How we score your site
The model behind every Accessrank scan, report and client summary
Every Accessrank report leads with one number: the accessibility score. It tells you how much of the web accessibility standard (WCAG 2.2, the rulebook ADA claims are measured against) your store meets, and nothing else is mixed into it. Two further numbers, SEO and AI readiness, sit beside it. This page explains what each one measures, how we work it out, and what none of them can tell you.
What do the three numbers mean?
A sample store after fixes, so you can see what a real result looks like. The numbers are illustrative; your results will differ.
The headline number. How much of the accessibility standard the store meets, with the most serious requirements counting for more. 100 means every requirement we test is met.
Fourteen things a search engine looks for on a page, from the page title to the product facts written in for machines. Reported beside the accessibility score, never mixed into it.
Eight checks on whether AI assistants and shopping agents can read the page and use it, from the buttons to the page title. Also reported on its own.
100 means every check we run passes, including the ones a person reviews. It does not guarantee conformance or protect against a claim.
Why they stay separate. The number a plaintiff's lawyer or an insurer cares about is the accessibility result. A combined figure could look healthy while real accessibility failures remain, or look poor when the only problem is a missing page description. Keeping the three apart means each one answers one question.
How is the accessibility score calculated?
We open every page in a real web browser (the same engine as Chrome, running on our servers without a screen), so the page is fully loaded exactly as a shopper would see it before we test it. We then run axe-core, the open-source testing engine most accessibility tools are built on. It checks the page against the requirements of WCAG 2.2, the standard. A requirement counts as met only if every check for it passes everywhere on the page; one failure anywhere and that requirement is not met. Each requirement then carries a weight, because some matter more than others.
Swipe sideways to see every column.
| Requirement level | Weight | Example |
|---|---|---|
| Level A (the basics every site must meet) | 3 | Every image has a text description 1.1.1 |
| Level AA (the level ADA claims cite) | 2 | Text is readable against its background 1.4.3 |
| Best practice (recommended, not required) | 1 | Extra checks that are good practice but not part of the standard |
The score is then calculated as
Worked example
The same sample store, before and after the fixes.
Swipe sideways to see every column.
| Step | Before the fixes | After fixes |
|---|---|---|
| Requirements checked | 31 Level A × 3 + 24 Level AA × 2 = 141 | 141 |
| Requirements not met | 8 Level A × 3 + 7 Level AA × 2 = 38 | 0 |
| Requirements met | 141 − 38 = 103 | 141 − 0 = 141 |
| Accessibility score | 103 ÷ 141 × 100 = 73 | 141 ÷ 141 × 100 = 100 |
The sample store had 76 individual problems behind those 15 unmet requirements before the fixes. The problem count tells you the size of the fix; the score tells you how close the store is to meeting the standard. Both are reported, and they are never mixed.
Why we score by requirement, not by how many things are wrong
If a product page has five images with no text description, it fails the requirement that every image have one, full stop. It does not matter that the same page has two thousand other elements that pass unrelated checks; those do not buy back the failure, because a shopper using a screen reader who lands on any of those five images still hits a wall. Scoring by counting elements would let a big page with a handful of serious failures, buried among thousands of trivial passes, report a flattering score. Scoring by requirement keeps it honest: it answers "how many of the things the standard actually asks for does this page get right?"
We still report how many elements each failure affects, as a separate figure. Five broken buttons repeated on every product page is a bigger job than one broken button on a single landing page, even though both count as one unmet requirement. Use the score to see how close a page is to the standard; use the counts to size the work.
Checks that need a person to decide
A few checks cannot be settled by software alone. Whether an image description actually describes the image, rather than just existing, needs someone to look at it. Where the automated test cannot decide, we count that requirement as half met rather than guessing either way, and flag it for a person to review as part of the fix.
Which versions of the standard we test against
The tests cover WCAG versions 2.0, 2.1 and 2.2 at Levels A and AA, which together make up the current standard, plus a set of best-practice checks that are widely recommended but not part of the standard itself. Best-practice checks carry the lowest weight. For the technically minded: axe-core runs with the wcag2a, wcag2aa, wcag21a, wcag21aa, wcag22aa and best-practice rule sets.
How is the SEO score calculated?
A checklist of 14 things search engines look for on a page, combined into a score out of 100. Each is checked as present and correct or missing, and weighted by how much it usually affects whether a page is understood by Google and how it looks when shared.
- Page title
- Meta description
- Single H1
- Correctly ordered headings
- Image alt-text coverage
- One official address per page (canonical link)
- Page language declared
- Works on phones (viewport tag)
- Product facts written in for machines (schema)
- Preview image and text when shared (Open Graph)
- HTTPS
- Page allowed in search results
- Descriptive link text
- Favicon
Several of these overlap with accessibility: image descriptions, heading order, link text and page language count in both scores, because screen readers and search engines read the same structure. That is why SEO tends to improve once a site is fixed for accessibility, though the reverse is not always true.
How is AI agent readiness calculated?
Eight checks on whether AI assistants and shopping agents can read a page and act on it. A check is skipped when it does not apply (no forms on the page, say), so a page is never marked down for something it does not have. The score is the share of checks that pass.
- Buttons have names agents can read
- Links say where they go
- Form fields are labeled
- Clear main region and heading structure
- Links can be followed by crawlers
- Search and AI crawlers are allowed in
- Descriptive page title and summary
- Page language is declared
For clients we add a fourth measure alongside these: an AI visibility baseline. We ask five AI assistants for brands in the client's category and record how many name the client. It is a count, not a score, re-run at 30 and 60 days, and it is never folded into any of the others.
What does a full audit test?
The free check and the monitoring scans are automated, which is why they can run in seconds and on a schedule. A paid engagement adds a person. At onboarding, and again in each remediation cycle, an Accessrank staff member works through the paths a shopper actually takes, landing on the store, browsing a collection, choosing a variant, adding to the cart and reaching checkout, using only a keyboard and then a screen reader.
That pass covers what software cannot decide: whether an image description says something useful, whether the order you tab through a page makes sense, whether a popup can be left, whether an error message tells you what to fix. What it finds is fixed alongside the automated findings and recorded in the same report. The conformance summary and VPAT you receive describe the pages assessed, at the time they were assessed, which is why monitoring continues after the fix.
What don't the scores tell you?
Read this as carefully as your score.
Automated testing, ours included, is generally accepted to catch between a third and a half of real-world accessibility barriers. It cannot judge whether an image description is meaningful, only whether it exists; whether the order you tab through a page makes sense, only whether each item can be reached; whether an error message is helpful, only whether one appears. Saying a site meets the standard is saying every applicable requirement is met, and backing that up takes a person testing with a screen reader and a keyboard. That human testing is part of every Accessrank fix, and it is why a perfect automated score is a milestone, not a finish line.
A good Accessrank score is a strong, evidence-based signal that a site's code is in good shape. It is not, on its own, a legal guarantee of ADA or WCAG conformance, and we do not represent it as one. For the legal context this score sits in, see our ADA guide and our Disclosure.