How can a restaurant tell whether AI search led to bookings?

A restaurant can be named in an AI answer and still receive no booking from it. A guest may choose the venue without clicking a link, book through a marketplace, call the host, or never visit at all. To judge the work, measure the journey in separate steps: correct recommendation, arrival, useful action, booking handoff and confirmed reservation. Each step answers a different question.

That distinction matters more than a single headline percentage. OpenTable's 2026 UK trends report, based on a 3 to 9 September 2025 survey of 2,005 UK respondents, found that 36% planned to use AI tools more for restaurant discovery and booking. It measured stated intentions, not completed AI-origin reservations. A 2026 research preprint measured how often venues appeared in AI recommendations in two Bali markets. The preprint was funded and conducted by a company selling hospitality AI-visibility tools; its results are bounded, have not been peer reviewed, and measured no bookings. Neither source replaces your own reservation records.

Older independent research in The Economic Journal linked displayed Yelp review ratings to restaurant reservation availability in its studied US market. It measured a booking-related outcome rather than counting online mentions. Its 2012 findings do not estimate the effect of AI recommendations or predict a UK venue's current bookings.

Five measurements, five different claims

On a phone, swipe sideways to read every column.

Use the same dates, venue and location for every row. A group-wide visibility chart beside a single branch's booking total creates an attractive but invalid story. If the question was about a late table near Soho, the answer that names your Manchester branch is a factual error, not a success.

Start by checking the answer, not just a brand mention

Choose a small, stable set of questions a real guest could ask without naming the restaurant: a weekday pre-theatre meal, a family lunch with a wheelchair user, or a table for six near a station. Use only occasions the venue actually serves. Record the exact wording, city, language, service, date and answer. Check the venue name, branch, opening status, menu, access information and booking destination. A correct recommendation with an obsolete booking link can be worse than a missing mention.

This is an audit of observed answers, not a claim that one answer is what every guest sees. AI results vary by service and context. The bounded venue-recommendation study found limited agreement across the systems it tested and documented stale closed-venue recommendations. Use repeated, comparable checks and keep corrections to public facts separate from changes in the recommendation count.

The National Restaurant Association's 2026 travel-and-dining research also describes several discovery channels and the continuing role of word of mouth. An AI answer is one path into a decision, not the entire market.

Track arrivals without calling every visit an AI referral

In GA4, the AI Assistant default channel groups arrivals from assistants such as ChatGPT, Gemini and Copilot. Google's definition explicitly excludes AI Overviews and AI Mode. A missing AI Assistant referral therefore does not prove that a guest never saw an AI answer. Equally, a referral tells you that a visit happened; it does not prove who booked.

Read the traffic acquisition report at the session scope, using the same date window as your answer checks. Keep sessions, engaged sessions and key events in separate columns. Check landing pages by venue before pooling them. Use Search Console for the Google Search questions and pages that earned impressions or clicks, but remember that its performance report is about search exposure and site traffic, not dining covers. If the property or booking system is inaccessible, write unknown instead of zero.

If you export analytics data, be careful with scope. Google's BigQuery attribution guide distinguishes first-user, session and event traffic-source fields. Mixing them can give one booking three different-looking sources. Choose a model and label it; keep the underlying event timeline available for review.

Instrument the booking handoff, then check the booking system

For an external booking provider, a website button can generate an outbound-click event. This establishes that someone clicked towards a provider. It does not say the provider had availability or that the guest completed the reservation. If the booking domain is included in cross-domain measurement, Google notes that the same click may not appear as an ordinary outbound event. Check the configuration before concluding that a button is unused.

Mark a genuinely useful action as a GA4 key event, and document its meaning. booking_button_click should stay a click; reservation_confirmed should only fire on a provider confirmation you can reconcile. Google notes that marking a key event does not recreate historic events, so do not compare a newly configured event with an earlier period as if the instrumentation had been identical.

The reservation platform remains the source for confirmed bookings, covers, changes and cancellations. Ask which system holds the accepted booking ID if the group uses several marketplaces. Restaurant Dive's 2026 reporting describes operators working across OpenTable, Resy and SevenRooms and the operational argument over a single table-management record. The precise setup varies by restaurant; do not assume one marketplace sees every table.

Where a booking platform and point-of-sale provider integrate, compare their records only after checking what each syncs. Nation's Restaurant News reported 2026 Square/OpenTable and Toast/Resy integrations intended to give operators a fuller guest and spend view. That is a reason to test your own data joins, not evidence that your current stack has that coverage.

Build one honest weekly view

For each location and guest question, keep an audit row with answer outcome and link accuracy. Beside it, show website arrivals by clearly defined channel, menu and booking-button actions, and booking-system confirmations. Add the tracking coverage and date window. A useful row might read: “Named correctly in 3 of 10 checks; 12 attributable visits; 2 booking handoffs; provider confirmation unknown.” It must not read “two AI bookings.” Those numbers are an illustrative format, not a report about any real venue.

Use booking codes, tagged links or provider referrals where they actually exist, and be candid about dark paths: a guest may copy the name into another app, arrive from Maps, call, or book later on a different device. A brief “How did you hear about us?” question can add context, but recall is imperfect. When attribution is ambiguous, report a range or an evidence grade instead of forcing a single channel to claim the covers.

Avoid treating a pageview increase as causal proof. The restaurant's own seasonality, local press, paid activity and events may move at the same time. The OpenTable UK trends report documents changing demand for spontaneous and experiential dining; your baseline should record those occasions and the dates on which they ran.

What should the team do next?

Choose one venue, one guest occasion and one month. Verify its public facts, run the same small question set each week, and test the actual booking path on mobile. Ask the reservations team which provider is authoritative. Instrument the website-to-provider handoff only after that is clear. At the end of the month, compare answers, useful site actions and confirmed bookings without rolling them into a single “AI revenue” figure. If wrong facts recur, fix the venue page and listings first. If people arrive but drop at the booking step, investigate the offer, availability and user journey before producing more articles.

For a view of the questions your restaurant or group is missing, start with a free AI-search audit. Bring the venue list, current booking routes and the questions your host team hears most. The audit can establish the discovery baseline; booking attribution still needs your own provider records.

Sources

1. OpenTable, 2026 UK Restaurant Dining Trends and methodology.
2. Google Analytics, default channel group and AI Assistant definition.
3. Google Analytics, traffic acquisition report.
4. Google Analytics, outbound-click measurement.
5. Google Analytics, cross-domain measurement.
6. Google Analytics, marking key events.
7. Google Search Central, Search Console performance.
8. Google Analytics Developers, attribution data in BigQuery.
9. National Restaurant Association, 2026 summer travel and dining research.
10. Pitenin, 2026 commercially funded bounded venue-recommendation preprint.
11. Nation's Restaurant News, reservation and POS integrations.
12. Restaurant Dive, operators' concerns about a primary table-management record.
13. Anderson and Magruder, 2012 peer-reviewed study of Yelp ratings and reservation availability.