AIRBNB AI PLANNER · CONCEPT 2026
The future of AI-powered trip planning
© Airbnb AI Planner
ROLE
Product Designer
TIMELINE
6 Weeks
TEAM
Inkyung Ryu
SKILLS
Product Strategy
AI Prototyping
Cursor
Claude
PROBLEM
Modern travelers are paralyzed by choice
Modern travelers spend an average of 10-15 hours researching accommodations, often paralyzed by Airbnb's vast inventory and the "paradox of choice." The challenge was to transform this fragmented search experience into a seamless, intelligent flow that anticipates user needs rather than just reacting to filters.
Conversational Search
Mapped how travelers naturally describe a trip in conversation, then designed the assistant to translate that into structured search criteria.
Field Research
Watched real people plan real trips, the tab-juggling, the re-filtering, the mental math, to see where the friction actually lived.
Cross-functional Collaboration
Partnered with engineering to validate that proposed AI interactions were technically feasible within real LLM latency and cost constraints.
Opportunity
An AI layer that plans, not just filters
Designed and prototyped an AI-driven orchestration layer that curates personalized itineraries in real-time. Moved beyond static design by using an AI-assisted development workflow (Cursor, Claude) to build a high-fidelity functional prototype. This allowed testing of complex LLM interactions and dynamic UI components that a standard static prototype couldn't capture.
CORE FLOWS
From open-ended search to a guided plan
The core experience walks a traveler from a broad, conversational starting point to a fully-formed itinerary, surfacing listings, dates, and logistics as the AI narrows in on what fits, rather than asking the user to filter everything by hand.
One Prompt to Start
Travelers describe their trip in plain language and the assistant starts building a plan immediately, no forms or filters required.
A Plan, Not Just Listings
Surfaces a full day-by-day itinerary with flights, stays, and activities already assembled, instead of a list of listings to sort through.
Book Directly in Chat
Reserve flights, stays, and activities from within the same conversation, with a clear review step before anything is confirmed.
RESEARCH
Understanding the paradox of choice
Early research centered on how travelers actually search today, jumping between tabs, saving listings, and re-filtering the same criteria multiple times. That fragmented, repetitive pattern became the throughline for the rest of the project.
Start with one open-ended prompt
Travelers start with a single free-form prompt, no filters or forms, and the assistant immediately begins generating a plan.
The assistant proposes a complete itinerary
It surfaces a complete plan (flights, stays, and a day-by-day itinerary) with pricing and keeps the conversation open to swap in different activities or hotels.
See the full day-by-day plan and book
Each day is broken out with timed stops, stays, and activities, all rolled into one price with a single Reserve button to confirm the trip.
research & Iteration
Testing Real Prompts, Not Just Mockups
Because a conversational entry point can't be evaluated as a static screen, I tested four distinct prompt patterns with real travelers to see which one actually reduced hesitation and got people to a usable itinerary fastest.
Strategic directions
Three Bets on How AI Guidance Should Work
The directions that shaped the final planning experience.
During a working session mapping the traveler journey, one direction we seriously considered was letting the assistant book automatically once it hit a high-confidence match, removing a step for the traveler entirely. Early testing made it clear people didn't want an AI making irreversible purchases on their behalf, even a good one. We kept a human confirmation step instead, trading a little convenience for something travelers said mattered more: staying in control of their own money and plans.
Show the Reasoning, Not Just the Result
Surface why the assistant chose these flights and stays, budget fit, dates, stated preferences, inline, so travelers trust the plan enough to act on it instead of re-checking it elsewhere.
One Conversation, Not a Form in Disguise
Keep every follow-up, swap a hotel, shift dates, add a stop, inside the same thread, so refining a trip never feels like starting over in a new tool.
A Human Checkpoint Before Anything Real Happens
No booking completes without an explicit review-and-confirm step, keeping the traveler as the final decision-maker even as the assistant does the heavy lifting.
Design decisions
We prioritized a simple, familiar, and clean interface.
The interface deliberately borrows patterns travelers already trust, a chat thread for input, familiar card layouts for flights and stays. So the one genuinely new thing (an AI planning your whole trip) doesn't also require learning a new way to interact with a screen.
reflections
Key Takeaways
Prompts Are a Design Material
How the assistant phrased its very first response changed how much travelers trusted everything that followed, wording turned out to matter as much as layout.
You Can't Spec an AI Conversation
Static mocks couldn't capture how the assistant should handle edge cases or bad inputs; only a working prototype (built with Cursor and Claude) surfaced the moments that actually needed design decisions.
Trust Is Built by Staying Visible
Travelers responded better to a slower assistant that explained itself than a faster one that didn't, a reminder that in AI products, disappearing into the background isn't always the goal.
U-Haul Truck Reservation
U-Haul Homepage