AI for hotels

Is AI Sending Guests to a Hotel That Isn’t There? A Hotelier’s Field Guide to AI Agents

AI Agents Hotel Ghost Hunt

What “booking with AI” really means today, what the research says about how AI picks a hotel, and the cheap, practical work that gets you ready.


Every hotel has a ghost.

I don’t mean the one in Room 237.

I mean the out-of-date version of your hotel that lives on in old web pages. Last year’s prices. The restaurant that closed in spring. The pool that’s been “opening soon” for three summers.

For years that ghost sat quietly in the dusty corners of the internet. Now AI assistants read those corners for a living, and repeat what they find with total confidence.

Here’s how far it can go.

On 29 July 2026, a luxury hotel marketing firm called Americas Great Resorts asked three AI assistants the same 60 traveller questions in a single day, and logged all 824 hotel recommendations. Two of the assistants recommended the Mandarin Oriental, Miami. One suggested it three times. The other did it twice, once in a list of the city’s top five luxury hotels.

The hotel closed in May 2025. It was brought down in a controlled implosion on 12 April 2026, 108 days before the test.

(I’d love to read those reviews.)

Nobody lied. The firm is careful to point out that a couple of documented mistakes don’t make an error rate. The assistants were simply reading pages nobody had got round to updating.

Your ghost is probably far less dramatic. But this is the clearest picture I’ve seen of how AI and hotels actually meet today, and it has very little to do with robots booking rooms.

Owners and GMs often ask what’s really happening with AI agents. So here’s a field guide: what’s real, what the research says, and what’s worth doing now.

Four ways to “book with AI”

“Booking with AI” gets used for four very different things. Picture yourself planning a family weekend at the beach.

You say to the AI…What happensHow common is it today?
“Help me choose.” Find a family-friendly beach hotel for May, under $300 a night.It gives you a shortlist and the reasons. You book later, yourself, on a hotel or travel website.Very common, and growing fast
“Show me where to book.” What does the second one cost for 12 to 14 May?It shows prices and a link. You tap through and book on that website.Common
“Book it, but let me check first.” Book that one with free cancellation.It fills in the booking inside the chat. You check the details and press pay.Real, but rare. The biggest launch so far is US-only, in English
“Just sort it out.” Somewhere nice by the beach in May. Under $300, free cancellation. Don’t bother me.It chooses and pays without asking you.Mostly demos and tests. Tiny numbers

The first two rows are booming. Adobe found that visits to US travel websites from AI assistants more than doubled in a year, and those visitors now book almost as often as everyone else.

The third row arrived properly in August 2026, when Google added hotel booking to its AI Mode with ten hotel and travel companies. US only, English only, and you still check and pay yourself.

The bottom row is where the hype lives, and the past few weeks have been busy:

Muse and Duet both ask for your approval before sensitive steps like paying. And most travellers on the planet can’t use any of the three yet without the right country, a premium plan or a spot off a waitlist.

Keep that in mind when you read the next numbers.

DoorDash, which handles hundreds of millions of orders every quarter, said this week it had seen “maybe several hundred orders in total” from outside AI agents. Booking Holdings said in the summer that AI-driven traffic was “less than 1% of total room nights.”

Travellers aren’t rushing to hand over their cards either. In a HUMAN Security survey of 2,400 Americans, 54% were happy for AI to plan a trip and 43% would let it book if it asked before paying. Only 12% would let it book without asking.

Even OpenAI let people buy directly inside ChatGPT, then pulled the feature in March 2026 (it called the move temporary). As Forrester put it, “Commerce isn’t just checkout.”

Small and early still matters. The pipes are being laid now, and access is likely to widen. But today the money sits in the top two rows. Which means your ghost matters more than any robot.

How AI actually picks a hotel

Researchers have spent the past year watching AI shoppers in lab tests. Four findings matter for hotels. (I’ve stuck to recent AI models where I can, and flagged it where a study used older ones. This field moves fast enough to embarrass a paper within months.)

1. It reads the same few pages, over and over. In that same hotel test, all ten of one assistant’s Los Angeles answers quoted the same two Michelin Guide list pages. In Chicago, every answer leaned on the same two Tripadvisor lists. Across six destinations, just 23 hotels took half of all 824 recommendations. Think of a well-read friend who only ever reads three magazines. If those pages are out of date, so is your hotel.

2. It changes its mind with every update. In the ACES study, a team including Columbia Business School researchers sent AI shoppers into a pretend online shop, then repeated the test with newer versions of the same AIs. One fitness tracker went from being picked 45% of the time to 77%. A phone case picked 63% of the time was picked just 5% of the time by the next version. Back in the hotel test, the three assistants agreed on a single best hotel in only two of six destinations, on the same day. There’s no permanent “number one in AI.” Anyone selling you one is selling a snapshot.

3. A wrong answer is worse than no answer. This one should change your to-do list. In the RITE study, published in October, ten recent AI models worked through realistic online shopping tasks. The researchers quietly swapped one piece of information for something believable but wrong. Success fell from about three tasks in four to about one in three. When the information was obviously broken, the AIs did better, because they noticed and went to check. A believable mistake sails straight through to the guest. So last year’s resort fee, an old cancellation rule or a restaurant that closed in spring will do more damage than a gap.

4. “Only 2 rooms left!” barely works on it. A Cue’s Hidden Price, another October study, measured how much sales wording swayed AI shoppers buying headphones and coffee machines, counted as how much more they’d pay. “Hurry, almost gone” was worth less than 2% extra on every model tested. A strong “number one best-seller” claim was worth about 21% on one popular model, and less on most others. The models were a generation old and nobody tested hotels, so treat it as a hint, not a rule. Still, it’s delicious: the industry’s favourite urgency trick may be the one machines simply shrug at.

The common thread? AI rewards facts that are true, consistent and easy to find. And the biggest platforms aren’t rewarding tricks. Google’s own guidance tells website owners to skip AI shortcuts, and says its search ignores llms.txt, a special file some sites add for AI tools.

Who does what (the food delivery version)

If you’ve ever ordered food through an app, you already understand how AI booking works. Three players, three jobs.

  • The app builds the shop window, takes the payment and decides which cities it launches in. In travel, that’s the AI assistants and booking websites.
  • The ordering system links each kitchen to the app, so the menu shows what’s really available and orders arrive correctly. In hotels, that’s your brand’s systems or, if you’re independent, your booking engine and channel manager (the software that sends your rooms and prices to booking sites). It’s how live prices, rooms and rules flow out, and how bookings, changes and cancellations flow back. This is the “pipes before they’re needed” work, and it’s happening across the industry. Hotel software firm Apaleo says hotels’ use of its AI connection grew 70% between July and September 2026, mostly helping staff. One operator used it to change 2,000 reservations in one go.
  • The restaurant keeps the menu right: the dishes, the prices, the photos, the opening hours. In hotels, that’s you. Nobody else can tell AI that your pool is closed for refurbishment until March.

And when a delivery goes wrong, it’s the restaurant the customer blames. Same here. When an AI booking goes wrong, the guest still walks up to your front desk.

If you own or run a hotel, the third job is yours. It’s the cheapest of the three, and it pays off whether AI agents take over next year or never.

AI Agents Hotel Ghosts Unveiled

The 30-minute ghost hunt

Once a month, put the kettle on and go looking for your ghost.

  1. Ask the AIs about yourself. Use three different assistants, logged out, in a private browser window, so you see what a stranger sees. Ask what guests ask: “best family hotel near [your landmark]”, “does [your hotel] have free parking?”, “what’s [your hotel]’s cancellation policy?” Write down every wrong or missing fact.
  2. Find where each ghost lives. Trace each mistake back to its source: your website, your Google Business Profile, your listings on travel sites, that “top 10” article from 2023. Fix it there. Start with anything that costs a guest money or sleep: fees, rules, room types, accessibility, outlets that have closed.
  3. Make your facts match everywhere. Same amenities, same rules, same room names on every channel. Contradictions are where wrong answers are born.
  4. Read what the AI reads. Note which websites the assistants quote for your area. If you’re missing, or described wrongly, that’s your PR and partnerships list.
  5. Count the AI visitors. Ask whoever runs your website data to show visits from AI assistants separately. If you use Google Search Console (Google’s free tool for website owners), it now has a report for Google’s AI features. The numbers will be small. Watch the trend, and whether those visitors book.

Then give AI something worth repeating

The ghost hunt removes the wrong things. This part makes the right things easy to find. It’s mostly good, old-fashioned hotel marketing, which is rather the point.

  • Great photos, explained in words. Google’s guide for marketers says to use high-quality images that are relevant to the page. Show every room type, the view you actually get, the bathroom, breakfast. Then describe them in captions and image descriptions, because words are what AI tools pick up most reliably. Adobe found that more than a quarter of what’s on a typical hotel home page isn’t easy for AI to read.
  • Answer the questions guests actually ask. Parking, check-in times, cots, pets, the walk to the beach. Put the answers in plain words on your own website, in one easy-to-find place. If you don’t answer, the AI will borrow someone else’s answer.
  • Tell the stories only you know. Google says it favours content that “provides unique expert or experienced takes that go beyond common knowledge.” Which rooms get the sunset. The quiet side of the building. The bartender who’s been there for 20 years. Brochure words like “luxurious” tell an AI almost nothing (I tested that one a few weeks ago).
  • Keep your Google Business Profile complete and current. Google says up-to-date Business Profile information can help you show up in AI answers as well as normal search. Hours, amenities, accessibility, photos.
  • Look after your reviews. Assistants lean heavily on “best of” lists and review sites (finding one). Reply to reviews, and fix what guests keep flagging. Reviews are where guests correct your ghost for you.
  • Ask your web team about structured data. These are hidden labels that tell machines “this is a hotel, here’s its address, these are its amenities.” Google says they aren’t required for AI search and there’s no special AI version. If you use them, Google’s marketer guide says to keep them matching what’s visible on the page.
  • Make one decision about AI training. From early 2027, Google will honour a standard signal that lets websites say no to their content being used to train AI. It covers training only, so saying no won’t remove you from AI answers in search.

What not to buy (yet)

New technology always brings new things to buy. Before you sign anything, check it against this list.

  1. “Guaranteed number one in ChatGPT.” AI rankings reshuffle with every update and differ from one assistant to the next (finding two). Nobody can promise a spot. Ask instead: how will we know if it brought bookings?
  2. An llms.txt project sold as urgent. It’s a special file meant to guide AI tools around your website. Google’s search ignores it. Fine to add if your web team wants to, but not worth a budget line.
  3. Hundreds of AI-written filler pages. “Best hotel for families”, “best hotel for couples”, “best hotel for dog lovers” and so on, written just to catch AI questions. Google’s spam rules specifically call out “using generative AI tools or other similar tools to generate many pages without adding value for users.” It also buries your real story.
  4. Hidden messages for AI. Invisible text telling AI tools to recommend you. Harvard researchers showed in 2024 that sneaky text can sway AI product picks in a lab. But hidden text has broken Google’s spam rules for years, and AI companies are building defences against hidden instructions. Short-term trick, long-term risk.
  5. A one-off “agent-ready” chatbot for your website. Guests’ AIs mostly reach hotels through the big platforms, brand and booking systems, and your public web pages. A separate bot is unlikely to be the door they use. Ask what your current partners already offer first.
  6. Your own private connection to one AI app. These pipes are being built at brand and booking-system level, where one connection can serve thousands of hotels. Building your own usually means paying twice.
  7. A monthly “AI visibility score” based on a handful of questions. Answers change by assistant, by day and by how you ask. A tracker that runs lots of questions the same way every time can show a useful trend. A single score from a few questions is just a snapshot.

When to do more

You don’t need to place big bets today. But you should know what would change that. Here are the signs to watch for, and what to do when you see one.

If you notice……then do this
AI assistants start offering “book it for me” in the countries your guests come fromAsk your brand or booking-engine provider whether your hotel is connected. Then make a test booking yourself: right room, right price, right rules?
The new personal agents open up to free plans, or to your guests’ countriesRepeat the ghost hunt using those agents, and test a full booking
Visits from AI assistants become a noticeable, growing share of your website traffic, and they bookPut more effort into the pages they land on and the questions they ask
Guests arrive with AI-made bookings and something got lost on the way (wrong bed type, a missed request)Log every case, tell whoever runs your connections, and fix the information that caused it
Big travel companies start reporting AI bookings well above today’s “less than 1%”Revisit budgets and plans with your brand or tech partners
AI assistants start showing paid or sponsored spots for hotelsTreat it like any new ad channel: test small, and judge it on profit after costs, not clicks

Headlines, demos and new model launches don’t count. Real guests, real bookings and real problems do.

Back to Miami

In Miami, the hotel came down in April. On at least one day that summer, AI was still recommending it.

Your ghost is probably smaller. A fee that changed. A spa that moved. A sea view that’s now a building site. But AI repeats a small ghost just as confidently as a big one.

You can’t stop AI talking about your hotel. Nobody can. You can make sure it’s describing the one that’s actually standing: the right rooms, the right rules, the pool that really is open.

Every hotel has a ghost.

Time to go and meet yours.

0 comments on “Is AI Sending Guests to a Hotel That Isn’t There? A Hotelier’s Field Guide to AI Agents”

Leave a Reply

Discover more from Hotelemarketer by Jitendra Jain (JJ)

Subscribe now to keep reading and get access to the full archive.

Continue reading