger models than the ones tested here, so those pairs compare a cheap model’s memory with its family’s searched product. The Muse search run has its own article: [How Muse Spark searches for hotels](/research/muse-spark-hotel-search-2026).\n\n## Are the hotels real?\n\nEach name was matched against a directory of real hotels for its destination. Between 68% and 86% of each model’s hotels are in it, and a further 4% to 6% were named by a search engine for the same city. I checked 25 of the remaining names per model by hand, after a Google Places lookup: of the 100, 81 are real hotels, 4 are invented, 14 are uncertain and 1 is not a hotel.\n\nModel\n\nHotels per answer\n\nIn the directory\n\nUnverified\n\nEstimated invented\n\nUpper bound\n\nMuse Spark 1.3\n\n11\n\n77%\n\n18%\n\n0.7%\n\n5.3%\n\nGPT-6 Luna\n\n6\n\n86%\n\n8%\n\n0.0%\n\n1.0%\n\nGemini 3.5 Flash-Lite\n\n7\n\n78%\n\n17%\n\n0.7%\n\n7.9%\n\nClaude Haiku 5.5\n\n8\n\n68%\n\n28%\n\n2.2%\n\n14.4%\n\nThe directory is incomplete (196 hotels for Queenstown, 12,620 for Beijing), so “unverified” means not found in these references. The upper bound counts every uncertain case as invented and adds the sampling error (95% interval). Claude Haiku named no hotel at all in 15 answers, giving neighbourhood advice instead.\n\nThe invented cases are mostly real names in the wrong city: “Circa 039 Hotel” in Buenos Aires resembles Circa 39 in Miami, and “Hotel NoMo Athens” borrows from NoMo SoHo in New York. When the prompt names a destination, these models rarely make hotels up.\n\n## Memory predicts what search finds\n\nFor each destination I counted how often a hotel appears in the remembered answers (its memory visibility) and how often it appears in the searched answers to the same prompts. A destination has about 11 prompts, so a hotel remembered in half of them is one the model reaches for whatever the question.\n\nRemembered hotels that the searched answers name at least once, by memory visibility\n\nMuse, search off vs on\n\nUnder a quarter of prompts\n\n41%\n\nA quarter to a half\n\n88%\n\nHalf or more\n\n97%\n\n0%100%\n\nGPT-6 Luna vs ChatGPT\n\nUnder a quarter of prompts\n\n47%\n\nA quarter to a half\n\n89%\n\nHalf or more\n\n96%\n\n0%100%\n\nGemini Flash-Lite vs Gemini app\n\nUnder a quarter of prompts\n\n24%\n\nA quarter to a half\n\n64%\n\nHalf or more\n\n87%\n\n0%100%\n\nGemini Flash-Lite vs AI Mode\n\nUnder a quarter of prompts\n\n24%\n\nA quarter to a half\n\n54%\n\nHalf or more\n\n77%\n\n0%100%\n\nUnit: one hotel in one destination. Memory visibility: share of that destination’s prompts whose no-search answer names it.\n\nA hotel Muse names in half or more of a destination’s memory answers is named by its searched answers 97% of the time. A hotel it remembers for only one or two prompts: 41%. GPT-6 Luna and ChatGPT give almost the same figures (96% and 47%), even though ChatGPT runs a different model. Google’s products follow the same order with weaker links.\n\nThe same holds without picking a model, counting how many of the four models remember a hotel in a destination:\n\nChatGPT and AI Mode: hotels named at least once by search, by how many models remember them\n\nChatGPT search\n\n1 of 4 models\n\n9%\n\n2 of 4 models\n\n36%\n\n3 of 4 models\n\n59%\n\n4 of 4 models\n\n84%\n\n0%100%\n\nAI Mode search\n\n1 of 4 models\n\n7%\n\n2 of 4 models\n\n23%\n\n3 of 4 models\n\n41%\n\n4 of 4 models\n\n64%\n\n0%100%\n\nSeen from the search side, 64% of ChatGPT’s hotel mentions go to hotels GPT-6 Luna names somewhere in that destination from memory. The figure is 59% for Muse, 63% for the Gemini app and 46% for AI Mode. Across all four memories, 77% of ChatGPT’s mentions and 67% of AI Mode’s go to a hotel at least one cheap model remembers.\n\n## The first hotel is usually a memory\n\nOverlap for a whole list is moderate: 38% of Muse’s remembered hotels are in its searched answer to the same prompt, and 45% of GPT-6 Luna’s are in ChatGPT’s. The top of the list is different.\n\nRank and memory, same prompt\n\nMuse: #1 also in memory\n\n67%\n\nMuse: memory top 3 kept\n\n60%\n\nMuse: memory 4th+ kept\n\n30%\n\nGPT-6 Luna / ChatGPT: #1 also in memory\n\n67%\n\nGPT-6 Luna / ChatGPT: memory top 3 kept\n\n59%\n\nGPT-6 Luna / ChatGPT: memory 4th+ kept\n\n33%\n\n0%100%\n\nIn two prompts out of three, the hotel a searched answer names first was already in the remembered answer. Hotels a model ranks in its top three from memory are about twice as likely to stay after search as those further down. Search mostly changes the bottom of the list.\n\n## Search queries carry the memory\n\nMuse exposes the queries it sends to its search engine, and ChatGPT’s are captured by the landscape. I had each of the 3,681 queries read by a model to find the hotels it names.\n\nSearch queries that name a specific hotel or a brand\n\nMuse: name a hotel\n\n58%\n\nChatGPT: name a hotel\n\n8%\n\nMuse: name a brand\n\n40%\n\nChatGPT: name a brand\n\n5%\n\n0%100%\n\nChatGPT usually sends one broad query that restates the prompt. Muse sends a median of four, and 58% of them name a hotel. Since the prompts almost never name one, those names come from the model. Of the 2,324 hotel names in Muse’s queries, 53% were also in its no-search answer to the same prompt, and 75% ended up in the searched answer. Muse appears to search to check and describe hotels it already has in mind, more than to discover new ones.\n\nChatGPT’s queries name a hotel less often, but when they do, 61% of those hotels are in GPT-6 Luna’s memory answer. One query reads “best boutique hotels Back Bay Boston The Newbury Lenox Eliot Hotel official hotel sites”: all three hotels were in GPT-6 Luna’s memory answer to that prompt. No query in either engine used a site: operator.\n\n## Brand or not\n\nI classified every hotel with the landscape’s chain classifier into three groups:\n\n-   **Big group:** Marriott, Hilton, Accor, IHG, Hyatt and similar, including luxury sub-brands such as Ritz-Carlton or St. Regis.\n-   **Luxury brand:** Four Seasons, Mandarin Oriental, Aman, Rosewood, Peninsula, Belmond, Shangri-La and similar.\n-   **Independent:** everything else, including members of networks such as Leading Hotels of the World or Small Luxury Hotels.\n\nBrand mix of hotel mentions\n\nBig groupLuxury brandIndependent\n\nMuse memory\n\n65%\n\nGPT-6 Luna memory\n\n63%\n\nGemini Flash-Lite memory\n\n69%\n\nHaiku 5.5 memory\n\n64%\n\nMuse search\n\n66%\n\nChatGPT search\n\n64%\n\nGemini app search\n\n61%\n\nAI Mode search\n\n71%\n\n0%100%\n\nValue shown: independent share.\n\nThe mix is close to identical across memory and search. Memory is not, as a whole, more branded than search. What differs is which hotels survive. A remembered luxury-brand hotel is much more likely to stay in the searched answer than a remembered independent:\n\nMemory vs search\n\nBig group kept\n\nLuxury brand kept\n\nIndependent kept\n\nAdded by search: independent\n\nMuse, search off vs on\n\n39%\n\n68%\n\n34%\n\n71%\n\nGPT-6 Luna vs ChatGPT\n\n44%\n\n69%\n\n41%\n\n70%\n\nGemini Flash-Lite vs Gemini app\n\n26%\n\n41%\n\n19%\n\n64%\n\nGemini Flash-Lite vs AI Mode\n\n19%\n\n32%\n\n15%\n\n76%\n\nThe strongest memories are also where luxury brands concentrate. Among hotels a model names in half or more of a destination’s prompts, 24% are luxury brands, against 4% among hotels named once or twice.\n\nBrand mix by memory visibility, four models pooled\n\nBig groupLuxury brandIndependent\n\nUnder a quarter of prompts\n\n4%\n\nA quarter to a half\n\n14%\n\nHalf or more\n\n24%\n\n0%100%\n\nValue shown: luxury-brand share.\n\nIndependents are still the majority at every level, because many famous hotels are independent: Le Bristol, Eichardt’s, La Mamounia, Hotel Monteleone. But the hotels search adds that were not remembered are 64% to 76% independent. For a smaller independent hotel, search is the main way into an answer. For a Four Seasons, memory already does much of the work.\n\n## One prompt, eight answers\n\n“Best hotels in Paris”, with the four memories on the left and the four searched answers on the right. The table lists the hotels named by at least two answers.\n\nMemory (search off)\n\nSearch on\n\nHotel\n\nMuse\n\nLuna\n\nGemini\n\nHaiku\n\nMuse\n\nChatGPT\n\nGemini app\n\nAI Mode\n\nHôtel de Crillon, A Rosewood Hotel\n\n●\n\n●\n\n●\n\n●\n\n●\n\n●\n\nLe Bristol Paris\n\n●\n\n●\n\n●\n\n●\n\n●\n\n●\n\nLe Meurice\n\n●\n\n●\n\n●\n\n●\n\n●\n\nLe Pavillon de la Reine\n\n●\n\n●\n\n●\n\n●\n\n●\n\nCheval Blanc Paris\n\n●\n\n●\n\n●\n\n●\n\nFour Seasons Hotel George V Paris\n\n●\n\n●\n\n●\n\n●\n\nHôtel Lutetia\n\n●\n\n●\n\n●\n\n●\n\nRitz Paris\n\n●\n\n●\n\n●\n\n●\n\nHôtel Chopin\n\n●\n\n●\n\n●\n\nRelais Christine\n\n●\n\n●\n\n●\n\nShangri-La Paris\n\n●\n\n●\n\n●\n\nHôtel Madame Rêve\n\n●\n\n●\n\nHôtel Particulier Montmartre\n\n●\n\n●\n\nLe Grand Mazarin\n\n●\n\n●\n\nPlaza Athénée\n\n●\n\n●\n\nNamed by this answer only\n\n4\n\n4\n\n3\n\n6\n\n10\n\n1\n\n0\n\n2\n\nHotels in the answer\n\n15\n\n10\n\n9\n\n14\n\n21\n\n6\n\n5\n\n5\n\nLe Bristol and the Crillon are the only hotels all four memories name, and both are in Muse’s and the Gemini app’s searched answers. ChatGPT’s six hotels include neither. Le Meurice, which only Haiku remembers, is named by all four search engines.\n\n## From June to October\n\nThe June study listed the ten hotels GPT-5.4 mini remembered most often in Paris, Dubai, London and New York. Four months and one model generation later, most of them still appear across each city’s 11 prompts:\n\nJune’s top 10 in\n\nGPT-6 Luna memory\n\nChatGPT search\n\nMuse memory\n\nMuse search\n\nParis\n\n8 of 10\n\n6 of 10\n\n9 of 10\n\n9 of 10\n\nDubai\n\n6 of 10\n\n5 of 10\n\n8 of 10\n\n8 of 10\n\nLondon\n\n8 of 10\n\n9 of 10\n\n10 of 10\n\n10 of 10\n\nNew York\n\n7 of 10\n\n6 of 10\n\n10 of 10\n\n10 of 10\n\nMuse names nearly all of them, with and without search. GPT-6 Luna names six to eight. In Paris, Mandarin Oriental and Hotel du Louvre, both remembered in more than 70% of June’s runs, are named by none of the October answers from GPT-6 Luna or ChatGPT. The two studies used different prompts, so this checks persistence and does not measure change.\n\n## How the remembered answers read\n\nAnswers\n\nWords (median)\n\nQuote a price\n\nName a year\n\nAsk you to check\n\nMuse memory\n\n685\n\n70%\n\n12%\n\n0%\n\nGPT-6 Luna memory\n\n196\n\n9%\n\n0%\n\n8%\n\nGemini Flash-Lite memory\n\n588\n\n13%\n\n1%\n\n4%\n\nHaiku 5.5 memory\n\n259\n\n10%\n\n0%\n\n39%\n\nMuse with search\n\n834\n\n38%\n\n32%\n\n2%\n\nMuse quotes a price in 70% of its answers without searching, more often than when it searches (38%). The other three models rarely quote prices from memory. Muse also mentions 2026 42 times across its no-search answers, often in openings such as “Here are my top picks for 2026”, for lists that come from training data. Claude Haiku tells the reader to check or verify in 39% of its answers.\n\n## What this means for hotels\n\n-   **Being remembered is a head start, not a guarantee.** A hotel a model names across many prompts for its city is very likely to appear after search too, and near the top. Memory is built from what was written before the model’s training cutoff, which for current models is several months before release.\n-   **Search is where most hotels get in.** Most hotels named by search were remembered weakly or not at all, and the ones search adds are mostly independents. For them, the pages search engines retrieve matter more than the model’s memory.\n-   **Engines differ.** Muse searches hotel by hotel, starting from names it already knows. ChatGPT sends one broad query and builds its list from what it finds.\n\n## Method and limits\n\n-   One answer per prompt and model, at default temperature. Visibility is measured across a destination’s prompts, not across repeated runs of one prompt.\n-   The memory models are the cheapest of each family. Except for Muse, the searched answers come from larger models, so “memory vs search” is a family comparison. Larger models are likely to remember more.\n-   The landscape captures are from the week of 19 October, eight days before the memory runs. Gemini and AI Mode lists combine several captures per prompt.\n-   Names are matched on their distinctive words, after removing words such as “hotel” and the city name. A brand word alone never makes a match. In a hand check of 40 matches, 4 pointed to a sister property, so overlap figures are slightly generous.\n-   The brand classifier works from the hotel name. A property listed under its own name with no brand in it counts as independent.\n-   Search queries were tagged by Claude Haiku 5.5. A spot check of 25 tagged queries found no errors.\n-   The invented-hotel estimate rests on 25 hand-checked names per model, judged by one person.\n-   Cost: $2.01 for the four memory runs, about $1.47 for hotel extraction, $4.10 for Google Places and under $1 to tag the queries.\n\nDisclosure: I work at Lighthouse, which owns The Hotels Network.\n\n### Summarize with AI\n\n### Related\n\n[How Muse Spark searches for hotels](/research/muse-spark-hotel-search-2026)[AI Hotel Memory 2026](/research/ai-hotel-memory-2026)[AI Chain Bias Distribution 2026](/research/ai-chain-bias-distribution-2026)[All research](/research)","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"]},"publisher":{"@type":"Person","name":"Nicolas Sitter","url":"https://nicolassitter.com"},"image":"https://nicolassitter.com/api/og/ai-hotel-memory-vs-search-2026","mainEntityOfPage":{"@type":"WebPage","@id":"https://nicolassitter.com/research/ai-hotel-memory-vs-search-2026"},"tags":["AI Search","Hotels","Parametric Memory","Brands"],"sameAs":["https://hotelrank.ai/research/ai-hotel-memory-vs-search-2026"],"alternateFormat":{"html":"https://nicolassitter.com/research/ai-hotel-memory-vs-search-2026","json":"https://nicolassitter.com/api/post/ai-hotel-memory-vs-search-2026","rss":"https://nicolassitter.com/rss.xml"},"datasets":[{"name":"summary","contentUrl":"https://nicolassitter.com/data/ai-hotel-memory-vs-search-2026/summary.csv","encodingFormat":"text/csv"}]}