citysense
How a citysense guide is made
A citysense guide is written in two layers that never mix. The first is a catalog of the city: its neighborhoods, landmarks, food, transport, phrases and local codes, researched and checked before anyone orders anything. The second is the writing: when you order, an AI reads that catalog together with your answers and decides what to put in front of you and how to say it. The rule it works under is short: choose from the catalog, never add to it. Below: how the catalog gets built, what the checks look like, what the AI does and does not do, and what arrives in your inbox.
The catalog comes first
Before a city goes on sale, it gets a catalog. For each city that means a pool of neighborhoods with their character, who they suit and what sits within a walk; the landmarks that matter and when they were built; the dishes and the kinds of places that serve them; how the transit works; the phrases worth knowing; the cultural calendar; and the codes, from greetings to meal hours to what to wear in a church. A catalog holds between six and ten neighborhoods, sixteen to twenty five landmarks, and twenty to thirty phrases, so the guide has room to choose and nothing in it is filler.
The drafter is an AI session working under one obligation: every dated or named claim gets searched and sourced at the moment it is written. Which year a palace was finished, who designed a bridge, when a market closes, whether a museum wing is open. Each source is logged in a file we keep as the audit trail.
The second pass shares nothing with the first
Then a second AI session audits the catalog from scratch. It does not see the drafter's sources, its reasoning or its notes. It reads the catalog file and nothing else, and it searches every claim again on its own. Every date, every name, every rule about tipping or shoes or train hours.
Its findings land in a log in four categories. Critical errors are fixed before anything else happens. Minor ones, wording and precision, are applied by default. Notes record what is defensible but worth explaining. And judgment calls go to a person: the editor rules on them one by one, and the ruling becomes a precedent for the next city.
Only then is the catalog locked. A locked catalog does not change until it goes through the same cycle again.
Facts that repeat live in one place
Some facts are not about one city. Entry rules for visitors, regional tipping habits, how transit cards work in a country. Those live in a shared registry that is re-verified before any new city is drafted, and a change to the registry needs the editor's approval, because one sentence there reaches several guides at once. We learned this the hard way: an outdated line about entry rules once copied itself from one European catalog into four others, and the second pass had to catch it five times. The registry exists so that cannot happen again.
What the AI does when you order
Two guides come out of one catalog. The City Guide is the read every visitor needs: how the city is put together, the three neighborhoods every first visit touches, how the city eats, transport, phrases, the codes. The same twelve pages for everyone, ready now. The Personalized Guide starts from that read and adds what depends on you: three neighborhoods chosen for your interests, eight experiences matched to them, and three days at your pace. Eighteen pages, built in minutes from five minutes of questions.
The personal part is where AI is good. Someone who travels for food and goes to bed late is pointed to different neighborhoods than someone who wants the old core at the door and an early start. Matching, ordering, framing: a model does that well. What it does not do is decide what exists. If a place, a date or a rule is not in the locked catalog, it does not appear in your guide. The model may leave something out; it may not make something up.
Types, not addresses
A guide that names a specific bar is wrong the week the bar changes hands. So a citysense guide recommends by type and anchors to landmarks: a family trattoria near the old market, a covered food hall, a hillside of terraced cafés. Landmarks stay put and types stay true, which is how a guide written months before your trip is still right the week you travel. It is also why you will not find restaurant lists in it. The guide tells you what to order and where to look, not where to book.
What arrives
Your guide is rendered as a PDF and delivered two ways at once: as a download the moment it is ready and by email. No account to create, nothing to install, nothing to renew. If the link ever goes missing, the recovery page on the site sends it again.
Questions people ask
Is the guide written by AI?
The words are chosen and arranged by AI, from a catalog of facts that was researched and checked before you ordered. The AI decides what to show you and how to say it. It does not decide what is true.
How do you check the facts?
Twice, by two separate AI sessions that share nothing. The first searches and cites a source for every dated or named claim as it drafts. The second starts from zero, searches every claim again and logs what it finds. Errors are fixed, judgment calls go to the editor, and the catalog is locked.
Can a citysense guide be out of date?
The facts are as current as the catalog's last check, and the guide favors things that change slowly: landmarks, neighborhoods, how a city eats, its codes. It does not list specific restaurants or shops precisely because those change. When a catalog is re-checked, every guide ordered afterwards carries the update.
Who is behind citysense?
A small editorial team and the process on this page. The byline on every article and guide is citysense editorial; the facts come from the catalog, never from memory.
Written by citysense editorial. How we work, as of October 2026.