What we audit, and what we don't
FetchGrade runs a fixed set of deterministic checks across five categories. Every result maps to something we observed on the page — not a model's opinion. Below is exactly what each category measures, how the score is calculated, and the boundaries we keep.
Readable
Visible text, headings, the page title, image descriptions, and content structure — whether key information is real, selectable text rather than locked inside images.
Navigable
Landmarks, link and button names, menu state, keyboard reachability, and whether sampled first-party routes respond.
Actionable
Forms and field labels, required-field clarity, contact paths, the primary call-to-action, and whether overlays block core controls.
Extractable
What an AI crawler that does not run JavaScript can read: how much content is in the raw server HTML, structured data and metadata, and whether robots.txt allows the major AI crawlers.
Automation friction
JavaScript errors, CAPTCHA or bot challenges, login walls, unsafe redirects, and how quickly the page settles after load.
How the score works
Each category scores (earned ÷ applicable points) × 100. Checks that don't apply to a page are removed from the denominator — never counted as a pass or a fail. The overall score is the sum of each category's contribution, weighted by the percentages above.
We weight checks by evidence of real effect. The Extractable category leads with whether content survives without JavaScript — the thing most AI crawlers actually depend on — and deliberately weights signals with weak evidence, like an llms.txt file or the sheer quantity of structured data, close to nothing.
- 85–100Strong foundation
- 70–84Good foundation
- 50–69Needs attention
- 0–49Significant friction
What we deliberately don't do
- No free-form clicking, typing, submitting, signing in, buying, or solving CAPTCHAs.
- No whole-domain crawl — we audit one URL plus a small set of public metadata.
- No security, WCAG, SEO, or AI-search ranking claims.
- An LLM only rephrases findings into plain language; it is never the source of truth.