Tour operator, AI assistant or engine

You have three weeks. You can hand the trip to a tour operator, or ask ChatGPT for it in ten seconds. Both will give you a programme.

Neither will ask you how many times you are willing to repack your suitcase.

Yet that question decides half of a long trip. And it is not the only one.

The operator: the ground, at the price of control

Most of our long trips were built by operators. They were good. The accommodation kept its promises, the logistics were handled, and the one time something went wrong, somebody picked up the phone.

A good operator knows the ground better than any map. It is their trade, and it cannot be improvised.

But it is also a sales trade, and that has consequences that do not depend on how good the operator is.

You hand over control. Every stage is a line on the quotation, and a programme that ticks twelve places looks more generous than one that ticks six. On our trips: a different bed every day and a half, with different operators.

It takes time. A quotation, a comment, a new quotation. By the third round, you accept the programme as it is.

You do not see what was left out. No quotation says what it gave up to hold together.

The AI assistant: an answer to the question asked

An assistant like ChatGPT or Claude knows a remarkable number of places. In seconds it offers a well-written itinerary, for free, and it is improving fast. We use one ourselves.

But it is a generalist, and that too has consequences that do not depend on its quality.

It answers what you ask it. If you do not tell it how often you are willing to repack, it will not concern itself with that. If you do not give it your temperature ceiling, it will not take it into account. If you do not know that a long trip needs a breather in the middle, it will not suggest one. A good plan starts with the right questions — and those are exactly what you do not know when planning your first three-week trip.

Its answer commits no one. It is a conversation, not a published plan: nobody signs it, nobody corrects it, nobody answers for it.

What both lack: a method

The gap is not a better answer. It is a method: knowing which questions to ask, and writing down what each choice costs.

The questions nobody asks, and that nonetheless decide a long trip:

How many times are you willing to repack? The most decisive parameter, and it appears on no quotation. → Base-change tolerance

Beyond what temperature does your day stop? Escaping heat is not a matter of latitude, it is a matter of ceiling. → Temperature ceiling

Where do the days with no plan go? In the middle, never at the end — otherwise they happen anyway, on a good day wasted. → Day with no plan

Is what you asked for what is planned? Counted in nights, not impressions. And a repetition you did not ask for is not an extra option. → Register saturation

And a method only counts if it shows: several variants that do not resemble each other, each with what it promises and what it gives up, written down.

What we built

An engine that applies this method, and publishes what it calculates.

The language model writes, the engine calculates and verifies. Each does what it does best. Knowledge of places and their descriptions come from a language model. Everything that can be measured comes from calculation: the month's climate from records, distances along the real road, rhythm counted stage by stage, consistency checked.

Every figure has a source, every check leaves a trace. An example. On 16 September 2026 we applied a single geographic consistency rule to forty-one itineraries. It found three real defects, including a night planned in Afghanistan on a Tajik circuit — a country under a formal "do not travel" advisory. The text of the stage was impeccable; the check caught it. And our own errors are published.

Every itinerary says what it gives up. Several variants, their promise, their trade-off, their rhythm, and the share of the territory they really cover. For the task-by-task detail of work done by hand, see How long to build a tour itinerary?.

And we go further

An itinerary is not a list of stages. It is what happens each day.

We build your route, day by day, with the nights where they make sense.
We prepare your hikes, stage by stage.
And an app follows you on the spot, so the plan does not stay in a PDF.

Early access — tell us what interests you.

Before you leave: three checks

Whoever made your itinerary — an operator, an assistant or yourself.

Check the driving times against a route planner. A driving day longer than planned shifts every day after it.

Check the state of connections — mountain roads, ferries, permits — against official sources, for your dates.

Count stages against nights. If you change accommodation more often than every two days, ask why: geography sometimes forces it, so does padding.

You can also send us the itinerary: we review it for free.

All seven checks, one by one, with the cases behind them: How to check an itinerary planned with ChatGPT.

What we do not do, and what we do not know

We book nothing, and we do not replace an operator for those who want to delegate everything.

We do not verify the world in real time. Ferry timetables, mountain-road opening dates and permit quotas have no worldwide source: we flag the constraint, we do not claim to have checked it for your dates.

And our thresholds — a day and a half of consecutive rest, a ceiling of twenty-four degrees — are parameters of the engine, set per traveller, not rules: yours may sit elsewhere, and the engine takes it into account.

Further reading

• Eleven hotels in twenty nights — the count that started it all.
• It isn't the north you're looking for — the ceiling, not the latitude.
• Where to put the day with no plan — in the middle, never at the end.
• The Baltics repeat themselves — and the 1,200 km of trail no circuit sells.
• Reading a feasibility check — one grade, six dimensions, assumptions.
• Overloaded tour: the corrected variant — what the engine proposes, gap by gap.
• The method, in six terms
• Are you an operator or a software vendor? — the engine can work for you.

Experienced something different — an operator who let you decide everything, an assistant that asked you the right questions? Tell us — corrections are published.