AI dictionary for tourism: 18 terms you'll keep hearing
In two years, the travel industry has absorbed an entire vocabulary that used to belong to engineers. These are the 18 terms you will meet most often in 2026, explained without hype and with the reason they matter. I group them in three blocks: fundamentals, visibility and transaction.
Fundamentals
LLM (Large Language Model). The engine behind ChatGPT, Gemini or Claude: a neural network trained on enormous amounts of text to predict language. The key thing to grasp is that it is a probability engine, not a database of truths — hence the next term.
Hallucination. When a model confidently states something false: a hotel that doesn't exist, an invented opening time. It is not an occasional bug but a property of the system, which can be mitigated but never fully removed.
RAG (Retrieval-Augmented Generation). A technique forcing the model to consult a controlled document base before answering — an open-book exam. It underpins almost every serious travel chatbot: the bot answers from your documents, not from its memory.
Grounding. The general principle behind RAG: anchoring every answer in a source verified in real time instead of inferring it. When a vendor tells you their assistant "doesn't hallucinate", ask how they do the grounding.
Context window. The working memory of an AI conversation: how much information it can hold at once. It explains why some assistants "forget" what you told them ten messages ago.
Visibility: getting the AI to find you
AEO (Answer Engine Optimization). Optimising to appear in the responses of answer engines (ChatGPT, Perplexity, AI Overviews), not just in link lists. Expedia declared in 2026 that it is its fastest-growing channel.
GEO (Generative Engine Optimization). Practically a synonym of AEO, with the nuance of optimising so generative engines understand and recommend the brand. A caution the industry itself shares: techniques promising guaranteed effects still lack solid evidence.
AXO (Agent Experience Optimization). A term SiteMinder picked up in 2025 from consultant Simone Puorto (Travel Singularity): optimising the website for autonomous agents — no captchas in the information path, data-rich text, complete metadata. If your visitor is no longer always human, neither is "user experience".
Structured data (JSON-LD, Schema.org). The machine-readable spec sheet accompanying a page: what you are, where you are, what you offer, what it costs. It is the language in which a search engine or an assistant reads you without interpreting your design.
llms.txt. A text file at a website's root summarising its content for AI models, the way robots.txt summarises permissions. Google says it doesn't need it; part of AEO practice uses it anyway. It takes an hour to implement — my website has one, and I'd rather be covered in both scenarios.
E-E-A-T. Experience, expertise, authoritativeness and trust: the credibility signals search engines and AIs try to detect. The first E (first-hand experience) is the hardest to fake — which is why it is the most valuable.
Algorithmic legibility. Making your brand and data understandable and processable by machines. The phrase that sums it up, from Accor's orbit (2026): AI agents don't browse — they filter.
Query fan-out. When the AI decomposes a complex question ("hotel in Granada to visit the Alhambra with my mother, who can't walk far") into several parallel sub-searches. The website covering the subtopics is the one that ends up cited.
Transaction: getting the AI to book you
AI agent. A system that sets goals, plans and executes multi-step tasks with little supervision. The difference from a chatbot: the chatbot tells you how to book; the agent books.
Agentic AI. The general paradigm behind those agents: goal-oriented reasoning with the ability to act (search, compare, pay).
Zero-click booking. A booking that happens entirely inside the assistant, without touching the business's website. The first flight booking completed this way end to end arrived in May 2026 (Mindtrip with Sabre and PayPal).
MCP (Model Context Protocol). The open protocol connecting AI models with external systems — booking engines, PMSs, databases. It is called "the USB-C of AI"; by 2026 it had been adopted from Sabre to the main hotel vendors. Complementary commerce and payment protocols exist: UCP (Google) and ACP (OpenAI and Stripe).
Reintermediation. The risk that the disintermediation AI promises ends up as its opposite: new middlemen (connector aggregators, assistant marketplaces) charging a new toll on the conversation, the way OTAs charged for search.
Superagent (or "centaur" professional). The human travel agent revalued by AI: half hospitality, half tool manager. The data point sustaining the concept: travel agent was Spain's fastest-growing profession according to LinkedIn (2025) — though it no longer appears on the 2026 list.
To wrap up
This dictionary has an expiry date: half these terms didn't exist three years ago, and some will have shifted meaning within two. I will update it as the industry's vocabulary moves. Meanwhile, one practical rule: when someone uses three of these acronyms in a single sentence to sell you something, ask for the sourced, dated figure behind them. New words don't exempt anyone from the old questions.
