Methodology: how scores are calculated

What we scan

When you submit an address, we fetch your homepage HTML, robots.txt, sitemap and llms.txt, read the HTTP response headers, and — when Google's PageSpeed service is available — run a real Lighthouse performance test. We read only; nothing on your site is changed.

The scan currently analyses the page you submit (typically the homepage). Site-wide crawling of every subpage is deliberately out of scope for a free instant report — the homepage carries most of the signals AI systems and search engines use to judge a brand.

Measured vs. estimated — the honesty rule

Every metric in the report is labelled either “Measured” (we directly observed it on your site: a missing tag, a blocked crawler, a slow response) or “Estimate” (derived from indirect signals). We never blend the two into one number without saying so.

That's also why the free scan does not show an “AI mention score”: we don't query ChatGPT or Perplexity with live prompts in the free tier, and showing a guessed number next to real measurements would be misleading. Tools that do this without saying so are selling you noise.

How priorities are calculated

Each finding has an impact rating (how much it hurts traffic, conversions or legal compliance: 1–3) and an effort rating (how hard it is to fix: 1–3). Priority = impact × 3 + (3 − effort). Critical ≥ 10, High ≥ 8, Medium ≥ 6, otherwise Low.

Findings with high impact and low effort get a “Quick win” badge and float to the top of “Fix these first”. The logic is deliberately simple enough to print in one sentence — a score you can't explain is a score you can't trust.

How scores are calculated

Each pillar starts at 100 points. Every finding subtracts a penalty by priority: Critical −28, High −16, Medium −8, Low −3. Scores floor at 0. The overall score is the plain average of the three pillars — they are equally important by design.

Colour bands: 0–39 red (needs urgent attention), 40–69 orange (room to improve), 70–100 green (in good shape).

Accessibility: what automation can and cannot see

We run the axe-core engine (the same one behind Lighthouse and most professional tools) plus our own static checks against WCAG 2.2. Important: automated tools can verify only roughly 30–50 % of WCAG criteria. Colour contrast, keyboard operability or caption quality need a human.

That's why the report shows an explicit manual checklist and why the conformance line says “estimated”. A green automated result is a good sign, not a certificate. The conformance estimate maps automated failures to WCAG levels: any level-A failure means “does not meet A”; only AA failures means “meets A, not AA”.

AI visibility: what the signals mean

The AI pillar measures what is directly observable: whether AI crawlers (GPTBot, ClaudeBot, PerplexityBot, Google-Extended, CCBot) may read your site, whether structured data describes your organization, whether the content has a citable structure (Q&A, lists, definitions, facts), and whether trust signals (about page, contacts, authorship) are present.

These are the levers you control. Whether an AI assistant actually mentions you also depends on your off-site authority — reviews, mentions, links — which a single-page scan cannot measure and which we therefore don't score.

llms.txt is reported as informational only. It is an experimental proposal, not an established standard, and we won't pretend otherwise.

Performance

When available, we use Google PageSpeed Insights (Lighthouse, mobile) — real lab measurements of Largest Contentful Paint, layout shift and total performance. If PSI is unavailable, we show only our own direct measurements (server response time, HTML size) and clearly say the full test didn't run. We never substitute an invented performance score.

Data & privacy

Reports are stored under an unguessable link so you can return to them and share them. They are excluded from search-engine indexing. Contact-form submissions are used solely to reply to your enquiry.