NOBODY is saying that automated tools are not useful, only that they cannot provide comprehensive coverage of WCAG 2.2 criteria.

You might think this feels like vindication, but it does not.

Sure, it’s nice to be validated by the FTC after 5 years of shouting to anyone who will listen that accessiBe is a net harm for the disability community. But the settlement is a pittance.

Reports put accessiBe’s 2024 revenue at $51.3 million. The settlement has a fine of $1 million. Subtracted from its 2024 revenue, that leaves accessiBe with $50.3 million dollars for the year. Dollars earned by lying to customers, misrepresenting itself to the community, and arguably harming disabled users.

According to the FTC, accessiBe not only fell short of its promises to clients, but engaged in misleading marketing. accessiBe “deceptively formatted third-party articles and reviews” to appear as if they were independent opinions by impartial authors, said the agency in a press release, and failed to reveal “material connections” to supposedly objective reviewers.

AI is not impacted by inaccessibility. It is not a disabled person. It cannot explain web accessibility from the perspective of a disabled person.

Removing the human factor of inaccessibility stories does not build empathy. It dehumanizes the stories. It dehumanizes US.

Additionally, inaccessibility is not a result of a lack of empathy. It’s a result of ableism. To still position lack of empathy as the main problem in almost 2025, is a failure to consider vital historical context.

Creating AI caricatures of disabled people does not help us dismantle systemic ableism.

I also take issue with the alleged need for comments to be “appealing” or “humorous.” Nothing appealing nor humorous about inaccessibility. Inaccessibility is PAINFUL in every single sense of the word. When disabled people encounter inaccessibility, we are harmed every. single. time.

Suddenly I knew exactly how Arthur Dent felt, dealing with the Nutrimatic Drinks Dispenser in The Hitchhiker's Guide to the Galaxy. Just like the cup of tea it dispensed to Arthur, the code provided by ChatGPT was, to paraphrase, almost, but not quite, entirely unlike the HTML for an accessible button.

Other times, however, Copilot is clearly just regurgitating irrelevant code samples that aren’t at all useful. Sometimes, it’s so far off base its suggestions are hilarious. (It regularly suggests that I start my components with about 25 nested divs, for example.)

But if we’re giving one of the world’s major corporations our money, in exchange for this tool that’s supposed to make us better…shouldn’t it be held to some standard of quality? Shouldn’t the results I get from a paid service at least be better than a bad StackOverflow suggestion that got down-voted to the bottom of the page (and which would probably come with additional comments and suggestions letting me know why it was ranked lower)?

As more and more of the internet is generated by LLMs, more and more of it will reinforce biases. Then more and more LLMs will consume that biased content, use it for their own training, and the cycle will accelerate exponentially.