We Audited 12 Hotels for AI Search. Most Fail on the Same Four Things
By Gordon Hwa, Founder, Khepri Digital
Published · 7 min read
What did we audit, and how?
We looked at 12 single hotels and resorts that we audited between 19 June and 19 October 2026: two in Bali, and one or more each in Malaysia, Japan, Spain, the United Kingdom and Singapore. Most are luxury or upper-upscale; one is a 3-star city hotel. Each audit checked the hotel's own site, its structured data, its robots.txt, its entity records (Wikipedia, Wikidata) and how AI assistants answered questions a guest might ask.
This is not a random sample. These are hotels we audited, mostly while pitching for their business, so properties with visible problems are over-represented. The audits were written by our team at different times, and not every audit checked every item. Where a check wasn't done, we leave that hotel out of the count rather than guess, so each finding below states its own base.
We don't name the hotels. Some are clients now; the others never agreed to be published.
What did most hotels get wrong?
Four gaps came up again and again. All four are about whether a machine can read and trust the hotel's basic facts. None needs a new kind of file.
| Finding | Hotels affected | Hotels checked |
|---|---|---|
| Name, address or key facts inconsistent across the web | 6 | 7 |
| No Hotel or LodgingBusiness schema | 9 | 12 |
| No structured data at all | 4 | 12 |
| No FAQ content on the site | 6 | 10 |
| Star ratings in schema that Google treats as self-serving | 2 | 2 with ratings |
| No Wikipedia article | 4 | 7 |
| Blocks an AI crawler in robots.txt | 1 | 12 |
| Named in at least one non-branded AI answer we tested | 9 | 10 |
Inconsistent facts
Six of the seven hotels we checked had a name, address or other key fact that differed depending on where you looked: the website, Google Business Profile, booking sites or Wikidata. An AI answer that finds two versions of the same fact either hedges or picks one.
Google's own list of what helps in AI features includes keeping Business Profile information up to date and making sure structured data matches the visible text on the page (Google Search Central). Both are about consistency.
Missing Hotel schema
Only three hotels marked themselves up as a Hotel (or the broader LodgingBusiness type) in schema.org structured data. Four had no structured data at all. Without it, a crawler has to infer from the page that this is a hotel, where it is, and what it offers.
No answers to the questions guests ask
Six of the ten hotels we checked had no FAQ content: no page that states, in plain sentences, things like the distance to the airport, whether children are allowed, or what the spa costs. Those are the passages an AI answer can quote. Copy written as mood ("a sanctuary where time slows") gives it nothing to quote.
Self-serving ratings
Two hotels had star ratings in their structured data. In both cases the ratings were the hotel's own or copied from elsewhere. Google treats reviews that a business publishes about itself as self-serving, and doesn't show them as review snippets for local businesses (Google Search Central, review snippets). They add risk and no benefit.
Does llms.txt matter?
Not much, on this evidence. Ten of the 12 hotels had no llms.txt, and the two that had one were broken: one linked mostly to pages that returned errors, the other was an auto-generated list with no useful content. Yet 9 of the 10 hotels we tested were already named in AI answers.
Google states that you don't need new machine-readable files, AI text files or markup to appear in AI Overviews or AI Mode (Google Search Central). llms.txt is a proposed convention. We still add one to the sites we build, because it takes ten minutes, but it is not why a hotel gets cited, and we no longer count its absence as a failure.
Blocking AI crawlers was rare too: one hotel of 12 blocked one in robots.txt.
What changes when the gaps are fixed?
When the facts are consistent and the pages answer real questions, both rankings and AI citations move. These are two hotels we work with, measured in their own Search Console and our AI-visibility tracking.
Neither result came from one change. Both came from the same unglamorous list: one consistent set of facts, Hotel schema that matches the page, and pages written to answer what guests ask.
How do you check your own hotel?
Five checks, in this order. The first three take an afternoon.
- Search your hotel's name and compare. Website, Google Business Profile, the two booking sites you sell on most, and Wikidata if there is an entry. Write down every place the name, address, room count or category differs.
- Test your structured data. Run your homepage through Google's Rich Results Test and check that a Hotel or LodgingBusiness item appears, with the same facts as the page.
- Read your site as a guest with a question. Can you find, in a plain sentence, how far you are from the airport, what time check-in is, and whether children are welcome?
- Ask the AI. Ask ChatGPT, Google AI Mode and Perplexity three questions a guest would ask without your name in them. Note who is named and which sources are cited.
- Remove ratings you wrote about yourself from your structured data.
Frequently asked questions
How do hotels get recommended by ChatGPT and Google AI Overviews?
AI answers name hotels whose facts they can verify from several sources that agree: the hotel's own site, its structured data, Google Business Profile, booking sites and independent coverage. Consistent facts and pages that answer guests' questions matter more than any AI-specific file.
Do hotels need an llms.txt file?
No. Google says no new machine-readable or AI text files are needed to appear in AI Overviews or AI Mode. In our audits, 9 of 10 hotels were named in AI answers although none had a working llms.txt.
What schema should a hotel website use?
The schema.org Hotel type (or LodgingBusiness), with the hotel's name, address, geo coordinates, check-in and check-out times and amenities, matching what the page shows. Don't add review ratings the hotel wrote or copied itself.
How many hotels were audited, and where?
Twelve single hotels and resorts, audited between 19 June and 19 October 2026, in Indonesia, Malaysia, Japan, Spain, the United Kingdom and Singapore. They are not a random sample: most were prospects we audited while pitching.
Can I see the data?
The table above gives every count with the number of hotels checked. We don't publish the hotels' names; the per-hotel findings stay with each hotel.
Sources
- Google Search Central: AI features and your website
- Google Search Central: Review snippet (Review, AggregateRating) structured data
- Google Search Central: Intro to how structured data markup works
- Google: Rich Results Test
- Schema.org: Hotel
- llmstxt.org: The /llms.txt file
- Khepri Digital audit files for 12 hotels, 19 June to 19 October 2026. Client results: Search Console and Khepri AI-visibility tracking for Jaya House (May to September 2026) and The Legian Seminyak (June to August 2025 against July to September 2026).
This guide supports our AI Search Visibility service.
