Methodology
How a PageReflect score is produced
Every report is a number you can act on, so the way we produce it is public. This page describes the six dimensions, their weights, what the Lighthouse lab measures versus what the model judges, and how per-page signals become one sitewide score.
The overall score
The overall score is a weighted average of six dimension subscores, each on a 0–100 scale. It is always derived from the six subscores below, never taken from raw model output, which can contradict its own subscores. The weights sum to 100%.
| Dimension | Weight | Source | What it measures |
|---|---|---|---|
| Performance | 20% | Model-judged | Judged from the mobile Lighthouse lab pass (LCP, INP, CLS, TBT) captured per page, read against current thresholds. |
| SEO & discoverability | 20% | Model-judged | DOM structure, title and meta, canonical and robots signals, structured data, internal linking, and indexability. |
| UX | 15% | Model-judged | Navigation, friction, and conversion fit: how quickly a visitor finds what they came for and can act on it. |
| Accessibility | 15% | Lighthouse lab | Lab-anchored directly to the Lighthouse accessibility category (contrast, ARIA, keyboard, screen-reader semantics), set server-side. |
| Content | 15% | Model-judged | Messaging clarity, CTA strength, trust language, and depth on every sampled page. |
| Design & trust | 15% | Model-judged | Multimodal read of visual hierarchy, trust signals, and mobile rendering across the sampled pages. |
Lab versus model
Accessibility is lab-anchored: the score comes straight from the Lighthouse accessibility category, set server-side at the time the report is persisted, so it is reproducible and not open to model interpretation. Performance is judged against the same mobile Lighthouse lab pass, LCP, INP, CLS, and TBT captured for each page. SEO, design, content, and UX are model-judged from the structured signals we extract from the page, held to a fixed rubric so a finding is categorized the same way everywhere it surfaces.
From pages to a sitewide score
PageReflect audits multiple pages, not just the homepage. We discover pages from the sitemap first, then a bounded crawl of internal links, and sample the most representative set: up to 5 pages on the free plan and up to 15 on Pro and Lifetime. Each sampled page runs the identical treatment, Lighthouse lab pass, structured DOM parsing, screenshots, and the model assessment, so pages are comparable. Per-page signals roll up into one subscore per dimension, and those six subscores produce the overall number.
Versioning
Weights and the scoring rubric can change as the product improves. Every report records the pipeline and prompt version it ran under, so a score is always interpretable against the methodology in force when it was produced. This page reflects the weights in the current pipeline.
Want the full tour of a report? Reading your report walks through every dimension, or multi-page audits covers how sampling and roll-up work in depth.