AI‑Generated Tbilisi Trip: How Accurate Are the Itineraries?

A test of four AI tools—ChatGPT, Gemini, Copilot and Perplexity—revealed that while each can draft a detailed Tbilisi trip, none delivers a fully bookable plan. ChatGPT offers the best structure and budget cushion, Gemini excels at mapping and activity variety, Copilot gives clear flight details bu…

By Felo News Desk · Published Sat, 19 Sep 2026 01:26:21 GMT

When you think of artificial intelligence, the image that often comes to mind is a quick, hassle‑free solution to everyday problems. Planning a holiday is no different. The idea that you could simply feed a destination, dates and budget into a chat box and receive a ready‑to‑book itinerary is tantalising. To test how close reality is to that promise, a group of AI tools—ChatGPT, Gemini, Copilot and Perplexity—were given the same brief: design a five‑night holiday in Tbilisi for two adults travelling from Dubai on 15‑20 October 2026, with a total budget of Dh8,000. The brief required direct return flights with checked baggage, a centrally located four‑star hotel, airport transfers, meals, local transport and attractions, plus a timed day‑by‑day schedule, individual prices, booking links, opening days and journey times. The results were mixed, and the experiment offers a clear picture of where AI shines and where it still falls short.

ChatGPT: The Most Practical Framework

ChatGPT’s response was the most balanced overall. It estimated a total spend of Dh7,130, leaving a buffer of Dh870 for fare changes or incidental costs. The flight cost was set at Dh3,300 for two flydubai tickets with 20 kg baggage, the hotel at Dh1,500 for five nights at The Gate Hotel City Centre, meals at Dh1,350 and attractions at Dh630. The itinerary was logically grouped: Old Tbilisi, Narikala Fortress, the Botanical Garden and the sulfur baths on day one; the Simon Janashia Museum on Sunday (to avoid Monday closures); the open‑air Chronicle of Georgia, Dezerter Bazaar and Fabrika on Monday; and a rest day on Tuesday. The plan also advised travellers to verify seasonal opening hours and to book flydubai’s Value fare rather than a cheaper Lite ticket that would require extra baggage fees. However, ChatGPT’s figures were based on target rates, not confirmed bookings. The flight and hotel estimates were not cross‑checked against live prices, and the booking links led to generic comparison sites rather than the exact options described. While the structure was solid, the final bill remained theoretical.

Gemini: Ambitious but Fragile

Gemini produced a more adventurous itinerary, adding two day‑excursions—one to Kazbegi and Gudauri, another to Kakheti and Sighnaghi—alongside standard Tbilisi sights such as Narikala, Sameba Cathedral and Mtatsminda Park. Its hotel choice, the Mercure Tbilisi Old Town, was clear and its use of Google Flights, Hotels and Maps gave it a robust geographical base. The budget, however, was razor‑thin: Dh7,937 with only Dh63 left for contingencies. The food allocation of Dh1,650 was especially vulnerable to small price swings, and the schedule contained timing inconsistencies—e.g., it allotted two and a half hours for a Kazbegi return trip that it also described as taking three hours. Flight times were listed without numbers or links, and optional costs such as a 4×4 trip to Gergeti Trinity Church could push the total over budget.

Copilot: Clear Flights, Confusing Arithmetic

Copilot’s strength was a specific flight suggestion: flydubai FZ721 outbound and FZ712 return, priced at roughly Dh2,740 for two with checked baggage. It raised the flight allowance to Dh3,000 to account for seat selection and price fluctuations. The itinerary included the Mercure hotel, airport transfers, local transport, food and attractions, with a total of Dh8,050. Copilot acknowledged the excess and suggested cutting a dinner or taxi rides, but that adjustment violated the original brief’s requirement to stay within Dh8,000. The claimed Dh7,600 total did not match the itemised figures, and the budget table was inconsistent.

Perplexity: Many Sources, One Calendar Error

Perplexity’s answer was the most heavily sourced, citing fare sites, hotel platforms, transport guides and local travel pages. It correctly noted that Narikala Fortress walls were under restoration, but it mislabelled the dates: it called 15 October 2026 a Wednesday and 20 October a Monday, when they are actually Thursday and Tuesday. This error could lead travellers to book a museum visit on a day it is closed. Its budget fluctuated throughout the answer: flights were initially quoted at Dh3,300, later rising to Dh3,960 with baggage; hotel costs ranged from Dh1,650 to Dh2,020, but the final estimate used Dh1,300; meals were rounded from Dh1,755 to Dh1,600. The final total of Dh7,872 left Dh128, but only if a four‑star hotel could be found at the lower end of the quoted range.

What the Experiment Reveals

None of the AI tools produced a fully bookable itinerary without further human verification. Each tool’s strengths and weaknesses are clear:

ChatGPT: Best overall structure and budget cushion, but relies on target rates.
Gemini: Strong mapping and activity variety, but budget is fragile.
Copilot: Provides a concrete flight option, but arithmetic is inconsistent.
Perplexity: Offers many sources and research leads, but contains a critical calendar mistake.

The common theme is that AI can generate logical routes, suggest neighbourhoods, identify closure days and tailor a holiday around personal interests in minutes. The real challenge lies in synchronising live prices across multiple services—flights, hotels, transport and attractions—so that the final plan remains accurate and bookable.

For travellers looking to use AI as a starting point, the best approach is to combine ChatGPT’s structured itinerary, Gemini’s mapping data and Perplexity’s source list, then double‑check every link and price on the official booking sites. A human touch—especially a calculator and a few open tabs—remains essential to turn the AI draft into a confirmed trip.

Why This Matters for Travelers

In an age where planning a holiday can be as simple as chatting with a bot, understanding the limits of AI is crucial. Knowing which tools provide reliable data and which require extra verification can save time, money and frustration for the modern traveler.

Key facts

ChatGPT offers the best overall structure but uses target rates.
Gemini’s itinerary is ambitious but budget‑tight and timing‑inconsistent.
Copilot gives a concrete flight option but mis‑calculates totals.
Perplexity cites many sources yet mislabels key dates.
None of the tools produce a fully bookable plan without human checks.
Combining the strengths of each AI with manual verification yields the most reliable itinerary.

Why it matters

AI can draft holiday itineraries in seconds, but without real‑time price checks it risks delivering plans that are out of date or over budget. Travelers need to know how to use AI responsibly to avoid costly surprises.

Frequently asked questions

Can I book a trip directly through an AI tool?

No, the tools provide estimates and links but do not confirm bookings. You must verify prices and availability on the official airline, hotel or attraction websites.