Can AI really plan a better holiday?

Planning a holiday through artificial intelligence sounds wonderfully efficient. Give it a destination, dates and budget, then wait a few seconds for flights, hotels, restaurants and attractions to fall neatly into place.

To see how well that promise holds up, ChatGPT, Gemini, Copilot and Perplexity received exactly the same assignment. Each tool had to plan a five-night holiday to Tbilisi for two adults travelling from Dubai from October 15 to 20, 2026. The total Dh8,000 budget had to cover direct return flights with checked baggage, a centrally located four-star hotel, airport transfers, meals, local transport and attractions.

They were also asked to provide a timed itinerary, individual prices, booking links, attraction opening days and journey times between stops.

Every tool returned a detailed holiday. Each plan looked useful at first glance. Closer inspection exposed unconfirmed prices, contradictory budgets, incorrect dates and itineraries that depended on several optimistic assumptions.

ChatGPT played it safe

ChatGPT proposed a total spend of Dh7,130, leaving a useful Dh870 for fare changes and incidental costs. It allocated Dh3,300 for two flydubai tickets with 20kg baggage, Dh1,500 for five nights at The Gate Hotel City Centre, Dh1,350 for meals and Dh630 for attractions.

Its strongest feature was the structure. The itinerary grouped Old Tbilisi, Narikala, the Botanical Garden and the sulfur baths sensibly. It scheduled the Simon Janashia Museum on Sunday because major Georgian National Museum venues commonly close on Mondays. Monday was reserved for the open-air Chronicle of Georgia, Dezerter Bazaar and Fabrika.

The plan also included realistic rest periods and warned travellers to check seasonal opening hours. Its advice on booking flydubai’s Value fare instead of adding baggage to a Lite ticket showed some awareness of the costs that can change a cheap fare.

The apparent precision came from “target rates”, however. The tool did not confirm that its chosen flydubai flights were available for Dh3,300 on the specified dates. Its Dh1,500 hotel estimate was also a target, with no dated room result confirming the rate or taxes. The booking references generally led to a hotel, airline or comparison site instead of the exact option described. ChatGPT therefore produced the most usable framework. The final bill remained theoretical.

Gemini planned the busiest holiday

Gemini selected the Mercure Tbilisi Old Town and estimated the complete trip at Dh7,937. It included two day excursions, one to Kazbegi and Gudauri and another to Kakheti and Sighnaghi. The schedule also covered Narikala, Sameba Cathedral, the sulfur baths, Mtatsminda Park and several well-known restaurants.

Its hotel choice was clear, and its geographical information benefited from access to Google services. Google says Gemini can draw on public information from Google Flights, Google Hotels and Google Maps, which gives it a useful base for comparing routes and locations.

The budget left only Dh63. That amount could disappear with one airport coffee, a baggage difference or a small movement in the exchange rate.

Several choices made the food allocation of Dh1,650 look especially fragile. Gemini recommended Barbarestan, Funicular Restaurant and Ethno Tsiskvili alongside daily lunches, coffees and wine. It also budgeted Dh540 for two full-day excursions for two travellers. Optional costs such as the 4x4 journey to Gergeti Trinity Church could push the figure higher.

The schedule contained smaller warning signs. It allowed two and a half hours for a Kazbegi return journey described in the same answer as taking approximately three hours. It assumed early hotel access soon after noon and added welcome drinks without confirming either. Its flight times appeared exact, yet no flight numbers or dated fare links were supplied.

Gemini created the most ambitious holiday. It also created the itinerary most vulnerable to one delayed journey or underestimated meal.

Copilot exceeded the budget

Copilot gave the most confident opening figure, estimating the holiday at around Dh7,600 with a Dh400 buffer. Its later budget table told a different story.

The listed costs came to Dh8,050: Dh3,000 for flights, Dh2,250 for the Mercure, Dh100 for airport transfers, Dh600 for local transport, Dh1,350 for food and Dh750 for attractions. Copilot acknowledged the excess and suggested saving money through one cheaper dinner or fewer taxi rides.

That solution failed the original assignment, which required the plan itself to stay within Dh8,000. The claimed Dh7,600 total also did not match the itemised figures. Removing Dh450 after presenting the budget as complete placed the work back on the traveller.

Copilot did provide a specific flydubai option, quoting FZ721 outbound and FZ712 on the return journey at approximately Dh2,740 for two with checked baggage. It then raised the flight allowance to Dh3,000 for fluctuations and seat selection. This was more useful than a general flight estimate, though the response still required the traveller to confirm the fare and baggage conditions on the airline’s website.

Perplexity lost track of the calendar

Perplexity offered the most heavily sourced answer. It cited fare sites, hotel platforms, transport guides and local travel pages throughout the plan. It also flagged a detail missed by the other tools: Narikala Fortress walls were reportedly under restoration, though the ridge, statue and viewpoints remained accessible.

Then it made a basic calendar error. It called October 15, 2026, a Wednesday and October 20 a Monday. They fall on Thursday and Tuesday. Every day in its itinerary was labelled with the wrong weekday.

This matters because the plan relied on museum closure days. Perplexity correctly stated that major museums close on Mondays, yet its incorrect calendar could cause a traveller to schedule a museum visit on the wrong day.

Its budget also shifted as the response developed. The initial flight estimate was Dh3,300 to Dh3,600. The detailed calculation increased it to Dh3,960 with baggage. The hotel section said typical rooms would cost Dh1,650 to Dh2,020, then budgeted only Dh1,300 in the final total. Meals calculated at Dh1,755 were rounded down to Dh1,600.

The finished estimate came to Dh7,872, leaving Dh128. That total depended on finding a four-star hotel at a price lower than the tool’s own quoted range.

Perplexity was the easiest answer to investigate because it displayed numerous sources. Several links led to third-party travel blogs or general booking homepages, and some hotel links did not clearly match the named property. A large number of citations created the appearance of verification without resolving the contradictions inside the plan.

What we feel

ChatGPT produced the best overall itinerary and left the healthiest financial margin. Gemini made the strongest use of location data and offered the most exciting programme. Perplexity was best for generating research leads. Copilot supplied the clearest flight option and then lost control of its own arithmetic. None completed the assignment well enough to book the entire holiday without further checks.

The experiment showed where AI genuinely helps. It can build a logical route, suggest neighbourhoods, identify closure days and revise a holiday around personal interests within minutes. It is less dependable when several live prices must remain valid at the same time. On this Tbilisi test, the best holiday would combine ChatGPT’s structure, Gemini’s mapping and Perplexity’s sources, followed by a human with a calculator and every booking page open.