I let ChatGPT plan my holiday. Here's why I wouldn't do it again

ChatGPT is out there. It can't be bargained with, it can't be reasoned with, it doesn't feel pity or remorse or fear, and it absolutely will not stop recommending I eat octopus. I hate octopus, and ChatGPT knows I hate octopus. Yet, for two hours, it has been telling me to eat octopus at various restaurants in Lisbon, Portugal.

How did this happen? Like everyone, I became obsessed with the game-changing artificial intelligence tool several months ago as it came crashing like an avalanche into the mainstream.

In simple terms, ChatGPT is a computer program that can have conversations with people. It understands what you say and aims to give you helpful answers, sort of like chatting with an emotionless but hyperintelligent friend.

Hailed as the internet's future, to some, it's a benevolent human-made creation that possesses the power to revolutionize our lives and work. To others, it's the catalyst of our ultimate demise. Let's not dwell on that part, though.

For some time, I’ve had the words "AI plans my holiday" jotted in my notes, but until recently, no freely available AI program could fulfill this dream. Then along came ChatGPT, and this futuristic travel fantasy suddenly appeared within reach.

My plan was simple: I'd use ChatGPT to craft a two-day city break itinerary, letting it select everything from the destination and hotel to restaurants, food and sightseeing. I would surrender myself entirely to the machine.

To do this, I gave ChatGPT several parameters: the ability to fly from London, a maximum flight duration of five hours, and no significant time zone changes.

I also told it that I loved "good food, arts and culture, magic, cool bars and great weather." Except for "magic," these requests seemed reasonable.

Mulling over my request, the bot suggested I visit Barcelona, citing its short flight times and abundant cultural offerings as the perfect fit for a sunny two-day getaway. It wasn't the most creative suggestion, but I couldn't fault the logic.

I could have stopped here, but I wanted to ensure ChatGPT wasn't just randomly throwing darts at an internet map. I opened a second chat and asked again. This time it declared Lisbon as the ideal destination.

Given its previous enthusiasm for Barcelona, this sudden shift made me apprehensive. If ChatGPT could change its mind so easily, how could I trust that its recommendations were anything more than arbitrary whacks at Google?

Opening a third chat, I asked again. The bot once again came back with Lisbon. Proceeding cautiously, I asked which of the two destinations — Barcelona or Lisbon— it believed I would enjoy more.

"Based on the criteria you provided earlier … I would say Lisbon might be the better choice for you," came the reply, "Plus, the city has a certain charm and magic that is hard to find elsewhere."

I wasn't convinced the AI had used genuine reasoning, but I didn't want to argue. Lisbon it would be.

Next was assembling accommodation options and the wider itinerary.

After some prompting, the bot provided recommendations for budget and points hotels. This seemed promising. However, on closer inspection, it was obvious little consideration had gone into the selection.

The location of the hotels ChatGPT recommended would mean I'd have to do a lot of walking, use public transport or take frequent Uber rides to reach the center of town, something I was keen to avoid.

The points hotels it recommended also missed several notable properties, including the Hyatt Regency Lisbon. ChatGPT acknowledged the hotel's existence but didn't recognize it as a points hotel.

Flight schedules also posed a challenge. I wanted two days in Lisbon, arriving early on the first day and departing late the second evening. This wasn't feasible with no midweek return flights departing later than midday without a stopover, which would turn the return journey home into a 10-hour-plus marathon and give me much less time to enjoy the city.

I opted to extend the trip to three nights. If I arrived in the afternoon of the first day to check in, I’d get two full days in Lisbon before an early flight back to London on the final day. To manage costs, I’d use points and stay the first night at the Hyatt Regency Lisbon. The following evenings would be at the ChatGPT-approved Lisbon Marriott Hotel. Apart from the first night, everything else would follow the bot recommendations.

Next up: the itinerary. The bot provided a list of well-reviewed breakfast and lunch venues, sightseeing spots, bars and even a magic show. I was impressed, but it was a lot to tackle, and I needed structure. So, I challenged ChatGPT to provide specific times for each item.

The machine pondered the request and responded with time slots. However, now, every item on the itinerary was different.

The AI had changed its mind again.

I pointed this out, and a new itinerary was generated, which deviated even more. Each time I asked the AI to revert to the previous version, it drifted further from the original brief.

Its suggestions seemed to be designed to please me rather than be realistic. For instance, it vaguely recommended seeing "a magic show at Casino Lisboa," but there was no indication that there was actually a show. Other suggestions, such as a visit to Time Out Market, were impossible as the market was closed on the specified days.

There were also odd logistical choices. One iteration said to have breakfast at Pastéis de Belém on the first day, and on the second day visit the historical Belém Tower. These locations were a mere 15-minute walk apart, but they were spread across two days instead of grouped in one.

When I asked for food suggestions, ChatGPT repeatedly recommended I eat octopus (a food I dislike) at various restaurants. Despite repeatedly telling it not to, ChatGPT kept suggesting octopus. At one point, after berating the bot, its very next suggestion was a glib: "Try the octopus, trust us, you’ll love it!"

ChatGPT's mischievous habit of following its own whims rather than my instructions made me feel like it was toying with me. After hours of failing to get a firm, logical plan in place, I decided to call it a day and just go with whatever itinerary the bot conjured up.

Like a petulant child that won't do as it's told, the bot had beaten me down.

Arriving in Lisbon midafternoon, I checked in to the Hyatt Regency Lisbon, a stone's throw from ChatGPT's Belém recommendations for the following day. This would be the only evening I had in the Portuguese capital that the AI hadn't plotted.

I grabbed dinner and, after explaining what I was doing, showed my itinerary to the restaurant staff. They seemed marginally impressed but not enthusiastic about the list and simply said: "Looks good … quite touristy. Wear comfortable shoes."

Pushing further, I asked if they’d recommend the things on the list. "It depends on what you like," was the reply. It turns out humans can be just as vague as ChatGPT. Feeling nervous, I headed back to the hotel for a full night's sleep.

The next day, my first stop was breakfast at the famed cafe and bakery Pastéis de Belém, often heralded as the birthplace of the iconic Portuguese pastry pastel de nata, a kind of sweet egg custard tart dusted with cinnamon.

Leaving the Hyatt Regency Lisbon just after 8:30 a.m., ChatGPT suggested I take an Uber or tram to reach Pastéis de Belém at 9 a.m. However, Google Maps showed I was only a short walk away, and with clear blue skies and the sun shining, a leisurely stroll among Belém's colorful buildings seemed much more fun, so I set off on foot.

Along the route, I took in the quintessentially colorful European architecture of Belém, with occasional views of the Tagus River and the picturesque Afonso de Albuquerque Square with its towering neo-Manueline monument.

Edging closer to my destination, the stark streets began to bustle more and more until I finally saw it: Pastéis de Belém. And oh boy, was it busy. Tourists and locals alike were crammed at its entrance, taking shade beneath its blue awnings and waiting to get inside.

I knew instantly I’d be queuing an age, and this pastel de nata jaunt would seriously eat into my morning, but I was relieved. The previous night I was worried ChatGPT could send me to places that didn't exist or weren't worth visiting. The crowds, inconvenient as they were, at least confirmed the recommendation was probably a good one.

Thankfully, the queue moved quickly, and I was soon inside, surrounded by glass cabinets filled with bottles of port (a little too early, thank you very much) and various baked goods that instantly made me salivate. ChatGPT wanted me to try the cafe's "iconic pastel de nata" with a cup of coffee. Which it purred would be a "delightful breakfast experience."

I placed my coffee order, and a warm pastel de nata was thrust into my hand. Waiting for my coffee, I took in the atmosphere of Pastéis de Belém. I was surrounded by beautiful food and smells, but the vibe was utter chaos. The staff was angsty to the point of rudeness, and the clientele was pushy and touristy. The entire cafe seemed to strain under the weight of its own popularity.

After waiting 20 minutes for my cappuccino, I retreated from the bustle to the Vasco da Gama Garden across the street to enjoy what a robot had described as one of the best pastel de natas in Lisbon and sip my ludicrously creamy coffee.

The pastry was still warm despite the long coffee wait, and as I took my first bite, I acknowledged that ChatGPT had done well. It was creamy, sweet and delicious. I ate it quickly and immediately wanted to order another. However, with the clock ticking, it didn't seem wise to queue again.

My second stop was the 16th-century UNESCO World Heritage Site Belém Tower, a former defensive structure against potential attacks from the Tagus River it overlooks.

Conveniently (thanks ChatGPT), the tower was only a 15-to-20-minute walk away. This time the walk was less scenic, taking me away from the pretty Belém streets and forcing me to cross an ugly urban bridge across a busy highway.

The small parkway leading to Belém Tower was already full of tourists when I arrived at around 10 a.m. The ticket kiosk informed me that, on this day, access was free due to International Museum Day, but would usually cost 9 euros ($9.62). I wasn't sure if this was a fluke or part of ChatGPT's meticulous planning, but when I asked the AI, it was none the wiser.

A limited number of visitors can enter (or leave) the tower due to its slim stairwells. I waited for quite a while before reaching the first level and its grand panoramic views of the Tagus River. With the beautiful weather and a pretty stream of boats plying by, it was a perfect picture spot. Had it not been for the crowds, at least.

The queue to go higher was growing fast, so I idly joined it and waited again for around 15 minutes. The next level was a bare limestone room with a balcony offering more views and more crowds. After this, I queued several more times to ascend more sets of stairs, each leading me to a similar-looking chamber.

Belém Tower is a striking testament to the rich history of the area. However, the long queue times, claustrophobic stairwells and sheer number of visitors made me happy to leave. Stunning views aside, the tower is just as pretty from the outside as inside (with fewer lines). I’d have felt shortchanged had I paid the entry fee.

Now that it was lunchtime, I was positively starving. ChatGPT recommended a restaurant in the Príncipe Real neighborhood, an annoying 1 1/2-hour walk away or 40 minutes using public transport. Verging on hangry, I decided to take a 20-minute Uber ride, and ChatGPT concurred.

While waiting for the cab, I checked out the nearby Monument to the Discoveries, which I’d seen from the tower's turrets. This impressive structure, ChatGPT explained, was erected in 1960 to commemorate the Age of Discovery and the maritime heritage of Portugal. The level of detail in its carvings of historical figures like Henry the Navigator and Vasco da Gama is outstanding. It seemed odd that ChatGPT hadn't told me it was nearby. Saddened to leave this impressive sight, I hopped in my Uber.

My lunch destination was A Cevicheria — a moderately fancy seafood spot that serves ceviche and other seafood dishes. Upon arrival, I was met with more queues, just like at ChatGPT's earlier recommendations. A busy restaurant can be a good sign. However, I began to feel like ChatGPT was simply sending me to Lisbon's busiest places.

After a short wait, I entered the restaurant and was confronted by a giant model octopus dangling threateningly from the ceiling as if it had leaped straight from the pages of a Jules Verne novel. After our earlier conversation, I was sure that ChatGPT was messing with me.

I tentatively asked ChatGPT what to order, conscious that it might reply with "octopus" repeatedly in all caps. Luckily, it behaved and suggested several ceviche dishes instead. Unfortunately, none of them were on the menu. Broadly speaking, though, it was at least in the same ballpark as what was on offer.

The bot also recommended a pisco sour. Asking the waitress for her recommendations, she also suggested a pisco sour, along with the tuna-and-foie gras ceviche. I decided both sounded delicious.

The ceviche was a tastefully presented masterpiece of tuna, foie gras, diced beetroot, raspberry, lychee and hazelnut. Its vibrant colors matched its zesty flavors and perfectly complemented the bot-approved cocktail.

As I ate, I chatted with a Dutch lady beside me called Alex, who visited Lisbon regularly. She said A Cevicheria was her favorite restaurant. I explained that ChatGPT had brought me there and asked what she thought of the AI's itinerary. "It's quite touristy," she replied, "and a little spread out. Your next place is back the other way." I checked Google Maps, and my next stop, LX Factory, was indeed another 20-minute Uber ride back toward Belém.

I couldn't guarantee ChatGPT's recommendations would be fun, but I at least expected them to be efficient. Right then, it seemed my time would be spent mostly in cabs, which eschews the basic appeal of getting lost in an exciting city.

Additionally, everywhere I’d been sent so far was so busy I felt I was being lured into tourist traps at every turn. Sure, the food at A Cevicheria and Pastéis de Belém was delicious, and Belém Tower was charming. But it was hard to shake that my itinerary felt like an internet listicle you'd find after a basic search for "best things to do in Lisbon."

I paid my bill and hopped into another cab to reach LX Factory, which ChatGPT promised was a "creative hub with unique shops and street art." Could this be the off-piste recommendation I craved?

Found in the Alcantara neighborhood beneath a mammoth highway bridge, the LX Factory entrance opens into a series of streets punctuated by small buildings and warehouse spaces filled with achingly cool streetwear and vintage shops, art studios and other boutiques.

It was busy and thriving but with a laid-back and authentic energy. Outside, various bars spilled onto a cobbled street, and Lisbon's cool kids congregated to sip cocktails over cigarettes. It was still touristy, but the hipster stylings gave the sense you'd stumbled into a well-kept secret. Maybe ChatGPT could claw things back.

By now, my iPhone was dying — a pitfall of conversing with an AI and Google Maps all day — so I headed back to the Hyatt Regency Lisbon to charge before heading to check in at the bot-recommended Lisbon Marriott Hotel, which was another 20-minute cab drive away.

My AI guide described the hotel as "contemporary and elegant" with a "modern and sophisticated design" — the type of vague description you’d find on a hotel's website. For the most part, the analysis was fair. The lobby, bar and breakfast area were relatively stylish, but the rooms were quite dated and stuffy. I didn't want to spend any more time there than I had to. It was now late afternoon, and I was itching to get back out.

The evening plan was dinner at an authentic and well-reviewed Portuguese restaurant called Taberna da Rua das Flores, followed by drinks in the Bairro Alto neighborhood. I showered and booked my fourth Uber of the day.

ChatGPT told me Taberna da Rua das Flores was a "charming and cozy restaurant in the heart of Lisbon's historic city center," serving contemporary twists on popular Portuguese dishes. It sounded perfect.

Unfortunately, I’ll never know just how perfect. The eatery only accepts walk-ins, and hopeful patrons were spilled out across the pavement in front of it when I arrived — the running theme of ChatGPT's recommendations. The queue time exceeded an hour. After waiting in line all day, I couldn't face it.

I whipped out my phone and tapped the bot up for an alternative. This plunged me into a frustrating back-and-forth with the AI repeatedly recommending places prohibitively far away or ones that would blow my entire budget for the trip.

After an hour of going in circles, it suggested an option specializing in traditional Portuguese cuisine just a 10-minute walk away: Taberna da Baixa. I marched in its direction and prayed that now, at 8 p.m., it wouldn't be too busy, and it wasn't.

I was seated on a small upper level by a friendly waiter named Daniel and asked ChatGPT to recommend a starter, main and a glass of wine. For the starter, it chose pica-pau — tender marinated beef cubes served with pickles and bread. I held my breath and checked the menu. It was there. ChatGPT got something right.

The next recommendation was bacalhau a bras, a classic Portuguese dish made with shredded codfish, eggs, onions and crispy potatoes. I scanned the menu. Home run. ChatGPT then proceeded to recommend various wines that weren't on the list. Still, two out of three ain't bad.

Daniel came to take my order, so I asked for his recommendations, too. He seconded the pica-pau but pushed me to try a creamy variant of the bacalhau a bras. Finding it hard to say no, I went with it. I was breaking the rules a little, but Daniel was so enthusiastic it felt rude to say no.

I’m humbled to say that the ChatGPT-selected pica-pau was the best part of the meal. Well done, my botty friend, well done.

As I dined, I asked Daniel what he’d do if he were visiting Lisbon. He suggested visiting Belém, where I’d already spent most of my day. Perhaps I’d judged the robot too harshly. His final suggestion was to explore Alfama and grab a coffee while enjoying its beautiful architecture.

The bot had recommended some bars, but they were now quite far away, and I was exhausted. I decided to cut things short and head back to the hotel. The day had been a mixed bag. None of it dreadful, but all of it hectic and full of queues. As I climbed into bed, I decided that tomorrow I would try a different approach.

It was time to change things up.

My problems on Day One, as I saw, were as follows: The itinerary was generic and touristy, everywhere was too busy, and each stop was too far away from the last, meaning I was constantly rushing from point A to point B.

However, ChatGPT eventually succeeded when I forced it to think on the spot, as with the evening restaurant recommendations. So, rather than the AI planning an entire day at once, I’d choose a starting point and request suggestions on the go. That way, I would hopefully avoid numerous Uber trips and tourist traps, and see more of the city.

Following Daniel's lead, I’d start in Alfama for a coffee in the sunshine. From there, we’d take things step by step.

I was Ubered out, so I bought a metro day pass and headed to Alfama by rail. Once there, I asked ChatGPT for a nice coffee spot. "Cafe A Brasileira," came the reply, a historic and traditional Portuguese cafe conveniently located by the train entrance. It will not shock you to discover it was full to the rafters. Drat.

I pushed for alternatives, but each option ChatGPT gave was a considerable trek away. Spotting a small street cafe in the adjacent Praca Luis de Camoes square, I took matters into my own hands, grabbed a coffee and sat beneath the square's statue to watch two flamenco-dancing street performers. I wanted to stick to ChatGPT's suggestions, but only when they didn't derail things.

Drinking my coffee, I asked for things to do nearby, and ChatGPT suggested some markets, local shops and street art. This seemed eminently doable, and I headed for the first place on the list: Rossio Square, a six-minute walk away. Had I finally tamed ChatGPT's wild recommendations?

I stretched out the walk, asking ChatGPT for its souvenir recommendations (tinned sardines, pastels de nata, port … canned octopus) and perused the shops while taking in the local street art. Eventually, I reached Rossio Square and grabbed a selfie in front of its fountain before wandering into its small market. The square was busy, but not like the places I’d experienced the previous day. Sure, there were tourists, but also locals enjoying coffee and drinks in the sun, buying knickknacks and eating giant ham sandwiches.

This inspired me to try some authentic Portuguese street food. The bot suggested I grab a prego — marinated beef served on a roll. Nowhere in Rossio Square sold the item, so ChatGPT directed me to nearby street food markets.

The first suggestion, Figueira Square, was just three minutes away. I was quickly disappointed when it turned out to be a vacant square that didn't even have a market.

ChatGPT had another ace up its sleeve: Martim Moniz Square, a short walk away and "known for its multicultural food scene" and "diverse range of street food." I turned 180 degrees and started walking. Once again, I found an empty square.

Now committed to eating the mythical prego, I asked where I could get the best one nearby. Numerous options came back, all more than an hour away apart from one, Café Império. It was a 40-minute walk away, so still a relative expedition in the afternoon heat. But ChatGPT said it was a "beloved local spot" and prego was its specialty. I decided to take my chances.

My inability to follow directions meant the walk took over an hour and took me away from the prettier side of Lisbon to a much grimier part of town, where the street art changed from colorful and political to just plain angry.

Outside Café Império, I surveyed the menu. There was no prego roll, but my stomach was rumbling, and I was too far away from other options at this point. I grabbed a table inside.

As was now the custom, I asked ChatGPT what to order. It reeled off a bunch of coffee and pastry options. I pointed out that Café Império is not a cafe but actually a huge Portuguese steak restaurant.

"I apologize for the confusion. Café Império is indeed a large restaurant in Lisbon," came the response, suggesting a dozen other dishes not on the menu. Giving in, I asked the staff, and they suggested the specialty, bife à Império da vazia.

The dish turned out to be an overcooked steak slathered in a brown buttery sauce with a fried egg dolloped on top and a side of fries. It was awful. I ate it in silence, furious that ChatGPT had dragged me so far out of town for this.

I paid up, set off back to central and decided to ask for something risk-free: a popular sightseeing spot in the city. I figured after my excursion to the outskirts of town, a tourist spot didn't seem so bad. ChatGPT sent me back to Alfama (at least I liked it there) to the medieval São Jorge Castle.

Knowing it would take a while to get there, I checked with ChatGPT when the castle closed. "9 p.m. during the summer season (April to October) and at 6 p.m. during the winter season (November to March)," it replied. I had plenty of time.

The route was steep and spiraling, taking me through an array of neverending cobbled streets. By the time I arrived just before 5 p.m. I was a sweaty mess.

Obviously, the castle was closed.

I wandered aimlessly through the surrounding streets and decided I’d tried my best but had reached my limit. I begrudgingly asked the bot to recommend a drinking spot with castle views. I could at least then see it from a distance while I drowned my sorrows.

ChatGPT suggested Miradouro da Graça, which it billed as having a "calm and peaceful ambiance" and "an ideal spot for a relaxed drink while enjoying the scenery."

My enthusiasm was entirely zapped, but it's fair to say, as I gazed out for miles across the Lisbon rooftops, that for all its earlier failures, ChatGPT had nailed this suggestion. I didn't want to push for more and decided to end my AI adventures on at least one victory.

After two days of navigating the whims of my friendly but infuriating robot guide, I had nothing left in the tank. The trip hadn't been terrible, but I can't imagine enlisting ChatGPT to plan my travels again.

I genuinely wanted ChatGPT to succeed and to experience a new, effortless way of traveling. Sadly, the intricacies of fulfilling human experience currently elude AI. Yes, it can furnish you with a list of things to do. However, it won't be any better than following the first list you find on Google. My standards might have been high, but I don't think that's bad. The world is here to be explored, not to stand in queues and follow the biggest crowd you see.

As I pondered all of this on the hotel's outside terrace, a man named Diego struck up a conversation. He was a chef and had just finished his final shift for the hotel. I told him my story while he laughed at my various mishaps.

Diego couldn't understand using a robot to plan a trip. "Why?" he asked. "We live because it's life. Listen to the birds sing. That's why you’re here. Follow life. The weather is beautiful. The people are beautiful; that's all that matters." He had the face of a grizzled Jeffrey Dean Morgan and the wisdom of an ancient traveling sage. I was enthralled.

In a few sentences, he’d encapsulated everything I love about travel and the one thing ChatGPT will never get. An understanding of the small things — fleeting human interactions, humor and conversation. The most fulfilling parts of this trip hadn't been AI-led conversations about statues or restaurants; it was meeting Alex in a seafood restaurant, learning about Daniel's favorite Lisbon spots and listening to the life philosophies of Diego.

"Have you eaten dinner yet?" he inquired. I suddenly realized I hadn't, and then Diego said something incredible: "Have you had a prego sandwich while you’re here?"

"No," I told him.

"We call it prego de novilho, and it's the best. I will order it for you in my name," he offered. "The chefs will make sure it's the best they cook today. If you don't love it, I will pay for it."

He stood up and wandered to the bar, shook hands with the bartender, pointed at me and, with a wave, he was gone. I’d come full circle, and I didn't even need ChatGPT's help, just a real human.

The prego was delicious.