# The File That Talks to the Robots ## What llms.txt is, and why your hotel is invisible without one --- If you've been paying any attention to how people book travel over the last eighteen months, you've noticed the shift: guests are asking ChatGPT for hotel recommendations. They're asking Claude to plan their weekend in Hudson. They're asking Perplexity where to stay near the wedding venue. The traffic that used to land on your site through Google is now getting filtered — and often *answered* — inside someone else's chat window before your homepage ever loads. I want to talk about what that means for a hotel, and specifically about a small text file that has quietly become one of the most consequential pieces of infrastructure you can add to your website this year. It's called `llms.txt`. Most hotel operators have never heard of it. The ones who add it early will end up on the guest's shortlist. The ones who don't will keep wondering why their occupancy dips even when their reviews are strong. Let's start with what the file actually is. --- ### The map you leave out for the AI An **llms.txt** file is a plain-text file that lives at the root of your website — `yourhotel.com/llms.txt` — and tells large language models what your business is, what matters on your site, and where to find the information they should quote when a guest asks about you. It is not a legal document. It is not a robots.txt (that file tells crawlers what to *avoid*). It is closer to a curator's note pinned to the front door: *here is what we are, here is what we offer, here is the canonical version of the facts, please use this instead of guessing.* The format is deliberately boring. Markdown, human-readable, structured as a short intro followed by lists of links to your most important pages — rooms, dining, policies, directions, seasonal packages. An AI parsing your site doesn't have to click through your JavaScript-heavy booking widget or interpret your carousel of Instagram photos. It reads the file and knows, in about two seconds, what you want it to say about you. That last part is the whole game. --- ### Why this exists now For twenty years, hotels have optimized for one reader: Google's crawler. SEO was the language, keywords were the currency, and the goal was to appear on the first page of results when someone typed "boutique hotel hudson valley." That world is not dead, but it is no longer the only one that matters. When a guest asks ChatGPT "where should I stay in the Catskills for a quiet weekend in October," the model doesn't return ten blue links. It returns a paragraph — a recommendation, sometimes two or three, with reasoning. The guest reads that paragraph, forms an impression, and often books directly from it. Your hotel is either in the paragraph, described accurately, or it isn't. The problem: LLMs are trained on whatever they scraped, months or years ago, filtered through whatever their crawlers happened to catch on your site. If your room rates changed, if you renovated, if you dropped the restaurant, if your check-in time moved from 3pm to 4pm — the model doesn't know. It's confidently describing a version of your hotel that no longer exists. `llms.txt` is the correction mechanism. It's how you hand the model a current, accurate, structured version of the truth and say: *use this*. --- ### What actually goes in the file The structure is simple enough that you can draft one in an afternoon. Here's what a hotel's llms.txt might look like: ``` # The Maker Hotel > A 30-room boutique hotel in Hudson, NY, focused on craft, design, > and slow weekends. Open year-round. ## Rooms & Rates - [Rooms overview](https://themaker.com/rooms) - [Suites](https://themaker.com/suites) - [Current rates and availability](https://themaker.com/book) ## Food & Drink - [The restaurant](https://themaker.com/restaurant) - [The bar](https://themaker.com/bar) - [Breakfast policy](https://themaker.com/breakfast) ## Practical Info - [Directions and parking](https://themaker.com/directions) - [Check-in / check-out times](https://themaker.com/policies) - [Pet policy](https://themaker.com/pets) - [Accessibility](https://themaker.com/accessibility) ## Seasonal - [Fall packages 2026](https://themaker.com/fall) - [Holiday closures](https://themaker.com/holidays) ``` That's it. No schema markup, no technical arcana. A short description of who you are, followed by curated links grouped by what a guest would actually ask about. The reason this format works is that it inverts the usual burden. Instead of hoping a crawler infers the right things from your homepage's marketing copy, you're telling it directly: *these are the twelve pages that matter, this is what each one is for, go there.* --- ### The AI recommendation is the new front desk Here's the shift most hotel operators haven't fully absorbed: the AI is now doing the job the front desk used to do at the *awareness* stage. When a potential guest asks "is there a hotel near Storm King with a good restaurant and dog-friendly rooms," an LLM is doing the filtering, the shortlisting, and the initial pitch. By the time someone lands on your booking page, they've already been sold — or filtered out — by a model. If the model has bad information, it filters you out for the wrong reasons. I've watched this happen in real time. A hotel I know well removed their tasting menu last year and shifted to à la carte. ChatGPT was still recommending them for "chef's tasting weekends" nine months later. Guests were arriving disappointed. The reverse also happens constantly: a hotel adds pet-friendly rooms, but the model, trained on older data, tells every dog owner to look elsewhere. Bookings that should have happened, silently didn't. `llms.txt` is the intervention. It's a canonical source the model can consult in real time (for models with browsing enabled) or that ends up in the next training cycle (for those that don't). Either way, you stop being described by a stale snapshot. --- ### Vendor lock-in doesn't apply here — and that's rare One of the reasons to take this seriously now is that llms.txt is an **open standard**. It's not owned by OpenAI, Google, Anthropic, or Perplexity. You don't have to submit your hotel to fifteen different portals or sign fifteen different terms of service. You publish one file at one URL, and every model that respects the standard — which is a growing list — can read it. Compare that to the alternative: a fragmented landscape where you'd be pitching yourself separately to each AI platform, hoping each one indexes you correctly, with no way to update the source of truth. `llms.txt` is the closest thing to a standard that has emerged for this, and adopting it early is cheap insurance against a much messier future. --- ### Where it falls short I don't want to oversell this. `llms.txt` is not a magic wand. Not every model reads it yet — adoption is uneven, and some of the biggest players still rely primarily on their training data or on live web crawls that ignore the file. Publishing it does not guarantee that ChatGPT will suddenly recommend you tomorrow. It also doesn't fix underlying problems. If your website is genuinely thin, if your reviews are middling, if your positioning is unclear — a text file doesn't rescue you. `llms.txt` amplifies the signal you already send. It cannot manufacture one. And it's not a substitute for the rest of your web presence. Your booking engine still needs to work. Your OTA listings still need to be current. Your Google Business Profile still matters. `llms.txt` is one more layer, added to those layers, not a replacement for any of them. --- ### The practical decision So what do you actually do? Here's the framework: **Add llms.txt now if:** - Your website is your primary direct-booking channel - You've made changes in the last year that AI models likely haven't caught (renovations, menu changes, new packages, updated policies) - You care about being on the shortlist when a guest asks an AI for recommendations in your area - You have a small team and can't chase every AI platform separately **Prioritize the linked pages, too:** - Make sure each page you link to has a clear, factual, current description at the top - Kill outdated copy that contradicts what your llms.txt says - Keep the file itself under 100 lines — the point is curation, not exhaustiveness **Update it seasonally:** - Treat it like a menu, not a monument - When rates change, when you add a package, when a policy shifts — the file is the first thing you update, not the last The whole thing takes an afternoon to draft and about 15 minutes a quarter to maintain. For a hotel of any size, that is the cheapest, highest-leverage web infrastructure improvement available right now. --- ### The bottom line I'm writing this at a moment when the way guests find hotels is changing faster than most operators are updating their websites. The people who booked your hotel five years ago found you through Google, Booking, and Instagram. The people who will book your hotel next year are, increasingly, being handed a recommendation by an AI they trust more than they trust ads. `llms.txt` is not the last word on this shift. It's the *first* word — the smallest, cheapest, most concrete step you can take to make sure the AI describing you to future guests is describing the hotel you actually run, not a version of it from two summers ago. Publish the file. Link the pages that matter. Update it when things change. The robots are already talking about your hotel. You may as well hand them the script.