ibin
All articles

Automatic checks or a human audit? What each one finds

Automatic checks find the common problems fast, but people must judge the rest. What each one finds, how to combine them, and what an audit costs.

9 minute read

You ran an automatic accessibility check, fixed everything it reported, and the report now shows no problems. Is your website accessible? Probably not entirely. Automatic checks are fast, free or cheap, and very good at the most common problems. But a large part of what makes a site usable for people with disabilities depends on meaning and context, and software cannot judge that. This article explains what automatic checks can and cannot find, how to combine them with a few simple tests of your own, and when it is worth paying an expert for a full audit.

What the numbers say

The W3C, the organisation behind web standards, puts it plainly in its advice on choosing evaluation tools: tools cannot determine whether a site is accessible, they can only help. Several studies show where the line falls.

  • Deque’s coverage study (2021). Deque, a company that makes accessibility testing tools, looked at more than 2,000 audits covering over 13,000 pages and nearly 300,000 problems. On average, 57% of the problems were found by automatic tests. But those problems fell under only 16 of the 50 criteria of WCAG 2.1 level AA. The automatic share is high because a few common problems, above all low contrast, occur very often.
  • The UK government experiment (2017). The Government Digital Service built a deliberately bad page with 142 known barriers and ran it through 13 automatic tools. The best tool found 40%, the weakest only 13%. Tools have improved since, but many barriers on that page simply cannot be detected by software.
  • The Dutch government’s view. DigiToegankelijk, the Dutch government’s accessibility programme, says that more than 80% of the required criteria must be checked by a person, and that the results of an automatic tool alone are never enough to meet the law.

These numbers measure different things, so they do not contradict each other. Counted by number of problems, automatic checks find roughly a third to a half. Counted by WCAG criteria (the 50 testable rules of the Web Content Accessibility Guidelines), only a minority can be fully tested by software. Both lead to the same advice: start with an automatic check, then let people check the rest.

What automatic checks catch well

Some problems are a matter of measurement or presence: a value is too low, or something is missing. Software is excellent at those, and it never gets tired on page 150.

  • Low contrast. Text that is too light against its background. A tool calculates the contrast ratio exactly. See the guide on colour contrast.
  • Missing image descriptions. An image without alt text, the short written description that screen readers (software that reads the screen aloud) read out. See image descriptions.
  • Form fields without a label. A field where nothing tells assistive software what to type. See form labels.
  • Empty links and buttons. An icon-only link or button with no name, which a screen reader announces only as "link" or "button". See link names and button names.
  • Missing page language. Without it, a screen reader may read Dutch text with an English accent. See page language.
  • Missing page titles and skipped heading levels, which make a page harder to navigate.

These are also the problems found most often on real websites. WebAIM’s analysis of the top one million home pages finds the same handful at the top every year. That is why an automatic check is such a good first step: it clears away the bulk quickly.

What only people can judge

The other half is about meaning, behaviour and experience. A tool can see that something exists, but not whether it makes sense.

  • Whether an image description is right. A tool sees that an image has alt text. It cannot see that the text is a file name, or that it describes a chart without saying what the chart shows.
  • Whether the order makes sense. Content can look logical on screen while a screen reader or the Tab key goes through it in a confusing order.
  • Keyboard traps in custom parts. A home-made date picker, chat window or pop-up may catch the keyboard focus and never let go. Only pressing the keys reveals it.
  • Whether link texts and headings are clear. A tool can flag "click here", but not judge whether "Our approach" tells people where a link goes.
  • Clear language and understandable error messages. "Error 422" passes every automatic test.
  • Whether captions are accurate. A video with automatic captions passes, even when the captions turn "Ghent" into "gents".

Here is a typical case. Both images below pass an automatic check, because both have alt text:

<!-- Passes the automatic check, but tells nobody anything -->
<img src="IMG_2041.jpg" alt="IMG_2041.jpg" />

<!-- Passes too, and actually helps -->
<img src="IMG_2041.jpg" alt="Wheelchair ramp at the side entrance of the town hall" />

Only a person knows what the photo shows and why it is on the page.

How to combine them: a routine in four steps

You do not have to choose between an automatic check and an audit. They work best in sequence, each step making the next one cheaper.

  1. Run an automatic check on your most important pages: the home page, contact, your main form, your checkout or booking flow.
  2. Fix what it finds, starting with the problems that block people. Shared templates first, so one fix repairs many pages. Then check again.
  3. Do the quick manual tests below. They take less than an hour and need no special software.
  4. Book an expert audit when the stakes call for it.

Quick manual tests anyone can do

  • Keyboard. Put the mouse away and use only Tab, Shift+Tab, Enter, Space, the arrow keys and Escape. Can you reach everything, see where you are and finish your main task? Our 10-minute keyboard test walks you through it.
  • Zoom to 200%. Press Ctrl and + (Cmd and + on a Mac) until the browser shows 200%. Is all text still readable, without text cut off or overlapping? Then try a phone-sized window: does the page still work without scrolling sideways?
  • Screen reader basics. Use the free NVDA on Windows or the built-in VoiceOver on a Mac or iPhone. You do not need to master it. Listen to the start of the page, jump through the headings, and fill in one form. Do the headings make an outline? Does every field say what it is for?
  • Read it aloud. Read your main page as if you were new. Are the sentences short and the words familiar? Do link texts make sense on their own?

When you need a human audit

An audit is a systematic review by a specialist of all criteria on a representative sample of pages, usually with a report that explains each problem and how to fix it. You probably need one when:

  • you are a public body. Your accessibility statement must rest on a real evaluation. In the Netherlands, a full evaluation that follows WCAG-EM, the W3C’s evaluation method, is required for the two highest statuses. Our article on public sector websites explains the rules per country.
  • you run a large web shop or booking service under the European Accessibility Act, especially with a complex checkout or login.
  • you want a label that shows visitors an independent party checked the site.
  • you are launching a redesign, or a new platform that many sites will share.
  • people keep reporting problems you cannot find yourself.

For a small business with a simple brochure site, an automatic check, the manual tests and some care with content often take you most of the way.

What an audit costs and how to choose an auditor

Prices depend mostly on the number of different page types and on complexity. Two organisations publish indicative prices:

  • AnySurfer, a Belgian non-profit project of the organisation for blind and partially sighted people Licht en Liefde, lists audit prices of 1,250 euros for a small site, 2,500 euros for a standard one and 4,000 euros for a complex one, excluding VAT, with cheaper follow-up audits to confirm fixes. AnySurfer also awards a label, valid for at most two years.
  • BIK BITV-Test in Germany, run by DIAS, combines a base fee with an indicative price per page of 420 euros for simple pages up to 840 euros for complex ones, with complex web applications priced by effort. A test usually takes four to six weeks.

In the Netherlands, organisations such as Stichting Accessibility carry out WCAG-EM evaluations and lighter quick scans. In France, an audit follows the RGAA method on a set sample that includes the home page, contact, legal notice, the page listing the whole site, login and complete processes such as a form across several pages. It can be done by a trained internal team or an outside auditor.

When you compare offers, ask:

  1. Which method do you follow (WCAG-EM, RGAA, BITV-Test or EN 301 549), and which WCAG version and level?
  2. How do you choose the sample, and does it include forms, PDFs and pages behind a login?
  3. Do people test with a keyboard and a screen reader, or is it mostly tool output?
  4. Does the report explain each problem with an example and a fix?
  5. Is a retest included, and at what price?

Be wary of anyone who offers a quick fix through an overlay, a toolbar widget added to the page. Overlays do not repair the underlying problems.

How to prepare so the audit costs less

Auditors are paid for their time. Every minute spent on problems you could have found yourself is money not spent on the hard questions.

  • Fix the automatic findings first. Contrast, alt text and labels are cheap to find with a tool and expensive to have an expert write up.
  • Fix shared templates and components before the audit, not page by page after it.
  • Prepare a list of page types and key tasks, such as "request a permit" or "pay an invoice", so the sample covers what matters.
  • Arrange test accounts and access to pages behind a login in advance.
  • Ask your platform supplier for its own evaluation report. The Dutch government suggests splitting an evaluation into the technical platform and the content, so the platform is tested once and reused.
  • Name one contact person who can answer the auditor’s questions quickly.
  • Plan time and budget for the fixes and the retest. An audit without follow-up mostly produces a document.

Start with an automatic check

Check a page for free with ibin to clear away the common problems first. The report explains each problem in plain language, shows where it is with a close-up picture and links to a repair guide. The optional second look from AI points out image descriptions, link texts and keyboard use worth a closer look, as leads for a person to judge. Then do the manual tests, and bring in an expert when you need certainty.

See how your own site does

Check a page for free, without an account. You get a plain list of what to fix first, with pictures of where each problem is.

For example yourshop.com. One page takes about 20 seconds and it is free.

Keep reading

All articles