One Tool, and What It Missed

One tool, and what it is actually for

There is a free accessibility checker called WAVE. It is made by WebAIM, it runs as a browser extension or on a page you point it at by web address, and it needs no account.

The thing that made it click for me is that it does not hand you a report somewhere off to the side. It draws its findings directly onto your page — small marks sitting on the actual elements they refer to. You are not reading a list about your site. You are looking at your site with the problems labelled in place.

I ran it over two sites I own, one after the other, on the same afternoon.

The first was this one. It came back with a single error — a form field with no label attached to it — and nothing else. No warnings, no contrast problems.

The second was a dashboard I built for a different project. One error again, a different one this time: a page title that does not say anything useful. But alongside it, eleven warnings. Skipped heading levels, and a run of things the tool wanted a human to look at rather than deciding for itself.

Two sites I built the same way, in the same habits, and they failed differently. Not worse or better — differently. That is the part I did not expect, and it is the reason the rest of this post is about what a checker can and cannot see rather than about my score.

What a checker actually tests

Automated checkers do not test whether your site is accessible. They test whether specific, machine-detectable conditions are present. Those are different jobs, and the second one is much narrower than it sounds.

Take image descriptions. A tool can reliably check whether a description exists. No tool can check whether the description is any good. An image labelled image1.jpg passes the existence test and fails the actual purpose completely, and there is no automated way to catch that — you have to look.

So every checker has to draw a line between definitely a problem, possibly a problem, and cannot tell from here. WAVE draws that line in a particular place and is open about where. Some things it reports as errors. Some it raises as things a human needs to judge — which is what most of those eleven warnings were. That second category is not the tool being unhelpful. It is the tool showing you the edge of what it can see.

WebAIM say as much themselves, and they are the ones publishing the largest annual survey of this. Detected problems are a subset of real problems, not a proxy for them. Real barriers are more common than detected ones, not less.

I find that clarifying rather than discouraging. A checker is a smoke alarm, not a building inspection. A silent smoke alarm is good news and it is not a certificate.

Which is also why I would rather learn one tool properly than collect five. Knowing exactly where a single instrument stops being reliable is worth more than a drawer of instruments you half-trust.

The six things, and where the number comes from

WebAIM publishes an annual survey of the top million home pages. In the February 2026 edition, they detected accessibility failures on 95.9 percent of home pages, averaging 56.1 errors per page. That figure went up rather than down, which was the first worsening in six years.

The part I find more useful than the headline: a small number of error types account for the large majority of what gets detected. Low contrast text. Missing image descriptions. Unlabelled form fields. Empty links. Empty buttons. A missing page language setting.

Six categories. Not dozens.

Both of the errors I actually found sit in that list, which is the ordinary outcome rather than a coincidence.

That the survey and the tool come from the same people matters here. WAVE is WebAIM's, and so is the research. The evidence and the instrument match, which is not something I could say if I were quoting one organisation's statistic while pointing you at somebody else's software.

Two things I will not do with that figure. I will not tell you it describes small businesses — it describes the top million home pages, which is the web at large, and small quiet sites are arguably excluded from that sample rather than represented by it. And I will not turn 95.9 percent into a reason to panic. It is a measurement of how common a detectable problem is, not a measurement of how broken the web is for people.

What I did about it, which is not the same as fixing it

Here is where I will be straight with you, because the tidy version of this post would end with me fixing something and showing you the green tick.

I have not fixed them yet. What I did instead was go through every mark the tool left on every page, write each one down with what it was and where it sat, and turn that into an ordered plan — what to change, in what order, and which items need a person to look rather than a rule to be applied.

That sounds like less of an achievement than a fix. I have come round to thinking it is the harder half. A checker hands you a pile of items with no sense of which ones matter, and the actual work — the last mile of it — is deciding what is a real barrier for a person, what is the tool being cautious, and what order to do any of it in. The list is not the plan. Turning the first into the second is the job.

In my experience most of these end up being template-level fixes — one change in one file, not a rebuild. That is my own working assessment from doing this kind of work, not a research finding, and I have not yet proved it on these particular items. No study I know of measures remediation effort, and I would rather label the claim than dress it up.

The part that stays with me is not technical. When people build things, they build for themselves, and building for yourself quietly means building for people like yourself. I did not miss that form label because a rule told me to care about it. I missed it because I could see perfectly well what the field was for, and it did not occur to me that seeing it was the assumption.

That is what I mean when I say accessibility is a practical place to start thinking about ethics in building things. It is not abstract and it does not require a position on anything. It is a short list of small corrections, each one of which is the same admission: I built this for someone shaped like me, and I can widen it a little.

If you want to try it

Run it on one page you care about. It is free, it takes a few minutes, and the marks land on your own site rather than in a report you have to interpret.

Then read what it links to. Every item WAVE flags comes with an explanation, and following those is how the tool stops being a thing that judges you and starts being a thing that teaches you.

The sources of truth are WebAIM and the W3C, and I would rather you read them directly than take my summary. That is the last of these posts, and it is the habit I most want to leave you with — go to the thing itself.

If you run it and you cannot tell whether something it flagged actually matters, that is the normal outcome and it is the reason I wrote this. Text me at 503-664-0546 or email eog@ernestofgaia.xyz and I will read it with you. That part is free.

The majority of things that are easy to fix are usually very little and boring. So if you do not want to do the boring stuff and you want to pay me to do it, I will do it. That is the whole offer, and I would rather state it plainly than build a funnel around it.

#ernestGoesToAI