19 October 2026
How the 300-agency list was built
Les cabinets individuels sont inclus : ils vendent le même service et font face au même problème.
300 agences, 36 pays, 170 issues du classement GoodFirms dans son ordre d'affichage — ce qui favorise les agences qui travaillent leur placement sur cet annuaire — et 130 de pages pays DesignRush.
Individual practices are included: they sell the same service and face the same problem. The list holds 300 agencies in 36 countries. 170 come from the GoodFirms SEO ranking in the order that directory displays them, which favors agencies that work on their placement there. 130 come from DesignRush country pages.
Comparé à l'étude des 500 sites, la médiane est 82 contre 59. Les signaux d'autorité sont la seule dimension en retrait (médiane 67). La structure du contenu est à 75, inchangée. Les trois autres sont à 100. C'est le résultat, pas un classement.
Against the 500-site study, the median is 82 versus 59. Authority signals are the only dimension still behind, at a median of 67. Content structure stays at 75, unchanged. The other three sit at 100. That comparison is the result. The index is the ordered reading of the same scans.
What was measured
Each homepage is fetched as the server sends it. JavaScript is not executed. Eighteen checks are grouped into five dimensions. The score is a weighted mean: brand clarity 25%, content structure 20%, authority signals 20%, AI accessibility 25%, schema coverage 10%. A check either passes or it does not. SoftwareApplication schema is recorded and does not move the score. A host is scored, unreadable, or not a website.
How a rank is assigned
The binary score comes first. When two scores match, the depth of each signal breaks the tie, in this order: a named person in JSON-LD, then any other author marker; a parseable date, then any other date marker; the number of distinct external registrable domains; how many of name, url, sameAs and logo are filled on an Organization; whether the title contains the agency name from this list; the byte length of llms.txt; the number of H2 headings; the number of list items. Depth is read from the HTML stored at scan time, capped at 2,000,000 characters. The llms.txt body and the sitemap body were not stored, so only the llms.txt length and the binary sitemap check are available.
Two agencies that match on the score and on every depth criterion share the rank. The next rank skips the tied places. The domain name is not a criterion. The index publishes the external-domain count, which separates the most agencies that share a score. Each fiche names the criterion that places that agency ahead of the next one. A free rescan reads the homepage again, replaces that fiche, and recomputes every rank. An unread fiche states the stored cause: an HTTP code, a robots.txt disallow, a certificate mismatch, or a timeout. An HTTP 403 means the site refused a crawler it does not know. That is not a defect of the site.