Guide

How the Websight accessibility score works

A score out of 100 is only useful if you can see how it was made. Here is the whole formula, three worked examples, and the list of things the number does not know.

By Luke Rallings. Published 26 October 2026. About 7 minutes to read. The method described here is the one in the extension's code and is printed in every report.

Harbour Bakery, home page. Many issues to fix

1 critical issue type on 4 elements, 4 serious issue types on 9 elements. 10 rules passed. 1 check needs a person. Fixing the top three would take it to 85.

The formula

  1. Start at 100.
  2. Each type of issue the scan finds takes points off, set by its impact: critical 10, serious 7, moderate 4, minor 2. The impact levels are the ones assigned by the axe-core engine's rules, not a judgement made by Websight.
  3. An issue that appears on many elements takes more. Each extra element adds 10 percent to that issue's penalty, up to 10 extra elements, so the most any one issue type can cost is double its base: 20, 14, 8 or 4.
  4. The result is kept between 0 and 100 and rounded to a whole number.

Written as a formula, for one page:

penalty(issue) = weight(impact) × (1 + 0.1 × min(elements − 1, 10))
score = clamp(0, 100, round(100 − Σ penalty(issue)))

That is the entire method. There is no hidden weighting for page type, no bonus for passed rules, and no adjustment for site category.

Three worked examples

One critical issue, one element

A page with a single image missing alt text. Critical, 1 element: penalty 10 × (1 + 0) = 10. Score 90. "Few automated issues", and the one issue is listed first.

The same issue on twenty images

Same rule, 20 elements. Extra elements are capped at 10, so penalty 10 × (1 + 1.0) = 20. Score 80. The cap exists so that one rule on a long gallery page does not wipe out the score and hide the other findings. The element count is still shown in full in the issue card and the report.

A typical small-business home page

Example penalties for a page with four issue types
Issue typeImpactElementsPenalty
Images must have alternative textCritical410 × 1.3 = 13
Elements must meet minimum colour contrastSerious37 × 1.2 = 8.4
Links must have discernible textSerious37 × 1.2 = 8.4
Heading levels should only increase by oneModerate14 × 1.0 = 4
Total1133.8, so 100 − 33.8 rounds to 66

Score 66, which is "Many issues to fix" (the next band starts at 70). Fix the first three and the projected score is 96: the "Fix these first" box in the results screen does exactly this sum for the three issues whose removal raises the score most.

The four bands

The band is a label for the number, chosen so that the words match what a developer would say looking at the list.

  • 90 to 100Few automated issues. Nothing or almost nothing failing. The manual checks are now the whole job.
  • 70 to 89Some issues to fix. A handful of issue types, usually fixable in a day.
  • 40 to 69Many issues to fix. Several critical or serious types, often from the template.
  • 0 to 39Significant issues to fix. The page is likely hard or impossible to use for some people right now.

Multi-page audits

Each page is scored on its own. The audit score is the rounded average of the page scores, and the report shows both, so one bad page stands out instead of disappearing into a mean. Issues that appear on more than one page are listed once as site-wide, because they almost always come from a shared template and are fixed in one place.

What is not in the number

This is the part that matters most, and it is printed in every report.

  • Checks the engine could not decide. axe-core reports some results as "incomplete": it saw something that might be a problem but cannot be sure without a person (text over a background image, for example). These are listed as needing review and take no points off.
  • Everything an automated tool cannot test. Whether alt text is meaningful rather than merely present. Whether the whole page works with a keyboard in a sensible order. Whether focus is always visible. How the page sounds in a screen reader. Whether captions are accurate. Whether the content is clear. Websight ships a 25-point manual checklist for these, and the score does not pretend to cover them.
  • Pages and states the scan did not see. Dialogs that were closed, error states that were not triggered, pages behind a login you did not open.
  • Any legal standard. A high score does not mean a page meets WCAG, the European Accessibility Act, the Equality Act 2010 or any other law, and a low score does not mean it breaks one.

So what is it for? Comparing scans of the same page or site over time, and deciding what to fix first. Scan, fix the top three, scan again, send the client the difference. Used that way it is honest and useful. Used as a pass mark it is neither.

Why a score at all

Because the alternative is a list of forty items with no order. Developers and clients both asked for a number they could track, and the common complaint about free automated checkers is that they give results without a summary. The score is that summary, with its method one click away on the results screen, so nobody has to take it on trust.

Questions

Is a score of 100 the same as WCAG compliant?

No. A score of 100 means the automated engine found no failing rules on that page at that moment. Automated tools find only part of the accessibility problems; keyboard use, meaningful alt text, screen reader behaviour and many other criteria still need a person to check. The score is a way to compare scans over time, not a conformance claim.

Why does one issue cost more than 10 points?

Because an issue that appears on many elements is more work and affects more of the page. Each extra element adds 10 percent to that issue's penalty, up to a maximum of double: a critical issue on 11 or more elements costs 20 points instead of 10.

How is a multi-page audit scored?

Each page is scored on its own with the same formula, and the audit score is the rounded average of the page scores. The report shows both, so a single bad page is visible rather than hidden in the average.

Do checks that need a person affect the score?

No. The engine reports some checks as incomplete, meaning it could not decide. These are listed separately as needing a person to review and take no points off, so the score never pretends to know something it does not.

See it on your own page

The scan, the score and the "fix these first" projection are free in the Websight Chrome extension, with no account. The PDF report and multi-page audits are in Pro, a paid upgrade with a free trial.