How Juno checks a website

From address to report

1. 01

   Capture

   A real Chromium browser opens the site on a desktop and a phone-sized screen, follows the main navigation to a few key pages and records the page, the rendered styles, accessibility tree, load performance (Lighthouse) and screenshots.

2. 02

   Measure

   The measured checks run on that capture: contrast ratios, text sizes and line lengths, tap target sizes, link labels, page titles, structured data, Core Web Vitals and more. The same capture gives the same result every time; only load times vary a little from one measurement to the next.

3. 03

   Judge

   Checks that need judgement (is the offer clear in five seconds, does the copy answer objections, can a test visitor finish a task) are judged by AI against the written criteria, three separate times. The median verdict counts; when the runs disagree, the check is flagged for a person instead of guessed.

4. 04

   Write

   Findings are turned into plain language with a concrete fix, ordered by impact and effort, and each one names the standard or research it rests on.

Six areas

The checks are grouped into six areas a visitor would recognise. The full list of checks, their thresholds and weights are Juno’s own work and not published; each finding in a report shows the check and its source.

• UX11

  Can people find their way?

  6 measured5 judged by AI

  • Do menu labels say where they lead?
  • Is it clear what happened after a click?
  • Can people get back from a wrong turn?

  Based onNielsen Norman Group · Apple · Baymard Institute

• UI13

  Does it look calm and trustworthy?

  9 measured4 judged by AI

  • Is body text big enough and the lines short enough to read?
  • Do type, colour and spacing follow one system?
  • Does motion help, or just distract?

  Based onNielsen Norman Group · Apple · Refactoring UI · Vercel

• Copy12

  Does it say the right thing, simply?

  6 measured6 judged by AI

  • Does the first screen say what you offer, and for whom?
  • Do buttons and links say what happens next?
  • Is it written for the visitor, in plain words?

  Based onNielsen Norman Group · Mailchimp · plainlanguage.gov · GOV.UK

• Accessibility12

  Can everyone use it?

  11 measured1 judged by AI

  • Is all text readable against its background (WCAG 1.4.3)?
  • Does everything work with a keyboard?
  • Are buttons and links big enough to tap?

  Based onW3C · Nielsen Norman Group · Deque axe-core · Apple

• SEO10

  Can the right people find it?

  8 measured2 judged by AI

  • Can Google crawl and index the pages?
  • Do titles fit in search results?
  • Does it load fast enough on a phone (Core Web Vitals)?

  Based onGoogle · Nielsen Norman Group

• Conversion10

  Do visitors get in touch?

  4 measured6 judged by AI

  • Is there one clear next step?
  • How long is the way from interest to enquiry?
  • Does anything make people hesitate?

  Based onBaymard Institute · WiderFunnel LIFT · Nielsen Norman Group · BJ Fogg

How the score is calculated

• Each check passes (1), partly passes (½) or fails (0), weighted by how much it matters to visitors.
• An area’s score is the weighted share of points earned, out of 100. The overall score weighs the six areas by the kind of site (a shop weighs conversion more than a portfolio).
• Blockers cap the score: any critical failure caps its area at 59, and a failed WCAG Level A check caps accessibility at 69. A site can’t look good on paper while it shuts people out.
• Checks that don’t apply to a site are left out rather than counted as passed. In the free quick check, an area with too few verified checks is shown as “not checked yet” instead of getting a score.

Standards and research

The standards and research the checks cite. Links go to the original publication.

What Juno can and can’t do