How this register is built

The map

OpenStreetMap knows where studios sit: an office tagged as graphic design, web design, advertising or IT, carrying a website tag. That is open data under ODbL, and it is used here with attribution.

Where a studio sits

A country comes from the national domain zone, from the map, or from the studio's own site. A city comes from the address our crawler read on that site, or from what the map holds for the office — never from a form somebody filled in. Because a street and a suite number look exactly like a city, a parsed name is only believed once it has turned up at least three times in its own country: a mistake is singular by nature, a city is not.

A city gets its own page at three studios. Below that, a parsed name is more likely to be a mistake than a city — and a city with two studios would not be worth a page anyway. So a city can be missing today and present next week: the register keeps reading, and the page appears when the third address turns up.

A district belongs to its city. A studio in Shoreditch is a London studio, and that is where a client looking for London expects to find it; the same goes for Surry Hills under Sydney or North York under Toronto. Places that sit inside a city — within eight kilometres of it and twenty times smaller — are shown as part of it. Boroughs people search for in their own right, Brooklyn and Manhattan among them, stay where they are.

One city, one name. Studios print their address in their own language, so the same city arrives as Warszawa and Warsaw, Praha and Prague. They are one page under the name the world uses, and the other spellings redirect to it rather than disappear.

To answer «near me» a city also needs a point on the map. Those come from GeoNames (CC BY 4.0) — the city, its name in every spelling, and its coordinates. Your own location is never part of this: the nearest-cities page does its arithmetic inside your browser and sends nothing back.

The footers

Every site a studio builds tends to carry its name at the bottom — Site by, Realisierung, Réalisation, Разработка сайта. Reading those footers gives two things at once: a studio that was not on any list, and a count of the sites it has signed. That count is the only figure here that is measured rather than claimed, which is why it sits in the first column.

Those signatures are published, not just counted: every studio page carries the list of sites it signed, and every one of those sites has a page of its own naming the studio and quoting the line we read. If you have an address in mind, look up who made it.

Reach: the one number we compute

Every other figure here is a count. Reach is the exception — it is derived, and it is built on somebody else's metric, so it is worth spelling out exactly what goes into it.

The input: Domain Rating by Ahrefs. DR scores a domain's backlink profile on a 0–100 scale. As Ahrefs describe it, the calculation walks the link graph: it collects the unique domains that carry at least one followed link to the target; it looks at how many domains each of those referrers links out to; and it passes on a share of the referrer's own rating, split across those outgoing domains — so a link from a selective site is worth more than one from a site that links to everyone. Because each referrer's rating is itself computed the same way, the procedure is iterative, in the manner of PageRank. The result is then mapped onto the 0–100 scale, and that scale is logarithmic: the distance from 70 to 80 is far longer than from 20 to 30. Ahrefs call the metric purely link-based and state that search traffic, domain age and brand strength are not part of it; nofollow links pass nothing, and a second link from the same domain adds nothing. They do not publish the exact formula, and DR is relative to the rest of their index — a domain can slip without losing a single link, simply because other domains gained.

What we add on top. A studio's own domain says how strong the studio's own site is. It says nothing about the work. So we sum the Domain Rating of the studio's site with the Domain Rating of every site we have recorded as built by it — the sites that carry its signature in their footer. One number, one meaning: the total link strength of everything that studio put on the web.

reach = DR(studio) + Σ DR(sites the studio built) + awards

One domain, one site. Some studios build a whole network for one client — fifty parish sites on one diocese's subdomains, a school board's every school. Domain Rating belongs to the domain, so each of those subdomains would bring the same rating again and turn one client into fifty works. We count such a group once: the domain itself if it is among the studio's sites, otherwise the strongest of them. The others stay out of Reach.

Awards from other juries. An award is a second, independent opinion about the work, so it adds to Reach. Awards come from The FWA's open data, which we read ourselves, and from studios, who send the link to the award page for our editors to check. Each award is weighed by the jury and by the level: the large industry juries (Webby, Cannes Lions, D&AD, The One Show, Clio, Red Dot, iF) start at 60, Awwwards, The FWA and CSS Design Awards at 40, specialist awards and curated galleries at 15, small galleries at 5. Year and Grand Prix count four times, month and gold twice, the day once, honourable mentions and kudos half; a nomination counts nothing. Several levels for one work count as the best one plus a fifth of each other. Awards from the last two years count in full, from two to five years ago by half, older by a quarter.

Two consequences follow, and both are intended. A studio with a modest site that built ten strong ones outranks a studio with a strong site and nothing behind it — which is the right way round for a register of people who build for others. And because the underlying scale is logarithmic, the sum is not a league table of quality; it is a measure of accumulated reach. Every card shows how the number breaks down: the studio's own rating, the rest, and how many sites it was summed across. We take the ratings once a month.

The jury: the one judgement here

Everything above is observation. Whether a website is any good is not something a crawler can read out of a footer, and it is not something we are willing to decide ourselves. So we ask the studios that have already taken a place in the register.

A seat is earned, not applied for. A place in the world top 100 of the monthly snapshot opens the whole register to that studio. A place in the top ten of a country opens the sites built by studios of that country. A place in a city opens nothing: a city holds too few studios to make a panel. The snapshot is retaken on the first of every month, and the panel is taken from the newest one — a seat is held, not owned.

Sites arrive one at a time, and the ones nobody has scored yet come first: a panel is a scarce thing, and spending it twice on the same website before the next one has been looked at once would waste it. Past those, the order is random. The juror gives a score from 1 to 10, and that is the end of it: one score per site, no editing, no withdrawal. Two guards make the scoring mean something. A studio is never shown its own work. And no juror is told who built the site on the screen — the address is there, because you cannot look at a website without opening it, but the studio behind it is not named. A panel of ten studios from one country knows itself by name; scoring with the name in view would be a vote for friends, not a judgement of work.

What a site page shows is the average of every score it was given, with the number of jurors beside it. Not the sum: a sum would compare luck with the number of voters rather than the work. And never on its own — one score from one studio is one opinion, and the page says so by printing the count in the same breath as the number. A site nobody has scored shows nothing at all, because a zero would state something about the work that nobody actually said.

What is not here

Contact addresses. They are collected, they are not published: putting a compiled list of addresses on an open page devalues it and changes what it is. Open reviews are absent too: anybody with a grievance and an evening can fill a directory with them, and a register that accepts them stops being a record of anything. The jury above is the single exception, and it is fenced on every side — the seat is earned by a measured place, the work is scored blind, and the score is published as an opinion with its jurors counted, never mixed in with what we observed.