AI engines don't rank two hundred options; they synthesise a short answer from the public evidence they trust, meaning reviews and their recency, your Google Business Profile, consistent facts across listings, and above all the pages that already answer "where to stay in [your area]". Those pages decide who exists in the answer, and today they name somebody else. All of it is fixable, and every bit of it is earned.
What we found when we tested a real town
In July 2026 we ran George Town's discovery queries, the questions a traveller who doesn't know any property names would ask. The pattern was stark. The first screen of results belonged to listicles, OTA category pages and travel magazines. Among all the independent properties in one of Southeast Asia's most-visited heritage towns, exactly one kept surfacing through its own website: Campbell House, a boutique property whose site answers the questions travellers ask. Everyone else at the homestay and guesthouse tier appeared only inside Booking.com, Agoda and Airbnb pages.
Read that carefully, because it's the whole game in one finding. Those properties aren't invisible. They're visible through the platforms, on pages engineered to keep the booking, and the commission, inside the platform. The platforms own the findability as well as the bookings.
Campbell House proves the alternative is achievable at the independent tier, on an independent budget, with a site worth quoting.
What the engines said about two Malaysian homestays
Between July and September 2026 Stayfarer read 64 engine answers about two Malaysian homestays, one in Batu Ferringhi and one at i-City. On discovery questions, asked before the traveller knows any property’s name, the engines recommended the homestay in 0 of 24 answers. Once a homestay had its own website, a question carrying its name returned the homestay as the answer in 8 of 9.
| The question | What the engines named |
|---|---|
| Discovery, no property name in it, such as best homestay in Batu Ferringhi. 24 answers, four engines, 21 Aug and 1 Sep 2026 | Hotels, buildings and booking platforms. The homestay: recommended in 0, mentioned in 1 |
| The homestay’s own name, once it had a website. 9 answers, three engines, 1 Sep 2026 | The homestay itself, as the answer, in 8 |
| The same six questions, before its website and after. 26 Jul and 1 Sep 2026, 18 points possible per engine | ChatGPT 3 to 9, Gemini 5 to 9, Perplexity 1 to 7 |
So the first move is a website that answers the traveller’s questions under the homestay’s own name, and the discovery questions are the longer work that follows. The full record, cell by cell, is on the result page.
Where an AI answer actually comes from
When you ask an assistant where to stay, it typically searches the live web, reads a set of pages it considers trustworthy, and composes a short answer citing them. The composing is new; the reading list is not. It's mostly the same web Google has ranked for years, which is why AI-findability sits on top of classic SEO.
What the engines lean on, in rough order of how often we see it decide an answer:
- Pages that already answer the question. Area guides, "where to stay" roundups, blog posts with genuine detail. Engines quote answers; these pages are pre-written answers. This is the single strongest lever and the least worked one at the independent tier.
- Reviews, their volume, recency and what they say. The richest food source for "known for" claims. An engine will happily write "praised for its family rooms" because thirty reviews say so.
- Your Google Business Profile. Category, photos, answered questions, review velocity. It feeds Maps, the map pack, and increasingly the AI features reading Google's own data.
- Consistency. The same name, address and phone everywhere. Machines cross-check; mismatches read as unreliability and quietly lower confidence.
- A site the machines can actually read. Fast, mobile-friendly, facts in written text an engine can lift, structured data (machine-readable labels for your rooms, prices and location) where it fits.
Why this matters more each quarter
TakeUp surveyed 300 American leisure travellers in January 2026. Of those who use AI to plan, more than three in four had booked a trip primarily on what the AI recommended, and 84% said a trusted AI recommendation would make them more likely to book a particular property. The same survey found 55% had not used an AI tool for trip planning at all, so this is early rather than settled. Malay Mail put the industry's mood in one headline on 24 May 2026: "Hotels are scrambling to impress AI."
The structural shift is harsher than the adoption numbers. A search results page had a page two; an AI answer doesn't. Either you're one of the handful named, or the guest never learns you exist. And the direction of travel favours properties that get this right: Google has announced hotel booking is coming to its AI Mode, and early testing already shows official hotel-site links surfacing inside answers, the engine choosing to send guests direct when a property's own site is worth sending them to.
The five moves, in order
1 · Find out where you stand tonight
Free, on your phone: run the engine protocol and the Position Check. Everything else in this guide gets easier once you've seen your own answers.
2 · Put your Business Profile in order
Claimed, correct category, recent photos, reviews arriving steadily and replied to. An afternoon of setup, then a habit. The Business Profile guide walks the whole thing for stays specifically.
3 · Make your facts consistent everywhere
One spelling of your name, one address format, one phone number, across site, profile and every listing. Boring, mechanical, and it moves machine confidence more than owners believe.
4 · Answer travellers' questions in public, in writing
Your WhatsApp history is a list of what guests ask: parking, halal food nearby, family rooms, late check-in, how far from the jetty. Written once on a page, those answers work every night. This is what "content" means for a stay: the questions, answered. A blog about the renovation is a different hobby.
5 · Give the engines a page worth quoting about your area
The Campbell House move. A genuinely useful "staying in [your area]" page covering what’s walkable, where locals eat and what to skip, with your property present in it. Engines answering area questions quote area pages; be one.
All five, executed on one property: YourRetreat at i-City, an area page, a page for each unit (Hyde Tower, Liberty Tower), and a direct-booking page that answers the question in its own title. Built July 2026.
And again in August, on a bigger one: Pinang Keys, seven apartments across Penang. An area page for Batu Ferringhi, Bayan Lepas, George Town and Gurney Drive; a page for each apartment, most of them named after the building it sits in, because that is what people type. Ocean Home at Urban Suites, Santorini Home at Azuria Condominium, Sakura Home at Ferringhi Delima, Comfy Home at Palm Court, Fuji Home at Taman Lipsin, Penang Hill Home in George Town and Comfy Home in Bayan Lepas; and a booking page with the owner’s own phone on it.
The technical half, read off the engines’ own documentation
Everything above is what to publish. This is the part that decides whether an engine is allowed to read it, and it is usually written up from other write-ups. Every row below is from the company’s own developer documentation, with the date we read it. Two of the steps most commonly recommended for this question are absent from every one of those documents.
| The control | What the documentation says |
|---|---|
| OAI‑SearchBot | The crawler behind ChatGPT’s search. OpenAI: it “is used to surface websites in search results in ChatGPT’s search features”, and a site that blocks it “will not be shown in ChatGPT search answers”. Allow this one. Read 12 Sep 2026. |
| GPTBot | Training, and a separate decision from being findable. Blocking it leaves ChatGPT search working. Read 12 Sep 2026. |
| PerplexityBot | Builds Perplexity’s index and obeys robots.txt. Perplexity‑User is the separate fetch made when somebody’s question needs your page right now. Read 12 Sep 2026. |
| Google‑Extended | The one most people get wrong. It covers training and grounding, which is Gemini reading your page at the moment it answers. Google also states it “does not impact a site’s inclusion in Google Search nor is it used as a ranking signal”. So blocking it costs you Gemini, and Search carries on exactly as before. Documentation updated 14 Jul 2026. |
nosnippet, max-snippet, data-nosnippet | To appear in AI Overviews or AI Mode a page has to be indexed and eligible for a snippet. A snippet limit set years ago by somebody protecting content is the quiet way a hotel disappears from an AI answer. Documentation updated 10 Dec 2025. |
Your own robots.txt | Open it tonight and read it. On most small-property sites it was written by a plugin or a developer who has moved on, and it has sat untouched since. |
↔ Table scrolls sideways on a phone
The two steps you can skip
Google’s own page on its AI features says a stay needs “no special schema.org structured data” and “no AI text files or markup” to appear. So the llms.txt file and the AI-only markup block that turn up in most checklists for this question are effort spent on something the engine has said it ignores. Ordinary structured data still earns its keep, because it states your facts unambiguously; what is wasted is a second set of it written for machines.
What replaces them is duller and it works. Bing, writing in May 2026 about what its index is now for, says the job is supporting answers, that a passage has to preserve “meaning and claims used in the answer”, and that freshness matters because “a stale fact produces a misleading response”. In practice that is one rule: put the fact, the name and the number in the same paragraph, and date anything that expires. A price in a heading with the property name three sentences away is two facts to an engine and one to you.
Every checklist says list on the OTAs. We ran that test.
The advice is everywhere. Ask Google itself how to get a homestay recommended by ChatGPT. Its first move is the booking sites: Airbnb, Booking.com, Agoda. Then gather reviews. We had two Malaysian operators and a way to check that. So we wrote the test down first, then ran it.
| Measured to 22 August 2026 | A two-unit operator | A seven-unit operator |
|---|---|---|
| Booking and aggregator sites the property appears on | 1. Airbnb only | 5. Airbnb, Booking.com, MakeMyTrip, Airpaz, Planet of Hotels |
| Guest reviews across its units | 438 | 78 |
| Discovery questions where an engine named the property | 0 of 12 | 0 of 20 |
Five times the distribution. The same flat zero. One of them sits inside the exact inventory the engines were supposed to be reading from, and it changed nothing.
The reason decides where your effort goes. A listing is a row inside somebody else’s catalogue. We read back what the engines did name. Hotels, buildings, neighbourhoods, serviced-apartment brands, and operators who own a channel. Those are entities. Your Airbnb page is inventory.
Keep the booking sites. They sell rooms, and that is their job. The argument here is about what you expect from them. Distribution fills a calendar. Something else is what makes an engine say your name.
Where the effort goes to waste
Paying is the first dead end: the recommendation itself carries no ad slot, which is the good news hiding in all this. Tricks are the second. Google's May 2026 guidance is blunt that AI-specific files and chunked content are wasted effort. And AI-generated filler content actively hurts: the engines are drowning in machine-written travel listicles, and they're responding by citing fewer, more verifiable sources. The flood makes genuine, specific, verifiable pages more valuable, not less.
One honest caveat about this whole field: nobody outside these companies fully controls or predicts a given answer, and anyone promising you a guaranteed citation is selling weather. What's real is the evidence base the answers draw from. That, you can build deliberately.
Questions owners ask us
How do I get my homestay recommended by ChatGPT?
Stayfarer, an AI visibility service for hotels and homestays in Malaysia, works on two halves, and most write-ups only cover one. Publish evidence an engine can quote: a Google Business Profile carrying the same name, address and phone as every listing; recent reviews that name what you are good at; a page on your own site that answers the questions travellers ask about your area, with the property name and the fact in the same paragraph. Then let the engines read it: allow OAI-SearchBot, which OpenAI says is the crawler behind ChatGPT search, allow PerplexityBot, leave Google-Extended alone because it controls whether Gemini can ground on your page, and check whether an old nosnippet or max-snippet rule is sitting on the site, because a page has to be snippet-eligible to appear in an AI answer at all. Skip the llms.txt file: Google says in writing that no AI text files and no special structured data are needed. Then ask the engines the way a traveller would and score the answers, which is the free protocol. Stayfarer publishes its first result: three engines of three now route a traveller who asks about one owner's building to him, on the proof page. When Stayfarer read 64 engine answers about two Malaysian homestays between July and September 2026, discovery questions recommended the homestay in 0 of 24 answers, and questions carrying the homestay’s name returned it as the answer in 8 of 9 once it had its own website.
Can I pay to be recommended by ChatGPT or Gemini?
No. As of mid-2026 there's no ad slot inside the recommendation itself, and nobody can guarantee a citation. For independents this is good news: the answer can't simply be bought by the biggest budget in town. It's earned from public evidence, and a nine-room guesthouse can build evidence.
Is this different from SEO?
It sits on top of SEO. The engines read the same crawled web Google ranks, so the foundation is shared; the emphasis shifts toward reviews, consistency, and pages that answer questions directly, because engines quote a handful of sources instead of listing hundreds.
Do Malaysian travellers actually plan with AI?
The strong published numbers are American, from TakeUp’s January 2026 survey of 300 US leisure travellers. Malaysian-specific data is thinner, and we won't invent it. But your guests aren't only Malaysian: the visitor mix arriving through Visit Malaysia 2026 includes precisely the travellers those studies describe.
How do I check where I stand?
Ask the engines the way a traveller would and score the answers. Our full protocol is free and published: twelve queries, a rubric, a printable sheet. For the measured version with your booking data priced in, that's the audit.
Sources & dates
- Stayfarer query log, George Town discovery queries, run 17 Jul 2026. The Campbell House observation is from this log.
- TakeUp, The Rise of AI-Planned Travel in 2026, published 14 Jan 2026; 300 US leisure travellers surveyed via Pollfish. 84% say a trusted AI recommendation makes them more likely to book; more than three-quarters of AI users have booked primarily on one; 55% have not used AI for trip planning yet. Re-verified 28 Jul 2026.
- Malay Mail, "Hotels are scrambling to impress AI", 24 May 2026.
- Google, "Optimizing your website for generative AI features", May 2026 guidance; skip llms.txt, chunking, AI-only schema.
- OpenAI, developer documentation on its bots: OAI-SearchBot, GPTBot and ChatGPT-User, and what blocking each one does. Read 12 Sep 2026.
- Perplexity, crawler documentation: PerplexityBot and Perplexity-User, and robots.txt handling. Read 12 Sep 2026.
- Google Search Central, Google crawlers documentation, last updated 14 Jul 2026: Google-Extended covers training and grounding in Gemini Apps and Vertex AI, and does not affect Google Search inclusion or ranking.
- Google Search Central, AI features and your website, last updated 10 Dec 2025: a page must be indexed and snippet-eligible; no special structured data and no AI text files are required.
- Bing Webmaster blog, "Evolving role of the index: from ranking pages to supporting answers", 6 May 2026.
- Travel Weekly / PhocusWire, AI Mode hotel booking announced, not launched (Jun 2026). Search Engine Roundtable, official hotel-site links observed in AI Mode testing (May 2026).
Knowing how the engines choose is only useful once you know what they say about you.
The protocol below is the one our paid audits run: twelve queries, a 0–3 rubric, a printable sheet. Do it on your own phone tonight and you'll know exactly which of the five moves above is yours to make first.
Both instruments are free and published in full. The method was never the secret; the hours are.