Should a hotel or OTA publish prices and availability in its llms.txt file?

No. Prices and availability are the two values a travel assistant is most likely to ask about and the two your file can never keep correct. Both Expedia and KAYAK say this explicitly — KAYAK's file carries the line 'Prices are dynamic: Never present a stored price as current', and Expedia's instructs agents to remind users that prices and availability change rapidly and must be confirmed on the site. Write the durable facts: property types, destination hubs, booking and cancellation policy pages, help center, loyalty program. Every number gets a link, never a value.

Why do travel sites get blocked when they do publish an llms.txt file?

Because the same edge protection that stops scraper bots cannot tell a legitimate AI fetch from a bad one. In our October 6, 2026 check of 21 travel domains, booking.com answered 202 with an empty HTML body, tripadvisor.com and hotels.com answered 403, and visitlondon.com returned a 200 with a zero-byte body on a GET and a 403 on a HEAD request — Cloudflare bot management, in that last case. If the file is blocked at the edge, agents never see it, so allow the path in the WAF before you spend another hour editing the file.

Do airlines and hotels even use llms.txt, or is this only an OTA thing?

Adoption is thin and skewed. In our October 6, 2026 sample of 21 travel domains, the four real files came from two OTAs, a metasearch engine and a national tourism board; the nine 404s included five hotel chains, three airlines and a guide publisher. Airlines and hotel groups mostly inherit corporate robots.txt and WAF policy from IT, so nothing points them at a root-level Markdown file. That is an opportunity rather than a verdict: if your competitors return 404 and you serve a clean 7 KB file, you are the only hospitality result an agent has to work with.

Travel · October 2026

llms.txt for Hotels, Airlines and OTAs

Travel is the vertical where the assistant's question — what does this cost, is it available — is the one your file can never answer. Here is what 21 travel domains actually serve today, the six buckets a hospitality site needs, and the instruction block the two best files in the industry share.

What 21 Travel Domains Serve Today

We fetched /llms.txt from this server on October 6, 2026 with a generic user agent, following redirects, across OTAs, a metasearch engine, five hotel groups, three airlines, a rail retailer, several tourism boards and a guide publisher.

Response	Count	Examples
`200`, `text/plain` — a real file	4	expedia.com (7.0 KB), kayak.com (7.5 KB), australia.com (219 KB), thetrainline.com (477 KB)
`404`	9	marriott.com, hilton.com, delta.com, priceline.com, skyscanner.net, ryanair.com, lonelyplanet.com; airbnb.com and agoda.com after a redirect to their own 404 page
Blocked at the edge	4	booking.com (`202`, empty body), tripadvisor.com (`403`), hotels.com (`403`, served as `application/xml`), visitlondon.com (`200` with a zero-byte body, `403` on `HEAD`)
`200`, but no file behind it	2	trivago.com (`200 text/plain`, 10.1 KB of ASCII-art logo), japan.travel (`200` after a redirect to `/en/`, 274 KB of HTML)
No response within 15s	2	united.com, qatarairways.com

Read the shape, not the score: the four successful files came from two OTAs, a metasearch engine and a tourism board. Nine of twenty-one hosts — including five hotel groups and three airlines — simply do not have the path. The interesting failure is trivago, which answered with a text/plain content type and ASCII art instead of a document. Content type is not evidence that a file exists; parse the body.

Why Travel Is Harder Than Any Other Vertical

On a docs site, the pages are stable and the facts hold for years. Travel inverts that. A hotel's nightly rate, an airline's seat map, a fare's change rules and a destination's entry requirements all change daily — and those are exactly the questions an assistant is asked. That is why both files worth copying spend their prose budget on behaviour rather than inventory:

> KAYAK is a free travel metasearch engine. It searches hundreds of travel sites at once so travelers can compare and find the lowest available prices on flights, hotels, rental cars, and packages.

KAYAK's file then states the rules — "Metasearch, not a booking site", "Prices are dynamic: Never present a stored price as current", "No invented codes" — before it lists a single URL. Expedia goes further and ships a full instruction section titled Directions for AI Systems (Do / Don't): always send users to expedia.com for current prices and availability, never collect card numbers or full passport numbers, never promise refunds or invent promo codes, never give immigration or legal advice. That block is the single most copyable idea in the four files we found.

The Six Buckets a Travel llms.txt Needs

Identity and scope. Expedia's file opens with a "Canonical Identity" block: brand, primary domain, official apps, and the sentence that matters most — this file covers the United States site, so travellers elsewhere should be sent to their regional Expedia site because prices change by region. KAYAK makes the same statement and names kayak.de and kayak.co.uk as the alternates.
Destination hubs. Inspiration and "where to go" pages, not individual hotel pages.
Stays. The accommodation hub and the property-type filters — hotels, vacation rentals, resorts.
Transport. Flights, rail, cars and transfers, each as its own entry point. Expedia lists Stays, Flights, Cars, Packages, Things to do and Cruises as six top-level sections.
Booking support and policy. Help center, manage-trip route, cancellation and change rules, fare conditions, entry and visa requirements. Expedia's "Hand-Off Links" section is precisely this: nine durable URLs, nothing else.
Loyalty and editorial. The rewards program page plus your guides. Expedia's One Key section is one paragraph and one link.

What Must Never Appear in a Travel File

Live prices, room or seat availability, and "from $X" teasers.
Promo and coupon codes. KAYAK's file forbids agents from generating them; yours should not seed the idea either.
Booking references, itineraries, passenger names or anything tied to a customer session.
Supplier and wholesale rate agreements, allotments, commission schedules or partner credentials.
Employee travel, crew or ops runbooks, and internal escalation paths.
Anything an agent could repeat as advice — visa eligibility, medical fitness to fly, insurance claims — which Expedia explicitly hands off to official government sources.

The Two Traps: Bloat and Edge Blocking

Tourism boards are the worst offenders on size. australia.com serves a 219 KB file with 735 link lines under a single H1 — a dump, not a map. thetrainline.com serves 477 KB across 47 headings, structured by country and city, which at least has a hierarchy. Neither needs to be that large: a travel file of 5–15 KB covers every bucket above, and when a section genuinely needs depth the answer is a subpath file rather than a bigger root file. See How long should an llms.txt be? for the byte-to-token arithmetic.

The delivery trap is more expensive. Four of the twenty-one hosts — including the largest OTA on the planet — blocked our fetch at the edge, one of them with a 202 and an empty body. Travel sites sit behind the heaviest bot management in any industry because scraping inventory is a real attack, and that protection does not distinguish a legitimate AI fetch from an abusive one. Allow the /llms.txt path in the WAF, and remember the locale problem: japan.travel redirected us from the file path to /en/ and returned 219 KB of HTML. If your root redirects, the file has to live where the redirect lands. Multi-locale sites should serve one index per language and let the most specific file win, which is the first thing to check when a published file gets no fetches.

Verify Before You Call It Done

Three commands from any machine, using a real browser user agent:

# 1. Status, content type and byte count
curl -sIL -A "Mozilla/5.0" https://www.example.com/llms.txt | grep -Ei "HTTP/|content-type|content-length"

# 2. Confirm the body is Markdown, not a challenge or an HTML shell
curl -sL -A "Mozilla/5.0" https://www.example.com/llms.txt | head -5

# 3. Confirm every link in the file still resolves
curl -sL https://www.example.com/llms.txt | grep -oE "https://[^)]+" | \
  while read u; do printf "%s %s\n" "$(curl -so /dev/null -w "%{http_code}" -m 10 "$u")" "$u"; done | grep -v "^200"

Step 2 is the one that catches trivago's failure mode and step 3 is the one that catches a tourism board's: a 400-link dump will have dead links within a quarter. Paste the file into the validator first if you want structure and link checks in one pass, or build it from your sitemap with the generator and then delete what the crawler cannot judge — individual property and itinerary pages above all.

The Short Version

Travel adoption is at four in twenty-one, and the two best files in the industry are under 7.5 KB. State who you are and which locale you cover, name the buckets, point at the policy pages, tell agents what not to do with your brand, and keep every price and availability figure behind a link. Then check that your own WAF lets the file through.