We Tested ChatGPT, Gemini and Perplexity for Trip Planning — Here's the Honest Verdict

Everyone in the travel industry is talking about AI. Most of what they're saying is either breathless hype or defensive dismissal. Here's something more useful: an honest test of how the three main AI tools — ChatGPT (GPT-4o), Google Gemini 1.5, and Perplexity — actually perform on real travel planning tasks.

We gave each tool the same brief: plan a 7-night trip to Thailand for a couple from Bangalore in May 2026, budget ₹1,20,000 total, mix of beach and city, no group tours.

What AI does well

All three tools produced a reasonable skeleton itinerary within seconds. Bangkok for 2 nights, then down to Koh Samui or Phuket for 5 nights — the logical structure is correct. Gemini and Perplexity both pulled live flight data and gave broadly accurate price ranges for May (₹18,000–26,000 return from BLR). ChatGPT, without browsing enabled, quoted prices from its training data that were 15–20% lower than current fares.

For general destination research — what to eat in Chiang Mai, how to get from Bangkok to Koh Samui, what the Thai visa-on-arrival process looks like — all three were excellent. Accurate, detailed, well-organised. This is the task AI does best: synthesising widely available reference information into a readable summary.

Perplexity stood out for up-to-date news. It correctly flagged that the Southern Thailand monsoon season runs May–October (so Phuket and Samui get hit), and recommended Koh Lanta or Krabi instead for a May trip — which is genuinely good advice that many people miss.

Where AI fails badly

Hotel recommendations were the biggest weakness. Every tool recommended well-known properties — the kind that rank highly in training data. When we asked for boutique hotels on Koh Lanta under ₹6,000/night, two of the three suggested places that no longer exist or have changed ownership. ChatGPT hallucinated a hotel with a glowing description that we simply cannot find in any booking system.

Flight routing was worse than expected. Gemini recommended a routing that doesn't exist as a single booking — it combined two separate tickets (IndiGo BLR to Bangkok, then a Thai domestic carrier with no interline agreement) without flagging the risk. If the first flight is delayed, you miss the connection with no protection. A travel agent would never do this.

Visa nuance was mostly handled correctly, but with important gaps. None of the tools mentioned that Indian passport holders on a Thailand visa-on-arrival need to show proof of onward travel — a requirement that catches people at Suvarnabhumi every week.

And critically: none of them can actually book anything. They can tell you prices but cannot access live inventory, cannot hold seats, and cannot respond when things go wrong.

The honest verdict

AI is now a genuinely useful research assistant for travel. Use Perplexity to get oriented on a new destination — what to see, roughly what it costs, what the weather is doing. Use ChatGPT to brainstorm itinerary structures or compare two destinations. These are real time-savers.

But AI is not a travel agent. It cannot negotiate rates, cannot access GDS inventory, cannot spot the routing trap that ruins your connection, and cannot call the hotel at midnight when your room is wrong. The travellers who will do best in the AI era are the ones who use these tools for research and then hand the actual booking to someone who can be held accountable.

That's still us. And we're not worried.