AI Agents Now Book Tables and Rooms for Your Customers. Can They Book Yours?

Your next customer may never open a search page. They may text an agent: book me a table for four on Friday, somewhere near Queen West. The agent picks a place. Then it tries to book it. Whether it can finish is now part of your visibility.

From recommended to booked: what changed

Being recommended by AI used to be the finish line. A person read the answer, chose a place and booked it themselves.

Now the agent tries to do that last step too. Here is what has changed in the past few weeks:

• Lola, September 29. The Booking Holdings-backed assistant launched with memberships and new partners. Travel Weekly reports it can already book restaurants and hotels, and that it draws on Booking Holdings brands such as Booking.com and OpenTable.
• Meta’s Muse, September 29. Meta expanded Muse to small businesses the same day. As a personal agent, Muse then ran into trouble on hotel sites, as Skift’s tests show below.
• Instinct, by text message. Instinct is a personal agent you message on iMessage or WhatsApp. TechCrunch’s October 10 roundup shows how many agents now live inside text threads.
• Google AI Mode, hotels. Google began rolling out hotel booking in AI Mode in the U.S. on August 27. People compare rooms, check cancellation terms and pay with Google Pay in the chat. The hotel or booking platform still takes the booking.

The pattern is the same everywhere. Recommendation is step one. Booking is step two. Agents only finish step two where they can get through.

We covered the discovery side when Google’s AI Mode started watching for new restaurants and deals. This post is about what happens after the shortlist.

The two doors agents use to book you

An agent that wants to book has two ways in.

When Door 1 fits

If you already take bookings through OpenTable, Libro or Booking.com, an agent that connects to those platforms can often see your availability. You rely on the platform to stay current. Your job is the listing.

When Door 2 matters

If you take direct bookings, or you are not on the platform the agent uses, your own page becomes the route. This is also the door chains want agents to use. Hilton’s technology chief told Skift the aim is to meet travellers in AI tools and bring them back to Hilton.com to book.

Why agents fail on real booking pages

Door 2 is where things jam. Two pieces of evidence show how.

What Skift’s hotel tests showed

Skift tested Muse on the Hilton, Marriott and Hyatt sites. Muse failed at the room-and-rate step on all three. Instinct completed searches on all three and returned matching member-only rates.

Hilton says it has not blocked Muse or any major agent, and that it accepts bookings from several agents, Muse included. Skift’s point is that letting agents in on paper and letting them finish are not the same thing.

These are big chains with large engineering teams. A small site with a custom widget and a pop-up has plenty to trip over as well.

What MERJ’s agent research found

MERJ began its research in November 2025 and warns that agent behaviour changes as models change, so its findings are version-specific. Even so, the failures are easy to picture:

• Many agents read the page’s accessibility tree, not just what you see. A div styled to look like a button can read as plain content with nothing to click.
• A button hidden with aria-hidden can be visible on screen and still invisible to an agent that reads structure.
• Native confirm dialogs can stall an agent that has no way to handle them.
• Misleading labels send agents to the wrong control. In one test, an agent asked to save a document clicked Cancel first in all 25 runs.

The lesson is simple. A booking page that looks fine to you can be a dead end for an agent.

Visibility without bookability: who gets the booking

If an agent recommends you but cannot finish with you, it may finish somewhere else. Platforms are built to be finished. Their listings are structured and their partner links exist. An independent’s booking page was usually built for a person with a thumb and a bit of patience.

That is a risk, not a certainty. But it points one way: independents can lose bookings to OTAs and big platforms without ever losing a recommendation.

Keep this in proportion. It is early. Skift reports that Booking.com says AI assistants still drive under 1% of its room nights. The fixes are cheap compared with being late. And the first step is still being recommended, which is the work of AI search optimization.

The Toronto booking checklist

Pick the doors that apply to your business, then tick each box.

OpenTable and Libro

OpenTable announced on April 20, 2026 that it had acquired Libro, the Canadian reservation platform. Reports say Libro keeps operating under its own brand. Both connect to Reserve with Google, and Lola draws on OpenTable through Booking Holdings. Keep hours, party sizes and availability current on whichever you use.

Reserve with Google

This puts a reserve button on your Google search and Maps listing. Google’s Canada blog describes AI booking that links diners to partners such as OpenTable and Libro to finish the reservation. If you use one of those platforms, switch the Google connection on.

Booking.com

Booking.com is a launch partner for hotel booking in Google AI Mode, and Lola members get hotel rates through it. Hotels should keep rates, policies and photos accurate. Note that Google’s AI Mode hotel booking launched for U.S. users, so check what is live in Canada before you plan around it.

Your own booking engine

This is the door you control, and the next section covers it. If you want help tying this together, our restaurant and hospitality marketing team works on exactly this mix of listings, reviews and booking pages.

Make your own booking page agent-friendly

You cannot change what Hilton does. You can change your page. Most fixes are plain web basics, which is MERJ’s point about accessibility and standard markup.

• Use real buttons and links. A styled div can look like text to an agent.
• Label everything clearly. Say “Book a table”, not “Go”. Give the date, party size and time fields visible labels that match what they do.
• Do not hide visible controls. If a button shows on screen, it should also show in the accessibility tree.
• Avoid blocking pop-ups. Newsletter pop-ups, cookie banners and native confirm boxes can stall an agent. Make them easy to dismiss and keep them off the booking form.
• Keep the path short. Fewer steps and no forced account means fewer places to get stuck. This one is our recommendation, not a finding from the research.
• Check your bot protection. Reports point to bot checks as one place agents get stopped. Make sure yours is not turning away the visitors you want.

Run Lighthouse’s Agentic Browsing check

Chrome’s Lighthouse now has an Agentic Browsing category. Per Chrome’s documentation and guides, it looks at an llms.txt file, WebMCP, the accessibility tree (names, labels, structure and visibility) and layout shift. Google describes it as a way to gather data and give actionable signals, not a final ranking.

Open Chrome DevTools, go to the Lighthouse tab, choose Agentic Browsing and run it on your booking page. You need a recent version of Chrome. If a developer maintains your site, technical SEO work is the right place to put the fixes.

Test it yourself this week

You do not need a tool to find out where you stand. You need an agent and twenty minutes.

1. Pick your agents. Use whichever you can get into: Google AI Mode, Lola, Muse, Instinct, ChatGPT or Claude. Two or three is enough.
2. Ask a real customer question. For example, “Book a table for four at [your restaurant] on Friday at 7” or “Find me a room at [your hotel] for two nights.”
3. Let it run and do not help. Watch the whole path. Note every step where it hesitates, asks you to take over or stops.
4. Write down where it got stuck. Was it the search result, the platform, your page, a pop-up, a bot check or the payment step?
5. Repeat without your name. Ask “Book a table for four in [your neighbourhood] on Friday.” See whether it picks you, and whether it can finish.
6. Run the Lighthouse check on your booking page. Fix the top item first, then rerun your agent test.
7. Test again in a month. Agents change fast. MERJ says its own findings are version-specific, so one test is a snapshot.

To see whether AI recommends you in the first place, run our AI Visibility Check.

Frequently asked questions

The verdict

Two doors, one test. Platforms are the smoother route when you are on them. Your own page is the route you control, and the one most likely to jam. Check both this week.

Being recommended gets you looked at. Being bookable gets you the table.