Project
AI Can Plan Your Trip.
Here's What It Gets Wrong.
AI travel planning is genuinely useful. It is also missing something that no language model can fix. Here's an honest breakdown of where it works and where it fails.
Let's start with something that might seem counterintuitive coming from a travel project: AI trip planning is genuinely useful. If you need a rough itinerary for a city you've never visited, ChatGPT will give you a serviceable framework in thirty seconds. It covers the obvious things. It won't forget to mention the famous museum or the neighbourhood everyone goes to for dinner.
That's not the problem.
The problem is what happens when you've already seen the obvious things. When you've done Barcelona, done the Costa Brava, and done a wine tour in the Priorat. When you open ChatGPT and type "what else is worth seeing in Catalonia," the answer feels exactly like every other answer you've gotten before. Not wrong, exactly. Just flat.
We've been thinking about this for a while, because it's the whole reason VIANÉDITA exists.
That's not the problem.
The problem is what happens when you've already seen the obvious things. When you've done Barcelona, done the Costa Brava, and done a wine tour in the Priorat. When you open ChatGPT and type "what else is worth seeing in Catalonia," the answer feels exactly like every other answer you've gotten before. Not wrong, exactly. Just flat.
We've been thinking about this for a while, because it's the whole reason VIANÉDITA exists.
AI Optimises for Completeness. You Need Selection.
The fundamental architecture of a language model pushes toward more information, not better information. Ask ChatGPT for places to visit in southern Catalonia, and you will get a list. It will be long. It will include things at the top because they have the most reviews, not because they are the most interesting.
The traveler's actual problem is the opposite. You don't need more options. You need someone to have already made the decisions for you, based on real judgment rather than aggregated data.
This is the difference between a list and a guide. A list is complete. A guide is selective. Those two things are not the same, and confusing them is where most travel planning goes wrong.
The traveler's actual problem is the opposite. You don't need more options. You need someone to have already made the decisions for you, based on real judgment rather than aggregated data.
This is the difference between a list and a guide. A list is complete. A guide is selective. Those two things are not the same, and confusing them is where most travel planning goes wrong.
AI Has Never Been to the Places It Recommends
This is not a criticism. It is just a fact worth sitting with.
Every AI travel recommendation is synthesised from text that other people wrote about places. The model has never stood in a rice paddy at 6 a.m., watching it turn into a mirror as the farmers flood the fields for spring planting. It has never sat across from a third-generation oyster farmer and understood, from the way he talked about the water, why the shellfish here taste different from anywhere else in the region.
What AI knows about a place is the publicly documented version of that place. Which is fine for the famous parts. The famous parts are famous because they've been documented extensively. But the interesting parts of any destination are, almost by definition, the ones that haven't been written up a thousand times.
When we spent three days in the Ebro Delta filming our first documentary, we learned things that don't exist in any database. A brewer in Barcelona ages his oyster stout submerged in the Delta's bay for six months. A family has been milling rice with 19th-century wooden machinery for three generations, producing a grain that isn't sold in supermarkets. Someone a few kilometres away is fermenting sake from locally grown river rice. None of this came from research. It came from showing up.
Every AI travel recommendation is synthesised from text that other people wrote about places. The model has never stood in a rice paddy at 6 a.m., watching it turn into a mirror as the farmers flood the fields for spring planting. It has never sat across from a third-generation oyster farmer and understood, from the way he talked about the water, why the shellfish here taste different from anywhere else in the region.
What AI knows about a place is the publicly documented version of that place. Which is fine for the famous parts. The famous parts are famous because they've been documented extensively. But the interesting parts of any destination are, almost by definition, the ones that haven't been written up a thousand times.
When we spent three days in the Ebro Delta filming our first documentary, we learned things that don't exist in any database. A brewer in Barcelona ages his oyster stout submerged in the Delta's bay for six months. A family has been milling rice with 19th-century wooden machinery for three generations, producing a grain that isn't sold in supermarkets. Someone a few kilometres away is fermenting sake from locally grown river rice. None of this came from research. It came from showing up.
The Timing Problem
AI can tell you that the Ebro Delta has flamingos. It cannot tell you that the campsite gates lock at night, meaning you need to park your car outside the evening before if you want to reach the lagoons before dawn for the best light. It cannot tell you that the kitchen at one of the region's best restaurants closes at 15:30 even though the dining room stays open until 17:00. It cannot tell you that a specific tasting experience is only available on Sunday mornings, or that a private boat tour requires a minimum number of passengers and often doesn't run at all in February.
These details sound small. They are not small. Getting them wrong means missing the thing entirely.
Operational knowledge of a place comes from being there repeatedly, talking to the people who run it, and paying attention to how it actually works rather than how it presents itself online. AI has access to the online version. The operational version lives in the heads of people who've done the fieldwork.
These details sound small. They are not small. Getting them wrong means missing the thing entirely.
Operational knowledge of a place comes from being there repeatedly, talking to the people who run it, and paying attention to how it actually works rather than how it presents itself online. AI has access to the online version. The operational version lives in the heads of people who've done the fieldwork.
AI Produces Consensus, Not Curation
There is a structural reason why AI travel recommendations tend to converge on the same places. The model is trained on the collective output of travel writing, which is itself heavily skewed toward the places that already get covered. Popular places get written about more, which makes them more visible to the model, which makes them more likely to appear in recommendations.
The result is a feedback loop that continuously surfaces the same destinations, the same restaurants, the same itineraries. AI is very good at telling you what everyone else has already done.
If that's what you want, it works perfectly. If you've already done what everyone else has done and you're looking for something past that, the model hits a wall.
This isn't a solvable problem with better AI. It's an inherent property of how recommendation systems work when trained on popularity signals. The things worth finding are, by nature, underrepresented in the data.
The result is a feedback loop that continuously surfaces the same destinations, the same restaurants, the same itineraries. AI is very good at telling you what everyone else has already done.
If that's what you want, it works perfectly. If you've already done what everyone else has done and you're looking for something past that, the model hits a wall.
This isn't a solvable problem with better AI. It's an inherent property of how recommendation systems work when trained on popularity signals. The things worth finding are, by nature, underrepresented in the data.
Where AI Is Actually Useful
In the interest of intellectual honesty, there are things AI does well for trip planning.
It is good at logistics frameworks. Flight connections, visa requirements, rough budget estimates, and general information about transport infrastructure. These are factual, well-documented, and AI handles them efficiently.
It is good for first-pass research for completely unfamiliar destinations. If you know nothing about a place, a ChatGPT overview is a reasonable starting point for orientation before you go deeper.
It is good at answering specific factual questions. Opening hours, entry fees, and address lookups. The kind of thing you'd use Google for anyway.
What it is not good at is the judgment layer. Deciding which four places out of fifty are worth your weekend. Understanding why the timing of your visit matters more than the itinerary. Knowing what to order, who to ask for, and what you'll miss if you follow the standard route.
That judgment layer is what we build into every VIANÉDITA guide. It comes from documentary expeditions, real interviews, and the specific knowledge that only exists in people who've spent time in a place on purpose.
It is good at logistics frameworks. Flight connections, visa requirements, rough budget estimates, and general information about transport infrastructure. These are factual, well-documented, and AI handles them efficiently.
It is good for first-pass research for completely unfamiliar destinations. If you know nothing about a place, a ChatGPT overview is a reasonable starting point for orientation before you go deeper.
It is good at answering specific factual questions. Opening hours, entry fees, and address lookups. The kind of thing you'd use Google for anyway.
What it is not good at is the judgment layer. Deciding which four places out of fifty are worth your weekend. Understanding why the timing of your visit matters more than the itinerary. Knowing what to order, who to ask for, and what you'll miss if you follow the standard route.
That judgment layer is what we build into every VIANÉDITA guide. It comes from documentary expeditions, real interviews, and the specific knowledge that only exists in people who've spent time in a place on purpose.
The Question Worth Asking
The next time you use AI to plan a trip, it is worth asking yourself one question: Is this telling me something I couldn't have found in thirty seconds on Google, or is it reorganising information that already existed into a format that looks new?
If the answer is the second one, you're not getting a curated perspective. You're getting a well-formatted summary of the consensus.
That might be exactly what you need. But if you've been to enough places to recognise the feeling of a consensus recommendation, you probably already know the difference.
If the answer is the second one, you're not getting a curated perspective. You're getting a well-formatted summary of the consensus.
That might be exactly what you need. But if you've been to enough places to recognise the feeling of a consensus recommendation, you probably already know the difference.
VIANÉDITA guides are built from documentary expeditions through overlooked Spain and France. Each recommendation is a decision made on the ground, not a data point pulled from a review aggregate.
