A benchmark is only as good as its rubric. Before we scored 771 agency websites, we wrote down exactly what a good one shows a buyer, and how to give it a number. Here is the rubric, so you can argue with it or reuse it.
The structure
- Looks and feel, 10 criteria: first impression, visual identity, typography, imagery, layout and whitespace, motion, consistency, mobile experience, craft and polish, and homepage storytelling.
- Usability and conversion, 10 criteria: positioning clarity, navigation, service findability, call to action, contact, form friction, pricing transparency, process clarity, readability and interactive aids.
- Proof and trust, 6 criteria: case studies, social proof, credentials, team presence, trust basics, and security and compliance.
- Offer and expertise, 4 criteria: technical depth, AI capability, thought leadership and industry focus.
- Technical, 6 criteria measured by script: speed, accessibility, on-page SEO, structured data, AI-search readiness and languages.
The scale
Every criterion runs from 0 to 20. A 20 sets the global standard for agency sites. 16 to 19 is clearly above the agency norm, 11 to 15 is solid, 6 to 10 is present but ordinary or flawed, 1 to 5 is weak or token, and 0 is absent or actively damaging. A feature that is absent scores 0 to 3 even if the business does not need it, because the benchmark measures what a site actually shows.
Rules that keep judging consistent
- Score only what the site shows a buyer. Brand fame, Clutch rank and company size earn nothing.
- The same evidence earns the same score on any site. Definitions say what a 20 looks like, in words.
- Be conservative when evidence is thin. If a page was blocked or a section is missing, say “limited evidence” and do not guess high. From still screenshots, motion is capped at 14.
- Every score carries a reason of about twenty words that cites concrete evidence, and every answer must pass a validator before it counts.
What we left out on purpose
We did not score the quality of the agency’s actual engineering, the truth of its claims, or client satisfaction. A website cannot show those, so a benchmark of websites should not pretend to measure them. The standards behind the measured part are WCAG 2.2 AA for accessibility and Core Web Vitals for speed; the judged part draws on usability heuristics, not on any single authority.
If you disagree with a weight or a definition, that is the point. A rubric you can read is one you can improve.