How we write the guide

Every fact comes from a source anyone can open: the hotel’s own website and its FAQs, the booking sites, Tripadvisor and Google reviews, forums, maps, YouTube and Instagram. We keep a copy of each source and the date we read it. A review picked out by one word tells one person’s story, and we treat it that way: one voice among many.

The page says so. If the hotel’s website gives one figure and a booking site another, we print both, each with its date, and say which we think is better supported and why. Anyone can check everything we write, without going through the hotel.

On every hotel page we tell five things as guests tell them: the location, the room, the staff and service, breakfast and the bill. To do it we sifted 115,111 Google reviews with text, from 33 to 500 per hotel. An AI model reads them one at a time and marks what each guest talks about and in what tone, from very badly to very well. Before trusting it we tested it on reviews whose answer we already knew. The counting is done by code. On the page we say it in words, “9 in 10 guests who mention it are happy with it”, and anyone who wants the figures finds them underneath, with how many voices stand behind each.

Some hotels in the guide hold one, two or three Lumi. A Lume says the hotel gives more than the others of its place and price range. We decide it on three readings: the guests, the critics, meaning other guides and the press, and our own. The data nominates and vetoes, then the person who signs decides. What each level means, how many hotels hold it and which they are is on the page “The Lumi”.

We measure walking and driving times ourselves, on the actual streets: from the hotel’s entrance to the public entrance of wherever you’re going, with two independent route planners, OSRM and Valhalla, on OpenStreetMap’s maps. The real time still depends on your pace and the rain.

We read them on Google Hotels, for two adults and one night, on several nights, usually in different months: 9,945 readings for the guide’s 307 hotels. Each figure carries the night it applies to and the date we read it, and taxes and resort fees are shown the way the source shows them, included or not. A figure seen today is no promise for your dates.

We use the hotel’s official photographs at the highest resolution available, press libraries included, or photographs from Wikimedia Commons with their author and licence. No photograph is retouched to add sky, light or detail that wasn’t there. We choose the 147 videos on the pages one by one and start each at the minute that matters; two films, for Portrait Milano and Il Ciocco, are our own.

We use it for the long work: reading hundreds of reviews and pages, working out routes, comparing prices, and gathering the questions people put to AI assistants before they book. It helps us write too, and our checks turn down any text that drops a number or describes a visit that never happened. Every price and every walking time a page quotes has to match a reading we have kept; if it doesn’t, the page doesn’t go live. Who a hotel suits, and who it doesn’t, is decided by the person who signs the page.

The date at the top of each page is the day its content last changed. When a fact changes, the page and its date change with it.

No one. No hotel pays to be in the guide, for the verdict it gets, for a Lume or for a badge, and none can ask for them: we award them. Lumina earns nothing from bookings: the links to the hotels carry no commission. The tag we add to those links is there only so the hotel can see that the visitor came from the guide.

Write to Marco at marco.dicesare@luminafrontier.ai. We’ll correct the page and update its date.