What a machine can check, and what it cannot
Automated testing is fast, repeatable and free. It also cannot tell whether a sentence is true. Knowing exactly where the line falls is what stops a clean report becoming a false comfort.
Learn
Short answer
Automated accessibility testing checks machine readable facts: contrast ratios, missing labels, heading order, attribute validity. Manual testing covers judgment: whether alt text is accurate, whether the tab order makes sense, whether a page can be understood aloud. Both are necessary, and a clean automated report is the beginning of the work.
What automation is genuinely good at
Anything that is a fact about the markup. Whether an image has a text alternative. Whether a form field has a connected label. Whether heading levels skip.
Contrast is the clearest case, because it is arithmetic on two colors. A machine does it perfectly, in a moment, on every element of every page.
It is also tireless. It checks the four hundredth page as closely as the first, which no human reviewer manages.
It is also the only half that scales. A person can review ten pages carefully; a scanner reviews four hundred on every publish and does the four hundredth as carefully as the first.
Where it goes quiet
A machine can confirm alt text exists. It cannot confirm that smiling team in the office describes a photograph of an empty car park.
It can confirm a heading exists and cannot judge whether the heading describes the section. It can confirm a link has words and cannot tell that those words are click here.
Every one of those passes automated testing while failing the reader completely. This is the gap that a clean report hides.
The failure mode is particular and worth naming: these pass silently. A scanner reports nothing, the report is clean, and the page is unusable for the reader it was supposed to serve.
What manual testing actually involves
Less than people fear. Walk the page with the tab key. Turn on the screen reader your computer already has and listen to a page or two. Zoom the text and see what breaks.
Read your own alt text against the actual images. Fill in your own form wrongly on purpose and read the errors.
That is under an hour for a small site, and it finds the category of problem no scanner will ever report.
- Walk the page with the tab key and nothing else
- Listen to a page with the screen reader already on your computer
- Enlarge the text substantially and look for overlap or clipping
- Read your own alt text against the actual images
- Fill in a form wrongly on purpose and read the errors
- View a page in grayscale to find meaning carried by color alone
How much each one covers
Published estimates vary and all of them agree on the shape: automated testing finds a meaningful minority of the issues, and the majority need a person.
That is not a criticism of the tools. It reflects that many criteria are about meaning, and meaning is not machine readable.
The practical conclusion is to use automation for coverage and people for judgment, rather than expecting either to do the other's job.
Testing with actual users
The strongest test is watching someone who uses assistive technology every day try to complete a real task on your site.
It finds things neither automation nor your own careful review will, because a daily user has habits and shortcuts you do not know exist.
It is also the most work to arrange, which is why it belongs at the point where the obvious problems are already fixed.
Budget for it after the obvious faults are fixed. Watching somebody spend twenty minutes on a keyboard trap you could have found in a minute is an expensive way to learn about the keyboard trap.
Where each fits in a routine
Automation belongs in the build. Run it on every page before it publishes, so problems are caught while they are cheap and while you know what caused them.
Manual review belongs at milestones. A new template, a redesign, a twice yearly walk of the main paths through the site.
Every page written by Website Builder Studio passes a check that runs before anything publishes, validated against Google Search Essentials and modern web standards. The in-app assistant says which findings still need your eyes.
What a clean report entitles you to say
That the checks you ran passed. That is a real and useful statement, and it is narrower than it sounds.
It does not entitle you to claim conformance, certification, or that your site is accessible to everyone. Those are different claims resting on evidence a scanner does not produce.
Nothing here is legal advice. Describe your process honestly and you will never have to walk a claim back.
The wording of that claim matters more than it sounds, and it is covered in writing an accessibility statement.
Questions people ask
Is a free checker good enough?
For the automated half, the good free tools are genuinely capable. The limits are identical whether you pay or not: neither can judge meaning.
How often should I run automated checks?
Continuously, on every page before it publishes. Once a year tells you about pages that changed eleven months ago.
Do I need an expert?
Not for the basics. An expert is worth it for a complex application, a procurement report, or when you need an independent view you can show somebody.
What should I test first by hand?
The paths that earn money. Your contact form, your booking flow, your phone number. Fixing a barrier there matters more than a caption on a blog post.
Keep reading
See your website built from a conversation
15-day free trial. Card required. Cancel before day 15 and you pay nothing.