The room where this stopped being a checklist
A few years back we got deep into discovery on a website rebuild for one of the largest disability organizations on the planet. We did not win it. We were young, our portfolio was thin, and it showed.
I still think about the room.
There were a dozen or so of their people in it, and every one of them needed the site to behave differently. Screen readers. Volume modifiers. Captions. People who needed the thing to slow down, and people who needed it to get out of the way entirely. Nobody in that room was describing a compliance standard. They were describing the daily negotiation of using the internet in a body the internet was not built for.
I had shipped accessible-ish sites before that. Ran the scans, cleared the errors, felt fine about it. What I had never done was sit in front of the people on the other end and listen to what our industry's output actually feels like to use.
I have not been able to look at an audit report the same way since.
I filed accessibility under "ranking signals" too
A few weeks ago I published a list of the SEO fundamentals I think are worth caring about. Original content, recency signals, question-driven structure, hierarchical HTML, FAQ schema, and, near the bottom, sitting politely between backlinks and JSON-LD: "great accessibility (shoot for WCAG AA if you can)."
So before I go after anybody else for this, I have done it myself, on this blog, about three weeks ago.
And to be clear, the benefit is real. Accessible markup and good search performance genuinely overlap, and I am not about to argue that clean semantic HTML does nothing for you. What I am arguing is that it belongs at the end of the sentence, not the start of it. Because the order changes what you build.
Why does the motive matter if the work still gets done?
It does not get done. You get a different piece of work with the same name on it.
Two teams can both be told to fix the alt text on a product page. One team is told to do it because a blind customer cannot tell what they are buying. The other is told to do it because the images are an untapped keyword surface.
Both teams populate the attribute and both turn the scan green. The websites they hand back are not the same website.
Your alt text is lying to blind people
We inherited a site once where the alt text on a product photo read, roughly, "person enjoying the benefits of [product name] on a summer day."
The product was not in the photograph.
It was two people on a blanket, having a picnic. That is the whole content of the image, and that is what a person who cannot see it needs to be told. Instead they got a sentence about a product that was nowhere in the frame, and they got it again on the next image, and the one after that, because every image on the site had been fed the same keyword regardless of what was actually in it.
Léonie Watson is a blind accessibility engineer, director of TetraLogical, and sits on the W3C Advisory Board. In July 2026 she published a piece called "Accessibility is resistance!" It contains this line:
Oh, and, every. Product. Image. On. The. Page. Has. Exactly. The. Same. Damned. Text description!
I did not go looking for that quote to make a point. She wrote it, unprompted, about the exact artifact I have watched agencies produce on purpose.
Elsewhere in the same piece:
And frankly, I'm fed-up. I'm fed-up with not knowing what's in an image, fed-up with not being able to activate a link or a button, fed-up with date-pickers I can't even open, fed-up with having to ask for help, fed-up with not knowing, fed-up with the uncertainty, and I'm really, really fed-up with having to do the hard work because someone else hasn't.
WebAIM's 2026 analysis of a million home pages found that more than one in four images carry alternative text that is missing, questionable, or repetitive, where repetitive explicitly includes alt text identical to that of an adjacent image. WebAIM counts that as a defect. The plugin that generated it counted the same bytes as a job completed.
The tooling actively rewards it
You could write this off as a few cowboys, except it is built into the software a large chunk of the web runs on.
- Yoast will not give you a green light on a post with more than four images unless your focus keyphrase appears in between 30% and 70% of the alt text. Describe your images accurately without the keyword in them and the plugin marks you down.
- There is a WordPress plugin with over ten thousand active installs whose advertised sources for alt text include, word for word, the focus keyword from Yoast, the focus keyword from Rank Math, and the target keyword from SEOPress. It writes the text into the rendered HTML only, never the media library, so the person editing the site never sees the sentence being read aloud to their customers.
I have never rated Yoast. It has always struck me as software that manufactures technical-looking busywork so you feel productive, rather than pointing you at the handful of fundamentals that actually move anything. This is that instinct doing real damage to real people.
Google calls this spam
The funniest part is that the tactic does not even work on its own terms.
Google's own image documentation says to avoid filling alt attributes with keywords because "it results in a negative user experience and may cause your site to be seen as spam." They publish a worked example of what bad looks like: an alt attribute stuffed with thirty keyword variations of the word puppy. Their example of best practice is "Dalmatian puppy playing fetch." No keyword in it at all.
So the practice is bad for the user, and the search engine it is aimed at pleasing has documented it as a spam signal. It survives anyway, because nobody checks and the scan stays green.
What a crawler cannot see
The slice of accessibility that overlaps with SEO is small, and it is exactly the slice that has the least to do with disability.
Alt text, headings, semantic elements, descriptive links. A crawler benefits from all of that, which is why it makes the SEO listicles. Now consider what a crawler has no opinion whatsoever about:
- Whether focus is visible, and whether it goes somewhere sensible when a dialog opens
- Whether a keyboard user can get out of a menu once they are in it
- Whether an error message is announced, or just turns red
- Whether a form tells you what went wrong and how to fix it
- Whether a tap target is big enough to hit with imprecise motor control
- Whether the page still works at 400% zoom
- Whether animation respects a reduced-motion preference
Googlebot has no use for a single item on that list, which is where disabled people spend most of their time getting stuck. A budget approved on an SEO business case buys the first list and stops, because that is where the story it was sold on runs out.
That is my answer to the pragmatists, and it is a fair objection worth taking seriously: surely a cynical yes beats a principled no, and if "it helps rankings" is what gets accessibility funded, take the money and do the work. I would agree if the money bought the same work. It does not. It buys the crawler-visible subset, plus a compliance score that says you are finished.
An LLM is not a non-visual user
The 2026 version of this is worse, because it has learned to borrow the language.
accessiBe has published material describing large language models as "non-visual users". Valtech, a global agency, ran a section header reading "Accessibility debt is now GEO debt". AudioEye paid for sponsored space in Marketing Dive to file accessibility alongside AEO and GEO as one of "the 3 forces that will define 2026 marketing."
A language model is not a non-visual user. It is a program owned by a company, parsing your page to build a product. It does not get frustrated, and it has never once abandoned a checkout and gone to a competitor. Reaching for the moral weight of blindness to describe a scraper is a genuinely gross piece of rhetoric, and it is now the pitch.
Adrian Roselli, who has been doing this a very long time, put the technical version plainly in June 2026:
SEO also drove bad accessibility practices, such as alt text keyword stuffing, low-contrast text, verbose headings, link text built around keywords, gratuitous landing pages, and more.
He has also said he has already found keyword stuffing inside aria-label attributes. That one deserves a moment, because aria-label is the worst possible place to put a keyword. It overrides the visible label. It is invisible to sighted users and to the developer reviewing the page. The only humans who ever encounter it are screen reader users, which makes it a string written for machines and delivered exclusively to the people accessibility is supposedly for.
ARIA usage is climbing fast. WebAIM found labels and descriptions up 28% in a single year, averaging over 31 per home page. They also found that pages using ARIA average 59.1 errors against 42 on pages without it. They are careful to say ARIA is not necessarily causing those errors, since the pages carrying it tend to be more complex anyway. Fair enough. It is still hard to look at that trend and conclude anyone is getting more access out of it.
The compliance angle is a side effect too
Same logic applies to the legal pitch, which is the other thing accessibility gets sold on.
In Ontario, the AODA has required organizations with 50 or more employees to meet WCAG 2.0 Level AA on their public web content since January 2021, with financial penalties attached. The European Accessibility Act came into force on 28 June 2025. If you sell into either market, that is real.
But look what happened in April 2026. The US Department of Justice pushed the ADA Title II web deadlines back to 2027 and 2028. Mark Riccobono, president of the National Federation of the Blind, called it "a betrayal of the promise America made when it enacted the ADA."
If the deadline was your reason, your reason just moved. That is the problem with borrowing a motive. Rankings shift, rules slip, and the person trying to use your booking form is still sitting there.
Build it for them and the compliance takes care of itself. The fine you never pay is a nice outcome. It is not the point, and if you make it the point you will do the minimum that survives an audit, which is not the same thing as a site somebody can use.
We graded our own site a C
Cobbler's children, naturally.
We rebuilt papercrane.ca a couple of years ago, and because it was our own site it got the treatment your own site always gets. We did fine on the obvious things. We did not go anywhere near as deep on the granular stuff as we would have on a client build.
Then one of our own team members told us.
He does not have a disability in the way most people mean it, but using a mouse for any length of time is uncomfortable for him, so he works almost entirely by keyboard and he notices immediately when a site has not been built for that. He was not on the build. He was just using our website, and he was surprised that an agency talking about accessibility had not covered the basics.
I would not say we had missed the bare minimum. We had done a good job. We had not done a great job, and there is a real difference between those two.
So we ran a proper audit on ourselves, WCAG 2.2 AA, axe-core through Playwright plus manual keyboard testing. Every page came back a C. No page had a critical failure, and there were twelve findings across the site, most of them in shared components: a missing language attribute, around a dozen invisible menu links sitting in the keyboard tab order, unlabelled icon links, no skip link, dimmed text failing contrast at 3.71 against a 4.5 requirement. Fixed, re-tested, clean.
The part I keep coming back to is that our tooling did not catch it. A person did. Automated testing catches somewhere around half of real accessibility problems, by the most generous estimate available, which comes from a vendor measuring its own product. A clean scan is necessary. It has never been sufficient.
How to tell which one you are doing
You can check this on your own site this afternoon. None of it needs a specialist.
- Read your alt text out loud. Just read it. If it sounds like a description, you are fine. If it sounds like a search query, you have found the problem.
- Diff it. Sort the alt attributes on a page and look for duplicates. Two images with the same alt text means either one is decorative and should be empty, or somebody automated it.
- Grep your aria-label attributes. Any of them read like marketing copy? A product name that is not in the visible label? That string is being spoken to somebody.
- Unplug your mouse. Tab through your own navigation and try to reach your main call to action. Watch where focus goes when you open the menu. Watch whether you can get back out.
- Turn your monitor off and try to buy something. With a real screen reader, on your own site. It is the single most uncomfortable and most useful hour you will spend this quarter.
- Then run the scanner. Not first. It cannot tell you whether a sentence is true, only whether the attribute exists.
Build it for people
More than a quarter of Canadians aged 15 and over, around 8 million people, live with a disability, and that number went up 5 percentage points in the five years to 2022. Globally the World Health Organization puts it at 1.3 billion, one in six of us. That is a market, and most of the web serves it badly on purpose, because serving it well was never what the work was for.
Lucy Greco is a blind accessibility evangelist at UC Berkeley, and she wrote the line that should end every argument about legal risk:
Nobody turns on their computer every day thinking, I need to find a website to take to court. Instead, they turn to the computer to find a new wardrobe or the best price on groceries or a place to go on a date.
That is who is on the other side of your alt text. Somebody who wants to buy trousers.
So when a client asks whether this is going to help their SEO, I do not give them a hard time about it. It is a fair question and it is their money. I just put it back in the right order: we do this for the users, and the amazing benefit of that is it also happens to be great for SEO and GEO. That is the whole of how we approach accessibility, and it is baked into how we build sites in the first place, because retrofitting it later costs a fortune and never fits properly.
Build it and they will come. Which is a terrible quote to reach for, given the "they" in the original are dead baseball players who show up with ticket money, but you take my meaning.
