Methodology

Popularity is not quality. The most-visited website is not automatically the best one. We start from a popularity-ordered seed only to decide what to research first — never to rank quality.

We model the web as a typed graph that distinguishes companies, products and domains. Google, Google Search and Gemini are separate products of one company. Relationships between products are typed as direct alternative, adjacent tool or complementary, each with a strength and a written rationale — so we never call two things competitors just because an embedding model found their descriptions similar.

Currently: 25 products · 24 companies · 46 relationships · 10,000 domains.

Every meaningful claim records its source, a timestamp, whether it was directly tested, the responsible model/reviewer, and a confidence. Evidence is classified by how we know it:

We never claim firsthand testing when only marketing pages were read, and we never fabricate prices, features, traffic, certifications, benchmarks or testimonials.

Scores are versioned and separated into distinct axes so they can’t be conflated:

• Website quality — task fit, UX, functionality, reliability, trust, accessibility, value, interoperability.
• Fit for purpose — recomputed per use case and per stated preference.
• Popularity — Tranco rank, shown separately and never mixed into quality.
• Evidence confidence — how well-supported the assessment is.
• Freshness — when it was last verified.

Dimension weights are configurable per category and use case. Crucially, an unreviewed property scores as “unknown”, with reduced confidence — never a silent zero, and never a fabricated number. On profiles today you’ll see most dimensions marked “not yet reviewed”, because the browser-tested review pipeline runs progressively.

Our domain seed is Tranco (list 8PX5V, generated 2026-10-09), a research-oriented ranking that combines Google CrUX, Farsight, Majestic, Cloudflare Radar and Cisco Umbrella using the Dowdall rule. We store the provider, list version, retrieval date and methodology for every domain.

License note: Tranco is intended for research use; its ranking is not an endorsement, and we do not resell the list. We treat it purely as a processing-order signal and record its provenance on every profile.

Infrastructure-only, CDN and tracking domains are flagged and kept out of editorial reviews rather than padded in to hit a number.

Research and drafting are AI-assisted and evidence-tracked. AI-written editorial carries disclosure, and our scores are not presented as verified aggregate customer ratings. Website owners and readers can submit verifiable corrections — which never buy a better ranking. Submit a correction →