Skip to Main Content

How We Measure

Every site in this gallery carries a set of signals – the CMS it runs on, how it scores on performance, how old the domain is, how a procurement buyer would rate its content. We measure more than 25,000 industrial marketing websites the same way, so the numbers mean something when you compare one site against the rest of the field.

Not every signal is the same kind of thing, and pretending otherwise would be dishonest. Some are measured mechanically and would return the same answer no matter who ran them. Some are judgments a language model makes from a site's content, with guardrails. One is pure subjective preference, built from thousands of head-to-head comparisons. Here's how each works, grouped by how much to trust it at face value.

Objective signals

These are mechanical. Point the same tools at the same URL and you get the same answer.

CMS platform. We fingerprint the live site against signature rules for 60+ platforms. That covers the usual suspects (WordPress, Craft, Drupal, HubSpot, Webflow, Sitecore, AEM) and some other industrial-specific commerce systems that few others track (CIMCloud, Unilog, BirdDog, K-eCommerce). A site only gets tagged when a concrete fingerprint fires, and we record which signal did it.

Performance. Four Lighthouse category scores (Performance, Accessibility, Best Practices, SEO) pulled from Google's PageSpeed Insights API on the mobile strategy, because mobile is the harder test and the more honest one. The three non-performance scores are effectively stable. The Performance score is a throttled timing run, so it moves a little between scans. We rescan quarterly, which is often enough given how rarely the stable scores shift.

Domain age. Registration date, expiry, and registrar. A registration date doesn't change, so we capture it once. It's can be a useful tell since a 2003 domain and a 2023 domain are usually two very different companies.

Security posture. Valid certificate, HTTP redirecting to HTTPS, the negotiated TLS version, and the security response headers (HSTS, X-Frame-Options, X-Content-Type-Options, Referrer-Policy, with CSP and Permissions-Policy as bonus). No valid SSL is an automatic F; a valid cert floors the grade at D; serving pre-1.2 TLS caps it at C. This is a read of header configuration, not a vulnerability scan. It tells you whether the basics are in place, not whether the site can be breached.

Size and architecture. We find a site's sitemaps through robots.txt and the well-known paths, then walk the sitemap graph to count real pages – filtering out images, PDFs, and off-host URLs so the number reflects actual content. Along with the count we capture URL depth, the top-level section tree, and any declared locales harvested from hreflang alternates, which is how we spot multi-language sites. Across the whole catalog this rolls up into something more useful than any single number, the URL and navigation patterns a typical industrial site actually uses.

Color palette. We run color extraction on each site's screenshot, pull the five most dominant colors, and classify each into a named bucket by hue and saturation. Low-saturation colors resolve to white, black, or gray; everything else is bucketed by hue. It's a rough read of what colors a brand actually uses on screen – enough to browse the gallery by color, not a substitute for a brand guideline.

AI-assisted judgments

These come from a language model reading a site's content. We use structured outputs and human-reviewable thresholds so the model reports rather than decides, but they're judgments, and we treat them that way.

Industry role. We classify each company into its role in the supply chain – original equipment, parts and components, distribution, and so on – with a confidence score. We auto-accept only classifications at 80% confidence or higher. Anything softer waits for a human.

Agency detection. Two passes. First a deterministic scan of footer links for credit phrases – "designed by," "site by," "built by." Detected agencies become their own entries, which is what powers browsing the gallery by the agency that built the work.

Preference

Head-to-head ranking. One signal isn't a measurement at all – it's aggregated taste. Sites are shown in random pairwise matchups and a winner is picked, and each site carries an ELO rating updated by the same math that ranks chess players. New sites move faster so they reach their real level quickly, and upsets move ratings more than expected wins. It's subjective by construction, and it should be read that way. Not "this site is better," but "when people compare these two, they prefer the design of this one."

Reading the CMS snapshot

On each CMS profile page, a panel of bars summarizes the industrial sites we've measured that run on that platform. Read them as a portrait of what people have built on the CMS, not what the CMS is capable of. A platform can power a 5,000-page catalog and still show a "Small business" scale bar if that's what its industrial users actually use it for. Every bar is the typical case across the sites we measured on that platform, and every share is counted over the sites we could measure, not the whole field.

Each bar fills to show roughly where a platform sits on an absolute scale, but the word next to it carries the real meaning. The thresholds are calibrated to our industrial index specifically – they'd look different for, say, restaurants or SaaS – and we retune them as the field shifts. Here's what each level means.

Scale

The median page count of sites on the platform, a read on the size of company that tends to build there.

  • Brochure: Under 30 pages
  • Small business: 30 to 149 pages
  • Mid-market: 150 to 599 pages
  • Enterprise: 600+ pages

Commerce

The share of sites that are catalog-shaped, meaning content concentrates heavily in one product or catalog section rather than spreading evenly across the site.

  • Capabilities-led: Under 15%. The platform's industrial sites mostly sell expertise and process, not a browsable catalog.
  • Blended: 15 to 34%.
  • Catalog-leaning: 35 to 54%.
  • Catalog-dominant: 55%+. Most sites here are built around a large product catalog.

Footprint

The share of sites that serve more than one language, a proxy for how far the typical customer reaches geographically.

  • Single-market: Under 10%. Almost all sites are one language, one region.
  • Regional: 10 to 24%.
  • Multi-market: 25 to 49%.
  • International: 50%+. Multilingual is the norm on this platform.

Publishing

The share of sites running a real authored-content operation – an active resource library, news, or technical content, not a static site that never changes.

  • Static: Under 5%. Set-and-forget sites.
  • Occasional: 5 to 11%.
  • Consistent: 12 to 24%.
  • Content engine: 25%+. A quarter or more of these sites publish like a media operation.

Security

A composite of the HTTPS and security-header posture across the platform's sites, blending the distribution of letter grades with how consistently sites force HTTPS. It scores 0 to 100.

  • Neglected: Under 55. The basics are missing on too many sites.
  • Baseline: 55 to 72. SSL is handled; hardening is spotty.
  • Fortified: 73 to 89. Strong headers are common across the platform.
  • Airtight: 90+. Rigorous security hygiene is the norm.

The bars pull from the same underlying signals described above – page counts, multilingual detection, and the security grade – rolled up to the platform level. What they add is comparison: not "is this site big," but "for the kind of company that builds on this CMS, is this typical."

Why the split matters

Keeping these three tiers separate is the point. A CMS fingerprint is a fact. A security grade is a fact. An audit score is a careful judgment with its reasoning attached. A head-to-head rating is subjective. Blend them into one number and you'd lose exactly the thing that makes the comparison worth trusting – the ability to know which kind of claim you're looking at, and how much weight to put on it.