We looked for 695 hostels on Hostelworld by name. We found 203.

Twenty-nine out of every hundred

Every morning our rate reader opens a city's public Hostelworld page, takes the listings on it, and reads what a dorm bed costs. Before it can do that for one specific property, it has to know which listing on that page is that property. There is only one way an outside system can work that out: compare the name it holds against the names on the listings.

As of this morning, 10 October, we have gone looking for 695 properties that way. We found 203. Twenty-nine out of every hundred. The other 492 are either not there or not findable under the name we hold, and the engine records which of the two on each record, so this is a count and not an impression.

The seven days to 9 October are more specific. Forty-three readings completed. One hundred and ninety-eight attempts came back empty, each carrying its own recorded reason:

• 91 — no listing matched the property's name among the listings we had for its city
• 47 — not on Hostelworld at all, checked and stamped as such
• 16 — a listing exists, but no night in the sample carried a price for that property and at least three competitors at once
• 44 — the morning's read budget ran out before the property's turn came

Take out the last two, which are our problem and nobody else's, and 138 of the 154 remaining misses are one problem: we could not connect a property to a listing. Nine in ten.

Where the matching breaks

A Hostelworld city page shows roughly the first 31 listings and no more, so a property further down its city's list can never be matched off the page alone. For those we read Hostelworld's own sitemap, which on 6 October carried 37,778 property URLs, 20,085 of them under hostels.

Then the names get compared, and three rules in that comparison are worth an operator's attention.

City words come out first. Half the listings in Madrid have Madrid in the name and it identifies nobody. What is left is your identity words, and that is the whole key you are found on.

Spelling is forgiven; identity is not. A listing written Latroupe Prado and a property that writes itself La Troupe Prado is one house, and the matcher is built to see that.

And a word that appears in two or more listing names in the same city is treated as a place, not a property. Alfama. Ribeira. Chueca. Condesa. A property whose only distinctive word is one of those matches nothing at all, deliberately: picking one of the four listings that share it would be a coin flip. Our threshold is strict for the same reason — a wrong match puts somebody else's price in an email with your name on it.

The first time we ran this, on 5 October, it read 23 properties and matched none. They were riads in Marrakech and hostales and residencias in Madrid: places that do not sell dorm beds on Hostelworld at all. That is a category problem rather than a name problem, and it is why the engine now stores not on Hostelworld as a real answer instead of retrying forever.

Why this is not only our problem

A miss here costs us one email. The same mismatch costs you in places where nothing tells you.

Your competitor set gets shorter and nothing says so

Our own hostel, Maka in Oaxaca, prices every morning against three named Hostelworld competitors. Three. If one renames itself — new owner, new brand, a city word added or dropped — the read that found it by name stops finding it, and the engine is reading a two-property market while believing it has three. The floor it computes moves. Nothing in the log says a competitor vanished, because from the inside a shorter list looks exactly like a cheaper market.

Any rate tool keyed on property names has this failure mode, ours included. Count the properties in your comp set today against the number you set it up with.

The name you are indexed under is not the one you say

You have at least five name strings in circulation: your Hostelworld listing, your website's browser tab, your Google Business Profile, whatever your channel manager pushes to channels, and your Booking.com page. They drift apart one small edit at a time, and each is somebody else's key for you. The version in the index decides whether a guest finds you, not the version you say.

The three hundred and seven

Yesterday morning's match run walked 13 cities. It stamped 14 properties and matched 2. It also counted 307 listings in those same cities it could not tie to any property in our list of 2,622. Some are houses we hold under a different name, on the page, in plain sight, invisible to us. We publish that number rather than drop it, because it is the honest size of the gap.

The ten minutes this costs to check

None of this needs software.

• Open your Hostelworld listing and copy the name exactly as it appears, accents and all.
• Put it beside the other four and compare them as strings, not as ideas.
• If they differ, decide which one is actually your name and change the rest to match. The version that matters least is the one you say out loud.
• Strike out your category word — hostel, hostal, hotel, casa, posada, riad, guest house — and strike out your city. If what is left is a neighbourhood other listings also carry, no name-based system can tell you from them.

What the reading looks like when the name works

For scale, one city. On 8 October the reader took Panama City: 25 listings on the page, 7 carrying a price for the night of 18 October, median USD 15.75 a bed, cheapest 9.00, dearest 36.18. Public Hostelworld prices, one guest, one night, read that morning, nothing modelled. A whole market in four numbers, and completely unavailable to any property the reader cannot name.

Nightfill takes that reading and makes the pricing decisions off it every morning on a property's behalf; the three tiers are at view pricing. It is independently operated, and what it does depends on your property, your market and your season.

Before any of that, there is the free read. Paste your Hostelworld listing URL at nightfill.co/#read with an email address, and the same reading we take of a city arrives for your own property by email in about ten minutes. We ask for the URL and not your name on purpose. A URL is an id and cannot be wrong. A name, on the evidence of 492 records out of 695 sitting here, can be.