The Future of Travel Planning: How AI is Revolutionizing Your Itinerary

The joys and pains of travel planning can be overwhelming when you’re attempting it on your own. The excitement of knowing you have the opportunity to plan a trip is interwoven with the many questions and details that you have to determine and navigate—how do you decide on where you should go in the first place? And once you’re there, which restaurants should you go to–and which of those do you need to get reservations for ahead of time? Are there even flights available on your dates?

What if you could have an intelligent assistant that organizes all of these details seamlessly for you? It’s a dream that many companies have been trying to make a reality.

Surprisingly, even the most advanced AI models today struggle with these types of tasks. A recent benchmark study found that LLMs (large language models) like Google’s Gemini Pro and OpenAI’s GPT-4 perform poorly in travel planning, with success rates of only 1.1% and 0.6%, respectively. These models often fail to make sequential, dependent decisions—such as ensuring a flight exists before recommending a destination—leading to incorrect, incomplete, or infeasible itineraries.

This is where Mobi’s AI comes in. As part of our latest research to support our Intent Driven Search product, we built a prototype system—not a product (yet)—to explore how AI can more intelligently and reliably plan multi-step travel itineraries. Mobi’s AI, powered by Context Variable Graphs (CVG) and topological sorting, outperforms existing models significantly, achieving an 84.4% overall performance and a 32.8% success rate—over 30 times higher than other LLMs.

Gemini Pro and GPT-4 statistics come from https://arxiv.org/pdf/2402.01622

Let’s explore how this works and why it could revolutionize travel planning.

How does Mobi’s AI plan a trip?

Imagine that you tell an AI assistant:

"Can you help craft a week-long travel plan for two people? We’ll be departing from Oakland and want to visit a city in Oregon from March 12th to March 18th. We prefer Italian and Mediterranean cuisine."

Instead of manually searching for flights, restaurants, and attractions, Mobi’s AI breaks your request down into a series of decisions:

✅ Finding cities in Oregon

✅ Checking flight availability from Oakland

✅ Looking for Italian and Mediterranean restaurants

✅ Identifying great attractions and accommodations

By systematically solving these steps, AI can quickly create a personalized and efficient travel itinerary. But, how can AI come up with these steps and make reasonable decisions?

🌐 Representing travel planning problem as a graph

To efficiently solve travel planning queries, our AI represents the problem using a graph-based structure called a Context Variable Graph (CVG). This method helps an AI system systematically break down the query into known inputs and unknown variables that must be determined.

In the graph representation (Fig. 1), blue nodes represent extracted information directly from the user’s query. These are facts explicitly stated, such as:

• Origin City: Oakland
• Destination State: Oregon

On the other hand, red nodes represent variables that the AI must determine to complete the itinerary. These include:

• Destination City? – The AI must find a suitable city in Oregon.
• Flight? – Checking flight availability between Oakland and the chosen city.
• Hotel? – Recommending accommodations based on location and availability.
• Restaurant? – Finding Italian or Mediterranean restaurants.
• Attraction? – Suggesting tourist spots based on user preferences.

With this graph, the AI knows which variables need to be determined. But how does AI decide the correct decision order? Let’s dive into topological sorting and how it ensures the AI solves variables in the right order.

🧩 Think of it like solving a puzzle in order

Imagine you're assembling a jigsaw puzzle. You wouldn’t randomly grab pieces and hope they fit—you’d start by finding the corner pieces, then build the edges, and finally fill in the center.

Mobi’s AI solves your travel plan in a similar way.

It figures out which decisions need to be made first before solving the rest. This is where topological sorting comes in—it’s like organizing the pieces of your trip in the right order before putting them together.

Instead of randomly picking locations and then checking if flights exist, AI follows a logical order:

1️⃣ Step 1: Find a city in Oregon

The AI queries the state-city database for possible destinations. It picks Portland as an option.

2️⃣ Step 2: Check if flights are available

AI looks for flight information from Oakland to Portland in the flight database. If there’s no available flight, it goes back to Step 1 and picks a different city, like Eugene.

3️⃣ Step 3: Find accommodations

Once a city is confirmed, AI searches for hotels in that city.

4️⃣ Step 4: Find restaurants that match your preference

• Since you asked for Italian and Mediterranean food, AI looks for restaurants in the chosen city.

5️⃣ Step 5: Find attractions

• Finally, AI searches for things to do in that city.

💾 Why do we need to query the database in each step?

Large language models (LLMs) like GPT-4 or Gemini often struggle with factual accuracy when handling structured decision-making tasks like travel planning. One major issue is hallucination—these models might make up flights that don’t exist, recommend fully booked hotels, or suggest restaurants that have long since closed. This happens because, unlike search engines that retrieve real-time, verified data, LLMs generate text based on "pattern matching." In simple terms, pattern matching is like assembling sentences: the model predicts the most likely next words based on its vast training data, rather than checking whether the information is actually true. As a result, it sometimes produces information that sounds plausible but isn’t actually correct.

To ensure accuracy, Mobi maintains a real-world verified dataset with more than 5 million hotels, 8 million attractions, and 4 million restaurants.  Our AI system cross-references this dataset at every step of the planning process, ensuring that recommendations are based on real, up-to-date information. This means that when our AI suggests a flight, hotel, or restaurant, it's pulling directly from verified sources.

The future of AI in travel

With AI advancing rapidly, we’re getting closer to fully automated travel planning. Imagine simply stating:

"Plan me a two-week trip to Japan with a mix of city life and nature experiences. Budget: $3000."

And receiving a perfect itinerary within minutes. The intent-driven approach isn’t just about convenience—it’s about making travel smarter, more enjoyable, and perfectly tailored to you.

So how close is this to being a reality? While still in a research prototype phase, this level of AI-powered travel planning is rapidly advancing. ​For example, this vision is becoming increasingly attainable with advancements like Mobi's Intent Driven Search (IDS), which leverages AI to interpret natural language requests and deliver personalized hotel recommendations.

Stay tuned to see how Mobi continues to lead the way in AI-powered travel planning, bringing the future of seamless, personalized travel closer every day.