Open each AI engine in a fresh session, ask the twelve questions below with your real area and property filled in, and score every answer 0–3 with the rubric. One evening of this tells you where you stand, who's winning instead of you, and which pages the engines are reading — which is exactly where fixing starts.
Why you should run this at all
When a traveller asks an AI assistant where to stay in your town, the answer names a handful of properties. Not a results page — a handful. In TakeUp's January 2026 survey of 300 American leisure travellers, 84% said a trusted AI recommendation would make them more likely to book a property, and more than three in four of the AI users had already booked a trip primarily on one. Those are exactly the travellers landing in Malaysia through the Visit Malaysia campaign now running.
Most operators have never once looked at what these engines say about them. The ones who look are usually surprised in one of two directions: absent where they assumed they'd appear, or present but described by an OTA's page rather than their own.
The run rules — these matter more than they look
- Fresh session, every engine. Log out or open a new chat. ChatGPT and Gemini personalise from your history; you want the stranger's answer, not yours. On Google, use an incognito window. The ask-links below are built to respect this: ChatGPT, for one, opens as a temporary chat.
- Phrase it like a traveller, not an owner. Travellers write "where to stay in Ipoh with kids", not your property's name. The battery below is built from real phrasing patterns.
- Screenshot every answer, with the date visible. This is your baseline. In three months, the comparison against these screenshots is how you'll know whether anything you fixed moved anything.
- Score as you go, not from memory. Print the sheet at the bottom, or copy it into a notebook. Memory flatters.
- Never retry a bad answer. Asking again until you appear is testing your patience, not your visibility.
The engines to run
ChatGPT (with web browsing on), Google Gemini, Perplexity, and Google itself — both the classic results and the AI Overview or AI Mode answer if one appears. That's the working set. Add Microsoft Copilot if your guests skew corporate; skip it otherwise.
The twelve queries
Replace the brackets with your real area, landmark and property. Say them the way a guest would.
Layer A — discovery. The traveller doesn't know you exist.
- best homestay in [your area]Ask →
- best boutique hotel in [your area] (or guesthouse — use your real category)Ask →
- where to stay in [your area] for [families / couples / a workcation — your real crowd]Ask →
- [your area] accommodation near [the landmark guests actually name]Ask →
Layer B — qualified. The traveller knows what they want.
- [your area] homestay with [parking / pool / big family room — your true differentiator]Ask →
- guesthouse walking distance to [landmark]Ask →
- is it better to stay in [your area] or [the rival area] when visiting [city]Ask →
Layer C — brand. They've heard of you and they're verifying.
Layer D — booking. The money layer.
- [your property] official websiteAsk →
- book [your property] directAsk →
- [your property] contact numberAsk →
Layer D is where platform capture shows itself. A guest asking these questions has already chosen you — if the answers route through Booking.com or Agoda, the commission is being charged on a decision you already won. When we ran George Town's Layer-A queries in July 2026, exactly one independent property surfaced through its own website; every homestay-tier property we could find appeared only inside OTA pages.
The rubric — score each answer 0 to 3
| Score | Meaning |
|---|---|
| 0 | Absent. You're not mentioned anywhere in the answer. |
| 1 | Mentioned. You appear in a list, with no reasoning attached. |
| 2 | Recommended. Named with a reason — "known for…", "good for families…". |
| 3 | The answer. Top recommendation, or your own site is the link given. |
↔ Table scrolls sideways on a phone
Two more columns to fill per query, and they're the valuable ones. Who is named instead of you — after twelve queries, a pattern of the same three or four names is your real competitor set, which is rarely who owners assume. And which sources the engine cites — the links under the answer are the pages that fed it. That list is a map of where visibility in your area actually comes from.
Reading your sheet
Mostly 0s in Layer A, whatever your Layer C looks like: you have a findability problem, not a reputation problem. Guests who know you can verify you; nobody new is being introduced. Start with how engines choose stays and your Business Profile.
Decent Layer A, but Layer D routes through platforms: you have a capture problem — you're being found and then handed to the OTAs at the last step. Start with the direct-booking playbook. This pattern is common and it's the expensive one, because it taxes your most convinced guests.
Strong scores, same three competitors everywhere: study their citations column. The pages feeding their answers are the rooms you're not in yet.
The scoresheet
Print this page — the sheet below is formatted for A4, and everything except the sheet and the queries drops away in print. One row per query per engine; most operators run twelve queries × four engines in about 45 minutes.
| Query (as asked) | Engine | Score 0–3 | Named instead of you | Sources cited |
|---|---|---|---|---|
↔ Table scrolls sideways on a phone
What this protocol can't tell you
It can't tell you why the engines answer the way they do for your specific property, what the gap costs in ringgit against your own booking mix, or what to fix in which order for the best return. That's the part that takes the hours: the full audit runs roughly two dozen queries per engine, tears down your public layer — site, structured data, Business Profile, review corpus, listing quality — and prices the leak with your numbers, not industry averages. If the sheet you've just filled in is mostly 0s and 1s, that's the conversation worth having: the AI-Visibility Audit.
And if your sheet comes back strong — genuinely, take the win. Screenshot it, date it, and re-run the battery each quarter. You're defending a position most of your market doesn't know exists yet.
Sources & dates
- 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.
- Search Engine Roundtable — Google testing direct hotel-site booking links inside AI Mode answers. May 2026.
- Travel Weekly / PhocusWire — hotel and flight booking announced for Google's AI Mode, not yet launched. June 2026.
- Stayfarer query log, George Town discovery queries — run 17 Jul 2026.
You'll have twelve answers and a score. Here's what to do with them.
If the picture was worse than you expected, that's the normal result and the fix list is knowable. The Position Check turns what you just saw into an ordered list of what to change first. The audit is the same work done properly, with your own booking data priced against it.
Keep your screenshots and the date. In three months they become the only honest measure of whether anything moved.
- The Position Check — the 15-minute self-audit that covers everything besides the engines
- How AI engines choose which stays to recommend
- The measured version: our AI-Visibility Audit