Practical AI (day 2): Travel Planning via Deep Research

Using AI to solve real-world problems, today.

Most of us use AI assistants in “standard mode” - getting those broad, surface-level answers. It's like chatting with someone who visited Mexico City once and remembers just the tourist spots. But these same tools can operate in "Deep Research mode" - delivering specific, actionable insights that actually help you make decisions.

In this post, I'll show you how to level up your AI game using travel planning as our example. We'll explore three ways to interact with AI - from basic questions to advanced prompting tricks - and see how the right approach can transform generic travel suggestions into a personalized adventure that fits exactly what you're looking for.

Let's see how to make AI work harder for you, starting with a simple trip to Mexico City.

Level 1: Standard Mode

Let's start with the simplest approach - just asking an AI assistant what to do in Mexico City. It's like asking that friend who visited once but doesn't remember all the details.

I'm planning a trip to Mexico City. What should I do there?

ChatGPT quickly served up a classic tourist highlight reel (see the full response here):

Nice! Mexico City is a vibrant, massive, and deeply layered city—history, food, art, chaos, calm, all rolled into one. Here’s a mix of must-do highlights, local gems, and things to plan ahead for:

---- 
🎨 Culture & Museums
• Museo Frida Kahlo (Casa Azul): Her home in Coyoacán. Gorgeous, emotional, book ahead.
• Museo Nacional de Antropología: One of the best museums in the world, especially for pre-Columbian history.
• Museo Tamayo or Museo Jumex: Modern and contemporary art with stylish vibes.
• Palacio de Bellas Artes: Stunning architecture and murals inside.
----

The recommendations are solid, but they're basically what you'd find in any travel guide from the last decade. There's nothing wrong with these suggestions - they're popular for a reason.

But I'm left with so many questions: How crowded will these places be? Which neighborhood should I stay in? Is Teotihuacan better in the morning or afternoon? What if I'm vegetarian? How do I get from the airport to my hotel?

Level 2: Using Deep Research Mode

Time to kick things up a notch. When I ask the same Mexico City question but with Deep Research mode enabled, I don't get an immediate answer - I get questions back!

Instead of dumping generic info, the AI wants to understand what I'm actually looking for:

That sounds like a great trip! Could you share a bit more about what kind of experiences you're looking for in Mexico City? For example:
1. Are you more interested in history, art, food, nightlife, or nature?
2. What's the length of your stay?
3. Will you be traveling solo, as a couple, with friends, or family?
4. Do you have any budget constraints or accessibility needs?
5. Are you looking for popular highlights, hidden gems, or a mix of both?

This will help me tailor the recommendations better for your trip.

This is already way more helpful! The AI is thinking like a real travel agent, not just a search engine. To be honest, I hadn't even considered specifying these details in my original question.

After I filled in these gaps (a four-day trip with my partner, focused on food and culture, mid-range budget), the AI went into research mode for several minutes. It's like watching someone frantically flip through guidebooks, check recent reviews, and consult local experts all at once.

When it came back, I didn't get a generic list - I got a full-blown custom itinerary that feels like it was crafted by a Mexico City insider. Here's just a taste of what it suggested for our first evening:

• Dinner at Azul Histórico – For your first night, dine in the romantic open-air courtyard of Azul Histórico (Isabel La Católica 30, Centro Histórico, Mexico City 06000) . This upscale yet authentic restaurant is set in a 17th-century palace (now Downtown Hotel), under a canopy of lights and trees. Chef Ricardo Muñoz Zurita, a renowned authority on Mexican gastronomy, serves elegant traditional dishes here – his Oaxacan mole negro and cochinita pibil are legendary . Handmade tortillas are pressed to order, and the atmosphere feels like a tranquil oasis in the bustling center . Booking tip: Reserve in advance for dinner, and request a courtyard table for the best ambiance.

• Nightcap with a View – Cityzen Rooftop Bar – After dinner, head to Cityzen Rooftop Bar on the 38th floor of the Sofitel Mexico City (Av. Paseo de la Reforma 297, Cuauhtémoc, 06500 Mexico City) . This stylish rooftop lounge offers panoramic views of the city skyline – you can often see the Angel of Independence and skyscrapers twinkling at night. Sip a craft mezcal cocktail or champagne as you take in the skyline from the outdoor terrace. (Dress to impress; upscale rooftop bars in CDMX enforce smart casual attire.) Booking tip: Arrive on the early side or reserve a table, especially on weekends, to secure a prime spot by the glass railing.

The entire itinerary goes on like this - each recommendation comes with specific dishes to try, insider tips about the best times to visit, current prices, and little details that only locals would know. It even warns me about which tourist traps to avoid and suggests alternatives.

This doesn't feel like generic advice anymore - it feels like I'm getting tips from a friend who lived in Mexico City for years and knows exactly what I'd enjoy. That's the magic of Deep Research mode. See the full response here if you want to take advantage of this plan. (Also please send pictures of the trip!)

Level 3: AI Assisted “Deep Research” Prompts

Deep Research is fantastic for getting those local insider details, but sometimes it can be information overload. For our Mexico City trip, we need to work around meetings and catch up with friends, so a full minute-by-minute itinerary isn't necessary. What we really need is a curated list of great spots we can flexibly visit.

This is where my favorite zero-effort AI hack comes in: getting the AI to craft its own prompt! Instead of struggling to write the perfect instructions, I'll let the AI do the heavy lifting.

Can you help me craft a Deep Research prompt to answer this question?
'I’m planning a trip to Mexico City. What should I do there?'

After a quick back-and-forth about our specific needs (flexible schedule, focus on food, mix of tourist spots and hidden gems), the AI delivered this masterpiece of a prompt:

You are a Mexico City travel strategist specializing in immersive food-and-history experiences. Your task is to identify, evaluate and rank the top 10 must-do experiences for a 4-day trip to Mexico City, mixing iconic highlights with local hidden gems.

Deliverable:
- Summary (150–200 words): Key themes and how they fit together.
- Top 10 List: For each entry include:
    • Name & category (e.g. Historic Site, Street-food Experience, Cocktail Bar)
    • Why it's unique (authenticity, wow-factor, historical significance)
    • Practicals (neighborhood, hours, cost, booking tips)
- Hidden-Gems Sidebar: At least 3 off-the-radar spots (with same detail fields).
- Cocktail Bar Roundup: Your top 5, highlighting signature drinks and vibe.
- Sources: Cite 5 authoritative, post-2023 travel guides, local blogs or news sites (with URLs and publish dates).

Research Steps (chain-of-thought):
1. Map out travel categories: cuisine, history, art/architecture, nightlife.
2. For each, compile 15–20 candidates from recent guides, blogs, forum posts.
3. Apply evaluation criteria: authenticity, accessibility, cost vs. value, "wow" factor.
4. Rank and distill to your Top 10 and Hidden Gems.
5. Cross-check practical info (hours, reservations, pricing).

Look at how much better this is than my original basic question! A few specifics that make this prompt so effective:

• Role assignment: "You are a Mexico City travel strategist..." immediately puts the AI in the right mindset. We don't want generic travel tips - we want specialized knowledge.

• Clear deliverable format: Instead of getting a wall of text, we've specified exactly how we want the information organized - a summary, a top 10 list with specific details for each item, and bonus sections for hidden gems and cocktail bars.

• Research methodology: The prompt even tells the AI how to think through the problem - casting a wide net first, then filtering down using specific criteria.

When I fed this prompt back to the AI, the results were exactly what we needed - a perfectly formatted list of experiences, each with practical details we could use to make quick decisions on the go. The hidden gems section revealed places I'd never have found in standard travel guides, and the cocktail bar roundup will be required reading before we start each night!

This approach gave us the flexibility to pick and choose activities based on our schedule, while still benefiting from Deep Research quality. It's like having a travel agent create a custom guidebook just for you, without the overwhelming detail of a rigid itinerary.

The best part? This technique works for literally any complex question you might ask an AI. Want restaurant recommendations in your city? Investment advice? Home renovation ideas? Getting the AI to craft its own research prompt first will dramatically improve the quality of what you get back.

Broad and Recent Context

Using Deep Research pulls much wider, current context for the model to use in forming its opinion. Just as a traveler might flip through a dozen guidebooks before a trip, Deep Research allows AI models to gather and organize a broad array of relevant information.

When you ask about Mexico City restaurants without Deep Research, the AI gives you what it broadly knows about popular dining spots. But with Deep Research, it can consider dozens of factors simultaneously: recent restaurant openings, temporary closures, seasonal menu changes, which places are currently trending with locals versus tourists, and how dining scenes have evolved in different neighborhoods.

This contextual expansion means recommendations aren't just common knowledge but are tailored to your specific needs at a specific point in time. It's the difference between a friend telling you “Pujol is a good restaurant” versus a local guide who knows that "Pujol requires reservations three months in advance, but if you're looking for innovative Mexican cuisine and can't get in, the head chef's casual taqueria Eno offers a similar culinary philosophy at a fraction of the price and with no reservation required."

Remember, the way these models work at the end of the day is by predicting the next token in a given sequence. They produce recommendations based on the patterns they've observed in their training data. The AI suggests visiting Frida Kahlo's Casa Azul because people collectively discuss it online as a must-visit attraction. The model has absorbed the general sentiment that this is an important cultural site.

However, the specific details that make a recommendation truly useful—like current ticket prices, reservation systems, accessibility features, or recent renovations—are buried within that generalized knowledge, mixed together with countless other opinions from across the internet. Without Deep Research, the AI defaults to the most commonly mentioned aspects without necessarily surfacing the most relevant details for your specific needs.

When to Use Each Approach

This doesn't mean you should always use Deep Research. Sometimes the quick answer is exactly what you need:

• Standard mode is perfect for general knowledge questions ("What's the currency in Mexico?"), creative requests ("Write a poem about Mexico City"), or when you're just beginning your exploration of a topic.

• Deep research mode shines when you need specific, current information for decision-making ("Which neighborhoods should I focus on for my 3-day trip next month?"), when accuracy and recency matter ("What are the current entry requirements for US citizens?"), or when you need comprehensive comparisons ("How do the different transportation options from the airport compare in terms of safety, cost, and travel time?").

That's the real magic of Deep Research: transforming an AI assistant from a generalist with broad knowledge into a specialist with the specific insights you need for your journey—whether that's a literal trip to Mexico City or any other complex decision you're facing.

What use cases have you found Deep Research most helpful for?