OpenAI has announced it’s launched a new browser, Atlas, with ChatGPT built in. For those familiar with ARIA, OpenAI outlines what to expect (I left the code as I found it, other than removing the target):

We’ll continue to make Atlas better, and our roadmap includes multi-profile support, improved developer tools, and ways Apps SDK developers can increase discoverability of their apps in Atlas. Website owners can also add ARIA ⁠(opens in a new window)tags to improve how ChatGPT agent works for their websites in Atlas.

That “ARIA (opens in a new window)tags [sic]” link points to this on the Developer FAQs page (I removed the heading and target but left the <b>):

What can I do to improve my website performance with ChatGPT agent in Atlas?

Making your website more accessible helps ChatGPT Agent in Atlas understand it better.

ChatGPT Atlas uses ARIA tags—the same labels and roles that support screen readers—to interpret page structure and interactive elements. To improve compatibility, follow WAI-ARIA best practices by adding descriptive roles, labels, and states to interactive elements like buttons, menus, and forms. This helps ChatGPT recognize what each element does and interact with your site more accurately.

What This Means

ChatGPT Atlas logo, a scalloped blue circle that looks like an anus with a rounded arrow-shaped butt plug, and two fists pulling it open at the edges. It’s not my fault LLM companies broadly chose buttholes as logos, and I didn’t tell OpenAI to use an arrow to plug its new browser, pejoratively known as Atlass.

There are some signals in here that accessibility practitioners can decode. They’re relatively straightforward:

  • Slurping / stealing content from div-soup React (et al) sites is hard when semantics and structure are ignored;
  • OpenAI sees ARIA as metadata which can guide its bots;
  • OpenAI wants authors to use ARIA to make it easier to classify and structure the data it slurps / steals;
  • Open AI doesn’t understand ARIA conceptually.

That last bullet requires a bit more context. But more advanced accessibility practitioners can tease some other signals from the noise:

Accessibility practitioners who’ve bothered to check in on the ChatGPT interfaces over the years have demonstrated the OpenAI team largely fails to understand HTML, let alone any aspect of accessibility. Instead, these feel like vibe-coded Vercel failures (I know that’s redundant), as Steve points out three times:

That’s a three year history of accessibility incompetence from the OpenAI team. From the same company asking authors to use ARIA to better slurp / steal their content.

What This Will Do

The web is already mostly or nearly mostly LLM-generated content. With this massive displacement of human content, sites are even more motivated to find tricks to stand out.

Professional SEO snakeoil shills are always looking for a new angle and way to set themselves apart. Pointing to OpenAI’s new poorly-informed request for ARIA is one lever they can pull. They’ve already been pulling on the ChatGPT search engine lever and are likely to wrap this in the same cloak.

SEO hacks have been keyword-stuffing alt text in images for literal decades. There is no way ARIA will get away unscathed. Every button, link, and div will be roled up to redundancy at best, and the wrong thing at worst.

Any ARIA attribute that accepts human words could be doomed. I expect to find an aria-roledescription="FREE no-cost free cheap free mouthwash lids" in my future, because I’ve already seen keyword stuffing in aria-label and now OpenAI is asking for it (even if it’s too daft to comprehend that).

SEO hacks have already done a great deal of damage to the web, including pushing keyword-laden spam sites meant to drive traffic. LLMs super-charged those efforts. Now the LLM makers themselves are finding newer ways to ruin the web.

What Can You Do

Maybe acknowledge the majority view on “AI” where you work, ignore ARIA for ARIA’s sake to stymie the content thievery, and then personally dismiss the privacy-breaching content-stealing OpenAI browser.

These are just my reactive thoughts. Maybe you have better ideas to protect your content, users, and the web.