AI things to do recommendations combine traveler intent with trusted experience data, time windows, location, travel relationships, and business rules so a customer can discover an activity that actually fits the trip. The language model interprets a request such as a three-hour indoor window before dinner and compares only valid options. The experience catalog remains authoritative for schedules, ticketing, duration, eligibility, and availability. Routing services calculate travel relationships, and Spatial AI explains the result.
The sections below separate nearby-place search from experience discovery, then cover source ownership, time-window feasibility, ranking, Kaleidr mapping, B2B products, destination measurement, and a narrow pilot. Related reading includes How to Build an AI Tourism Map, AI Guest Concierge for Hotels, and Traffic-Aware Journey Planning. Teams already choosing an implementation shape can skip to the Kaleidr mapping; teams still naming the data boundary should start with source ownership.
Things-to-do essentials
- Approved catalog first: Recommend experiences the business actually offers or endorses.
- Time is eligibility: Duration, start time, and the next commitment filter candidates before ranking.
- Travel is a relationship: Hotel to experience and experience to dinner matter more than a radius.
- Spatial AI second: The language model interprets intent and explains grounded options.
- Measure the decision: Selection, itinerary add, and host handoff beat chat length alone.
Things-to-do recommendations become more useful when the system evaluates both the experience and whether it fits the customer's trip.
Why Are AI Things to Do Recommendations a B2B Spatial AI Problem?
Hotels, destination platforms, tourism marketplaces, attraction networks, convention products, and city guides already hold first-party context such as a booked property, an approved member catalog, a ticket inventory, or an editorial list. Plotting attractions as markers is no longer the scarce capability. The product problem is helping a customer choose something to do that fits the remaining time, the next commitment, and the host's rules, without asking a language model to invent the activity.
Kaleidr currently lists Things to Do as an AI-driven customer journey that finds activities, experiences, and places to explore based on what customers want (AI-Powered Map Experiences for Business). The AI Map Chat for Customer Discovery page currently describes travel as turning intent into itineraries, routes, and destination recommendations. Those pages are authoritative about Kaleidr's own positioning. The same pages are not evidence that Kaleidr operates an experience catalog, a ticketing system, or a destination-wide inventory feed.
Destination organizations are already attaching conversational trip planning to trusted local content. Visit Orlando launched an AI-powered trip planner in June 2026 that combines local-expert curation with broader travel data (Visit Orlando, 2026). The San Diego Tourism Authority later integrated conversational trip planning into its official visitor site, with recommendations and itineraries on SanDiego.org (Mindtrip, 2026). Those launches are evidence that destination discovery is becoming a conversation over curated supply. The same launches are not Kaleidr case studies, and they do not prove that every host must replace its catalog or booking stack.
How Do Things-to-Do Recommendations Differ From Nearby Places Search?
A nearby-places map answers which venues sit around a pin. A things-to-do system answers which experience fits what the customer wants to do in the remaining trip. The distinction matters because a museum can be geographically close and still be closed, sold out, too long for the window, too far from dinner, or outside the approved catalog. A walking tour can sit on the right block and start two hours too late. A concert can match the traveler's interests and still conflict with a reservation.
The useful unit is therefore the experience plus its place plus its time, not the place alone. Nearby search can still supply coordinates, categories, and public context. The recommendation job starts after those facts exist: the product must decide whether the customer can reach the experience, complete it, and still make the next commitment. AI Restaurant Search and Table Booking covers a similar eligibility problem for dining. Things-to-do discovery generalizes that contract to museums, tours, performances, classes, and events.
Location Intelligence Customer Experience covers the same Discover → Compare → Act shape for customer-facing location products. Discover retrieves eligible experiences. Compare makes duration, travel time, and availability inspectable. Act is add-to-itinerary, directions, a ticket handoff, or a booking start. Ranking a closed or impossible option because the description scored well inverts that order.
Which Systems Should Own Experience Facts?
Catalog, schedule, ticketing, routing, and itinerary systems should remain authoritative for the facts a recommendation depends on. The language model can turn a request such as something indoor, nearby, and complete before a 7:00 PM dinner into inspectable fields: origin, next commitment, time window, indoor constraint, travel limit, and experience type. Those fields are queries against the catalog and routing owners, not invented values. The structure in any example is illustrative. The important contract is that vague language becomes state the customer can correct without restarting the conversation.
Hard constraints are binary and belong with the catalog, schedule, and routing owners before ranking. Closed today, sold out, duration that overruns dinner, a travel budget the routing service cannot meet, or a missing accessibility attribute should remove the candidate. Soft preferences such as neighborhood, format, or family fit then rank the remaining valid set. An experience that misses dinner should not win because it is more popular.
| Customer question | Authoritative source |
|---|---|
| Which experiences may this product recommend? | Host catalog, member list, or partner inventory |
| Is the experience open at the requested time? | Schedule or hours system |
| Can the customer still get a ticket? | Ticketing or availability system |
| How long does the visit take? | Experience duration field |
| How long is the trip from the hotel? | Routing service |
| Where is “back” or “dinner”? | Itinerary, reservation, or property context |
| May this user see this offer? | Host identity, tenant, and permissions |
Exact hours still belong in a place or experience system. Google's current Places API documents currentOpeningHours as the next seven days of operation, including special hours, and regularOpeningHours as the typical schedule (Google, 2026). That page is evidence of one production place provider's hours contract. The same page is not evidence that every Kaleidr deployment uses Google Places, and it does not describe Kaleidr inventory.
The language model interprets and explains the request; business and geospatial systems remain authoritative for the facts that make an experience usable.
Why Is Time a First-Class Spatial Constraint?
A recommendation is useful only when the customer can reach it, complete it, and still make the next commitment. Straight-line nearness is not that test. Two exhibitions can sit at similar distances from the hotel while one adds twelve minutes of travel and seventy-five minutes on site, and the other consumes the entire window. The product should evaluate origin to experience, experience duration, buffer, and experience to the next destination as one feasibility object.
Arrive-by and free-time windows are different intents. “I have three hours before dinner” is a remaining budget. “Get me to the venue by 7” is an arrival deadline on a later commitment. The language model should preserve which intent the customer named. The routing service should calculate the travel legs. The catalog should supply duration and start time. Do not ask the model to invent those minutes after the customer has already stated the constraint.
The following comparison is illustrative, not a measured Kaleidr or destination result. Use it only to show why options need the same columns. Real products should fill those columns from the current catalog, schedule, and routing responses.
| Candidate | Start | Travel in | Duration | Travel to dinner | Fits the window |
|---|---|---|---|---|---|
| Experience A | 4:10 PM | 8 min | 165 min | 12 min | No: too long |
| Experience B | 5:45 PM | 10 min | 75 min | 15 min | No: starts too late |
| Experience C | 4:15 PM | 12 min | 75 min | 15 min | Yes |
A generic “best nearby” score can hide those tradeoffs. If one option wins, name the reasons: indoor, available, completes before dinner, and stays inside the stated travel limit. Do not manufacture a quality score the catalog does not supply. Traffic-Aware Journey Planning covers the travel-leg side of the same hotel-to-dinner object when the customer then needs a route.
A recommendation is useful only when the customer can reach it, complete it, and still make the next commitment.
How Should Eligibility, Ranking, and Explanation Stay Separate?
Eligibility is a filter. Ranking is an order over the survivors. Explanation is a grounded account of why the shortlist exists. Mixing those jobs produces the familiar failure: the assistant recommends a famous attraction that the customer cannot actually do. Popularity is a weak proxy for the current trip. A less famous indoor exhibition next to the hotel can be the better answer when the headline attraction is a forty-five-minute trip plus a three-hour visit.
Diversity belongs after fit. A shortlist of five nearly identical walking tours is less useful than a smaller set that still satisfies the window and then varies neighborhood or format. Do not sacrifice a hard customer constraint merely to produce variety. Ranking should operate only over an eligible set; the things-to-do layer supplies the time and catalog constraints that ranking must not override.
Shared map state keeps conversation, cards, and itinerary on one canonical object. Selecting an experience should highlight the place, show the travel relationship, and preserve the hotel and dinner anchors. Asking “what is shorter?” should keep the same window. Asking “what about outdoors?” should re-run eligibility rather than invent a new trip. A second, invisible assistant-only list breaks that contract. The map is a view. The itinerary object is structured data, and the product should not reconstruct it from whatever happens to be visible in the viewport.
How Does Kaleidr Map Onto Things-to-Do Discovery?
A Kaleidr implementation can attach a conversational spatial layer to a map and experience stack the host already operates. Kaleidr currently documents Chat as a product that mounts over a map the host already renders, plots resolved places, and frames the camera as the conversation resolves locations (Chat attach). The attach contract confirms map-aware conversation exists in the current public developer surface. The same docs do not promise a native experience catalog, ticket inventory, or destination-wide hours feed.
Those catalogs and booking systems should remain explicit deployment dependencies. Kaleidr can provide the conversational spatial layer and map-aware coordination while the deployment uses the appropriate authoritative experience, schedule, and routing sources. Do not imply that Kaleidr itself is the attraction operator or the ticket ledger unless a specific integration is documented for the deployment.
The hotel, a dinner reservation, or a selected map point should be the default origin for things-to-do search, because those anchors already sit on the itinerary. A publishable key is for browser SDK use; server credentials belong in the application layer. Kaleidr currently documents that split and states that a publishable key presented as a bearer is rejected (Auth & scopes). Device location is a separate permission: the current W3C Geolocation Candidate Recommendation Snapshot requires express permission from an end user before any location data is shared with a web application (W3C, 2026). Device location helps when the customer asks to start from the current position, and it should not be required when the hotel or dinner already names a better origin. Private Location Data for AI Map Workflows covers authorization for movement data the host does not expose publicly.
Kaleidr currently describes a hospitality template with curated destinations and tap-to-ask place summaries; the live starter is Kaleidr Hospitality. Confirm current plan allowances on Pricing & Plans before depending on a specific production workflow. Treat the current developer documentation as the integration contract; marketing pages describe the use case, not the experience-feed list.
Which B2B Products Need a Things-to-Do Layer?
A destination marketing organization can connect member businesses, attractions, tours, and editorial picks to a visitor who has one free afternoon and does not want the major tourist site. The DMO keeps control over the candidate universe. The assistant interprets the preference. The catalog, hours, and routing systems still decide what is possible. How to Build an AI Tourism Map covers the destination-map architecture around that catalog.
A hotel can combine the active property, approved partners, a guest time window, and a dinner reservation. “What can I do for two hours before dinner?” is then a property-anchored feasibility question rather than an open web search. AI Guest Concierge for Hotels covers the property-side conversation; the things-to-do layer adds duration and the next commitment. An attraction network can ask which of its own sites still fits before 4:00 PM. A convention platform can fill the gap between a 3:00 PM session and a 6:30 PM dinner without turning the product into indoor venue wayfinding. A travel marketplace can narrow a large bookable catalog to trip-compatible options while remaining authoritative for price, availability, and cancellation.
Keep the AI-to-map contract narrow: show eligible experiences, focus one place, show the travel relationship, add to itinerary, open directions, or clear the shortlist. The host validates the action. Do not let the model emit arbitrary map code. Experience selection should stay user-controlled. A conversational system can recommend one option. The customer should still be able to pick another experience, another window, or another origin.
How Should Destination Discovery Stay Measurable?
Map pans and chat opens are diagnostics. Outcome metrics include query starts, eligible results returned, experience selection, itinerary add, directions opened, and ticket or booking handoff. Quality metrics include no-result rate, stale-hours rate, duration-missing rate, and travel-calculation failure. Business metrics depend on the host: member referral, ticket conversion, hotel engagement, or destination-content use. Preserve a structured no-result reason such as no approved experiences, schedule conflict, sold out, too far, duration does not fit, or unavailable accessibility data, rather than a bare failure flag.
Visit Orlando reported more than 14,800 itineraries created and more than 57,000 chat interactions after a soft launch earlier in 2026, and stated that conversation insights were informing destination content and digital channels (Visit Orlando, 2026). Those figures are Visit Orlando's own operating report. The same figures are not Kaleidr Analytics numbers, and they are not a benchmark every destination should copy. The product lesson is the loop: traveler questions can become content, partnership, and coverage signals when the host logs intent and outcome rather than chat volume alone.
Map Engagement and Location Analytics currently documents map and place engagement, spatial patterns, and activity that product, inventory, and growth teams can act on. Host systems still own ticketing and booking. A travel or destination platform can use that model to ask which neighborhoods receive activity searches but few selections, which categories are requested but undersupplied, and which trip times generate no-results. Spatial Analytics vs. Web Analytics covers why those questions are geographic rather than page-view questions. The same measurement discipline applies to other map products: task completion over raw interaction volume.
Things-to-do discovery can become a growth loop when customer search and place engagement inform destination content, inventory, and partnership strategy.
How Should a B2B Pilot Start?
Start with one high-value task, such as recommending something to do near the hotel before a fixed dinner time. Keep the experience catalog, hours, and ticketing in the systems that already own them. Attach conversational map interaction to the existing map. Limit candidates to approved experiences, require duration and operating hours, calculate travel from the hotel and onward to dinner, and measure selection plus the host action that follows. Expand categories and cities only when that first window works.
Conversational discovery does not replace catalog quality, hours freshness, or fulfillment discipline. Travel times remain estimates. Availability claims are only as good as the inventory behind them. Attaching an assistant to an existing map is usually cheaper than replacing the renderer, but the host still has to own authorization, supplier contracts, and the next business action.
Explore Kaleidr Spatial AI to add conversational experience search on an existing map. Explore Kaleidr Analytics to measure place engagement and geographic demand around that journey. Confirm the current public pages before treating any example in this article as a shipping contract.
FAQs
What are AI things to do recommendations?
AI things to do recommendations combine traveler intent, an approved experience catalog, schedules, duration, location, travel time, and business rules so a product can suggest activities that fit the current trip. Spatial AI interprets and explains the request; catalog and booking systems remain authoritative for the facts.
How is this different from nearby places search?
Nearby search lists places around a pin. Things-to-do discovery asks whether a specific experience is open, available, reachable, and completable before the next commitment.
Should the language model invent attractions?
No. The host catalog, member list, or partner inventory should remain the candidate universe. The assistant can interpret constraints and explain grounded options.
Why is time a spatial constraint?
An experience that the customer cannot reach, finish, and still make dinner is not a valid recommendation, even when the place is geographically close.
Is this the same as an AI tourism map?
An AI tourism map is primarily a destination-map architecture. Things-to-do recommendations focus on one customer decision: choosing an eligible experience that fits the current trip context.
Can a hotel use this with a guest concierge?
Yes. The concierge keeps the property as the anchor and the approved-partner policy. The things-to-do layer adds duration and the next commitment, such as dinner or checkout.
Does Kaleidr replace ticketing?
No. Current public Kaleidr pages describe conversational map discovery and analytics. Ticket inventory, price, and purchase remain in the host or supplier systems unless a specific integration is documented.
What should a product measure?
Measure eligible results, experience selection, itinerary adds, directions, ticket or booking handoff, and structured no-result reasons. Chat length is a diagnostic, not the outcome.
Do recommendations require a permanent travel history?
No. A temporary origin, itinerary anchors, and explicit preferences are often enough. Persist precise movement patterns only when the product needs them and has permission.
References
- Kaleidr. AI-Powered Map Experiences for Business. Accessed 9 September 2026. https://kaleidr.com/
- Kaleidr. AI Map Chat for Customer Discovery. Accessed 9 September 2026. https://kaleidr.com/ai
- Visit Orlando. Visit Orlando Expands Free Vacation Planning Services with New AI Trip Planner. 25 June 2026. https://www.visitorlando.org/media/press-releases/post/visit-orlando-expands-free-vacation-planning-services-with-new-ai-trip-planner/
- Mindtrip. Mindtrip Partners With The San Diego Tourism Authority To Reimagine How Travelers Discover San Diego. 28 August 2026. https://www.prnewswire.com/news-releases/mindtrip-partners-with-the-san-diego-tourism-authority-to-reimagine-how-travelers-discover-san-diego-302862145.html
- Google. REST Resource: places. Places API. Accessed 9 September 2026. https://developers.google.com/maps/documentation/places/web-service/reference/rest/v1/places
- Kaleidr. Chat attach. Developer documentation. Accessed 9 September 2026. https://docs.kaleidr.com/sdk/chat-attach
- Kaleidr. Auth & scopes. Developer documentation. Accessed 9 September 2026. https://docs.kaleidr.com/platform-api/auth-and-scopes
- W3C. Geolocation. W3C Candidate Recommendation Snapshot, 26 March 2026. https://www.w3.org/TR/2026/CR-geolocation-20260326/
- Kaleidr. Kaleidr Hospitality. Template. Accessed 9 September 2026. https://template.kaleidr.com/customize/?template=hospitality
- Visit Orlando. How Visit Orlando Is Using AI to Power Smarter Trip Planning. 26 June 2026. https://www.visitorlando.org/about/corporate-blog/post/how-visit-orlando-is-using-ai-to-power-smarter-trip-planning/
- Kaleidr. Map Engagement and Location Analytics. Accessed 9 September 2026. https://kaleidr.com/analytics
@misc{kaleidr_home_things_to_do_2026_09_09,
title = {AI-Powered Map Experiences for Business},
author = {{Kaleidr}},
note = {Accessed 9 September 2026},
url = {https://kaleidr.com/}
}
@misc{kaleidr_ai_things_to_do_2026_09_09,
title = {AI Map Chat for Customer Discovery},
author = {{Kaleidr}},
note = {Accessed 9 September 2026},
url = {https://kaleidr.com/ai}
}
@misc{visit_orlando_ai_planner_2026_09_09,
title = {Visit Orlando Expands Free Vacation Planning Services with New AI Trip Planner},
author = {{Visit Orlando}},
year = {2026},
month = jun,
url = {https://www.visitorlando.org/media/press-releases/post/visit-orlando-expands-free-vacation-planning-services-with-new-ai-trip-planner/}
}
@misc{mindtrip_sdta_2026_09_09,
title = {Mindtrip Partners With The San Diego Tourism Authority To Reimagine How Travelers Discover San Diego},
author = {{Mindtrip}},
year = {2026},
month = aug,
url = {https://www.prnewswire.com/news-releases/mindtrip-partners-with-the-san-diego-tourism-authority-to-reimagine-how-travelers-discover-san-diego-302862145.html}
}
@misc{google_places_resource_2026_09_09,
title = {REST Resource: places},
author = {{Google}},
year = {2026},
note = {Places API; accessed 9 September 2026},
url = {https://developers.google.com/maps/documentation/places/web-service/reference/rest/v1/places}
}
@misc{kaleidr_chat_attach_things_to_do_2026_09_09,
title = {Chat attach},
author = {{Kaleidr}},
note = {Developer documentation; accessed 9 September 2026},
url = {https://docs.kaleidr.com/sdk/chat-attach}
}
@misc{kaleidr_auth_scopes_things_to_do_2026_09_09,
title = {Auth \& scopes},
author = {{Kaleidr}},
note = {Developer documentation; accessed 9 September 2026},
url = {https://docs.kaleidr.com/platform-api/auth-and-scopes}
}
@misc{w3c_geolocation_cr_2026_09_09,
title = {Geolocation},
author = {{W3C}},
year = {2026},
month = mar,
note = {W3C Candidate Recommendation Snapshot, 26 March 2026},
url = {https://www.w3.org/TR/2026/CR-geolocation-20260326/}
}
@misc{kaleidr_hospitality_template_2026_09_09,
title = {Kaleidr Hospitality},
author = {{Kaleidr}},
note = {Template; accessed 9 September 2026},
url = {https://template.kaleidr.com/customize/?template=hospitality}
}
@misc{visit_orlando_ai_results_2026_09_09,
title = {How Visit Orlando Is Using AI to Power Smarter Trip Planning},
author = {{Visit Orlando}},
year = {2026},
month = jun,
url = {https://www.visitorlando.org/about/corporate-blog/post/how-visit-orlando-is-using-ai-to-power-smarter-trip-planning/}
}
@misc{kaleidr_analytics_things_to_do_2026_09_09,
title = {Map Engagement and Location Analytics},
author = {{Kaleidr}},
note = {Accessed 9 September 2026},
url = {https://kaleidr.com/analytics}
}