The address a machine cannot read
We read 1,004 small-business websites across six countries. Eight in ten publish an address a machine can read. The other two are not offline: they never write their city anywhere.
81%
of sites give up an address a machine can read
In this article
Short answer: one business in five does not publish, on its own website, an address a machine can read. We read 1,004 small-business websites across six countries. 812 gave us a city. The other 192 are not offline: probed the next day, 196 out of 200 served their pages normally. Their address is missing for no technical reason at all - it is simply not written anywhere a machine looks.
61%
in Brazil, against 99% in Switzerland
196 of 200
silent sites answer normally: the absence is not a fault
A site showing its address in an image, or only inside an interactive map, is fine for a customer and invisible to the machine deciding whether to name it.
What we measured
On 6 September 2026 we visited up to seven pages per site - home, contact, legal notice, about, in the country's own language - and looked, in the page TEXT, for a full postal address: a postcode next to a city name. Six markets: the United Kingdom, France, Belgium, Switzerland, Poland and Brazil. Ten seconds per page, no JavaScript executed. We read what an ordinary crawler reads, not what a human sees in a full browser.
These 1,004 businesses share one trait: in our own records their city was wrong or missing - a region, a county, a country. We were trying to repair it by reading it at the source. This is therefore not a random sample of the world economy, and we say so before the numbers.
Eight in ten, and the country changes everything
812 sites out of 1,004 gave us a city, or 81%. The gap between countries is far wider than any gap between trades, and it is striking.
| Country | Sites read | Readable address | Share |
|---|---|---|---|
| Switzerland | 111 | 110 | 99% |
| Belgium | 144 | 136 | 94% |
| France | 140 | 112 | 80% |
| Poland | 129 | 103 | 80% |
| United Kingdom | 322 | 254 | 79% |
| Brazil | 158 | 97 | 61% |
Switzerland and Belgium lead by a distance, and the obvious hypothesis is legal: both countries require a company identification page - the Swiss *Impressum*, the Belgian mandatory particulars - which carries the registered address. Brazil, where nothing compels a business to publish an address on its site, comes last at 61%. We do not prove that link; our measurement does not test it. We note the coincidence and name it for what it is.
An address is not a sentence
The most instructive part is not the rate, it is what made us fail. An address buried in running text does not read: we came back with "Code APE", "Numéro", "Let's" and "Contact" in place of cities, because those words followed a postcode directly in the page flow. What we had to learn to handle, an engine has to learn too - and it does not always.
| What the page writes | What the machine understands | The real city |
|---|---|---|
| 1400 Let's talk | Let's | Nivelles |
| 75008 Paris Adresse email | Paris Adresse email | Paris |
| 01-355 Warszawa zobacz na mapie | Warszawa zobacz na mapie | Warsaw |
| SIRET … Code APE 8690F | Code APE | none |
The British case deserves its own line. In the United Kingdom what precedes the postcode is very rarely a city: across the 152 records we repaired, the original value was "England" 55 times, "West Midlands" 8, "West Yorkshire" 7. A county places nobody: there are dentists all over Yorkshire. Writing "Leeds, LS1" rather than "West Yorkshire" is not a matter of style, it is the difference between being placeable and not.
Why your city decides, more than it used to
A directory could guess your city: it inferred it from your phone number, your postcode, a form somebody filled in ten years ago. An AI assistant answering "a plumber in Ghent" works differently: it reads pages, pulls names out of them, and has to attach each name to a place. What it cannot attach, it does not offer.
A business whose site does not state its city is not ranked poorly in the answer: it is not among the candidates. That is a difference in kind, not in degree.
What this measurement does not prove
- The sample is not random: these are businesses whose city was already wrong or missing in our data. They may publish worse than average - or better, since their own site is sometimes their only correct shop window. We do not know.
- We read the HTML text without executing JavaScript. A site that builds its address by script, or shows it only inside an interactive map, publishes it for a human and not for us. That is also what many crawlers see, but not all of them.
- Seven pages per site at most. An address filed on an eighth page escaped us.
- Reading an address is not the same as being cited by an AI assistant. This measures how readable a piece of data is, not a position in an answer.
- The link between a legal duty and the publication rate is a hypothesis this measurement does not test.
What it changes for you
Write the address as text, in a block of its own.
Not mid-sentence, not in an image, not only in a map widget. The postcode and the city must sit together.
Give the city, never the region.
"Normandy", "England", a Brazilian state: none of those places a business. The city, and its postcode.
Put it on the home page, or on a contact page one click away.
We looked across seven pages; an engine is not always that patient.
One form, everywhere.
The same wording on your site, your listings, your profiles. Two spellings of one address are sometimes two different businesses to a machine.
Check what a crawler sees
, not what your browser shows: open your home page source and search it for your postcode. If it is not there, neither is your city.
You can see in a minute, free and without an account, what AI assistants answer when someone looks for your trade in your city.