Restaurant Discovery: How AI Agents Choose Where to Send Customers
"Find me a restaurant near the station that does good vegetarian food, has outdoor seating, and can take a booking for four people this Saturday at 7pm." This is a typical request people make to AI agents. It sounds simple. Getting a useful answer depends entirely on whether restaurant websites expose the right information in the right way.
For a multi-site restaurant group, the difference between being readable and unreadable to agents can be the difference between capturing this demand and being skipped entirely. When you run an audit on a typical group website, the same problems come up again and again.
The Audit Findings
A common setup is one page per restaurant on the group's website, plus a standalone microsite for a few locations.
The menu is usually the biggest problem. Many restaurants display their menu only as a PDF download, with no HTML menu anywhere on the site. An agent asked to check whether a restaurant has good vegetarian options cannot answer without downloading and parsing a PDF, which most agents either cannot do or do poorly.
Opening hours are often displayed as an image in the footer of each page: a stylised weekly timetable in the restaurant's brand fonts, with no text equivalent. Agents cannot determine whether a restaurant is open on Saturday evening.
The reservation system is commonly a third-party widget embedded via iframe. The iframe loads a different domain with no connection to the restaurant's structured data. Agents can find the "Book a Table" button, but interacting with the contents of the iframe is unreliable, and the booking widget itself requires selecting date, time, and party size from custom dropdown components that most agents cannot operate.
Location data is frequently sparse. The address may appear in the footer as text (good), but there are often no geo-coordinates in the markup, no Schema.org LocalBusiness data, and no connection between the website and the restaurant's business profile.
What to Change
HTML menus with structured data. Convert every menu from PDF-only to an HTML page with Schema.org Menu and MenuItem markup. Include each dish's name, description, price, and dietary flags (vegetarian, vegan, gluten-free) as structured data. You can keep the PDF available as a download, but make the HTML version the primary menu page.
Machine-readable hours. Add opening hours as Schema.org OpeningHoursSpecification markup on every restaurant page. Supplement the visual timetable image with a plain text version in an HTML table. Include special hours for bank holidays and seasonal changes as validFrom/validThrough entries.
Booking system accessibility. Work with your reservation provider to implement a direct booking URL scheme. Rather than relying on the iframe widget, include a link in a format such as book.provider.com/restaurant-slug?date=YYYY-MM-DD&time=HH:MM&party=N on each restaurant page. An agent can construct this URL from the user's requirements and send them directly to a pre-filled booking page. Keep the iframe widget as a fallback for users who prefer to browse availability manually.
Rich Schema.org markup. Give each restaurant page full Schema.org LocalBusiness markup including name, address, geo-coordinates, cuisine type, price range, accepted payment methods, and links to the menu, reservation URL, and opening hours.
The Results You Can Expect
With these changes, agent-referred reservations can grow noticeably across a group. The gains tend to concentrate in locations with high foot traffic, where agents are recommending restaurants to people already nearby.
Average party size from agent bookings tends to run higher than from a website's direct booking widget, because agents are often used for group planning, which means higher per-booking revenue. No-show rates for agent-booked reservations also tend to be low, possibly because agent bookings are more intentional: users have to specifically ask for a restaurant and confirm the details, rather than casually browsing availability.
Why Restaurants Specifically Benefit
Restaurant search is one of the most common agent tasks because it involves combining multiple filters (location, cuisine, dietary requirements, availability, price range, ambience) that humans find tedious to cross-reference manually. AI agents are ideally suited to this if they can access the data.
The fixes are not complex. Converting a PDF menu to HTML with structured data is straightforward, and the Schema.org markup can be templated so that additional sites take very little time once the first is done. The booking URL scheme requires coordination with the reservation provider but no custom development.
Any restaurant with a website can do this. The ones that do it first will capture a growing share of the agent-referred dining market. The ones that stick with PDF menus and image-based hours will become invisible to the agents that customers increasingly rely on.