Meeting Minutes
Wendy Reid: Good morning.
So, what I wanted to do today was,
I think just continue the discussion about
Both the research side,
I know there's been, progress on that. I saw the thread.
and then also, talking a little bit more, I think, about the use cases and requirements.
side of things.
Just because, I mean, they're kind of intrinsically tied to each other. They
It helps to have use cases and requirements to know what are,
important areas of research.
But I don't know, Lisa, I missed the last meeting, and I wasn't sure if you wanted to give kind of, an overview of what happened.
Lisa Liskovoi: Sure, yeah, and I owe you the meeting notes, which I do have, so
I'll send those to you. I'm trying to remember what we had talked about, it feels like it was forever ago now. Ned, I don't know if do you remember anything you that we should
catch folks on?
Ned Zimmerman: I feel like I'm conflating the last meeting with the one before.
Let me think about that for a little bit longer.
If anything comes to mind, I'll jump back in.
Jutta Treviranus: I think we certainly, Mike brought up a conversation, not not
not this Mike, but the other Mike. A conversation about current authoring tools, and
the comparison between them and some of the
issues regarding, challenges of compliance with requirements. We talked about a little bit about
how to influence
Which authoring developers, and the difficulty in doing that.
And yeah, I'm trying to remember what else it was quite a lively conversation, Wendy.
And we got into, some of both the specifics, but also then the overall general
systemic issues in
encouraging certain features or certain things within authoring tools. Also, especially with AI.
the changing the quickly changing things that are happening, and and is there actually control over what the AI does in terms of accessibility, and how would that happen? Or that's what I'm recalling some of the conversation being.
Wendy Reid: That is, a really interesting point. I think especially with AI and
And, accessibility together. It's I've had this conversation, or I've had this conversation, and I've seen this conversation happening in other spaces where
everyone in the accessibility industry who's using AI right now
Like, everyone can acknowledge that
the current models?
that are out there, all
I'm not even gonna be, diplomatic. They all suck.
Add accessibility.
And what's really interesting is that everyone agrees that this is a problem, even people who, work, I think, at some companies that could actually have, some say in this.
And our current solution is to do things like
some of the benchmarking that's been happening, with, the, A11y LLM eval project and, AIMAC.
if people haven't haven't seen those.
I can try and provide links,
But
the, core problem is that
essentially, like
these AIs are built for a bunch of different use cases, but accessibility does not seem to be one of them.
And I wonder if, obviously this isn't really a problem for, a standards
group dissolve.
but it does seem like
there is an element of,
I mean, maybe this is something a standards group could solve, maybe not us, but
is is one of the requirements.
for conformance. Let's say, I'm an AI
I'm an AI tool vendor.
and I want to claim that my product is
accessible.
how might I do that?
And obviously, there's, my interface that should be accessible.
my output that should be accessible.
But I wonder if there's, something to be said for
that I also benchmark.
I also have an accessibility benchmark that I adhere to.
and can confirm conformance, confirm
success against.
And I and I share that information.
Wendy Reid: https://aimac.ai
Wendy Reid: https://microsoft.github.io/a11y-llm-eval-report/
Mike Masey: There's some in
It sounds like there could be some,
Again, whether this group is it's the right place for it, but, you know,
Saif Altalib: Can we talk to people with disabilities who vibe code or use IDEs, CMS authoring tools?
Mike Masey: There's a multitude of things that kind of
that question has to answer, because it's if there's an AI tooling that you want to use for benchmarking.
There's also, you know, what is
what is the what are the goals of how you're using that?
AI tool, is it to create
code for front-end, for back-end? Is it for creating content for something else. So, in terms of Benchmarking,
could this group be a good position for creating almost, some prompts?
that could be reused.
with the aim for, you know, creating seeing an example of how well this certain tooling can create
some accessible type of
web content, whether it be code or
or otherwise.
Because then you spent? I don't know, I don't know.
Ned Zimmerman: Something else that I've come across in doing a bit of a
research scan is
some,
some studies which have shown that
in order to get LLMs to generate accessible web content,
you have to be explicit about the accessibility and the prompting.
So you can't just ask them to create web content. You have to specify the accessibility requirements.
And so I think this is something that
it seems like will be really relevant for ATAG, because
the knowledge of what to ask for around accessibility requirements will have to be built into the prompting, at least with the way things are working currently.
and that's not necessarily something that everyone would know, and so that
like, there's that feedback loop of guidance and
remediation suggestions that are referred to in ATAG right now for how to fix accessibility issues.
But with the LLM,
field the way it is, it seems like there's also gonna need to be some expertise at a certain level by the people doing the prompting using the authoring tool.
In order to direct the LLM on how to build the content in an accessible manner to begin with.
I just yeah, which is which is, I think, germane to this as well.
Mike Masey: I've found the same thing Zed. It's one of those things where it can create accessible code / content, but the level of knowledge needed upfront and the amount of reviewing needed to get it to succeed probably outweighs using it
Ned Zimmerman:👍
Jutta Treviranus: And there's no consistent, actual following of that prompt, right? Which was, I think, another point that was made last week.
So it could be working, and then all of a sudden it won't.
Lisa Liskovoi: Yeah, yeah, no, I used to think you, I actually wanted to make a pretty much the same point, but in relation to one of the conversations we were having earlier, which is,
One of the things we have been talking about was that ATAC, in an ideal case, would create a kind of tool where somebody doesn't have to know everything about accessibility in order to create accessible content.
And that becomes so much more challenging in this situation where, A, you have to ask know exactly what to ask for,
But then, also, you have to be prepared that the tool is not going to follow those instructions consistently, so you have to verify that it followed those instructions.
And if you need to both ask and verify, that requires a pretty in-depth understanding of accessibility, so I think it gives us an
extra complicated challenge in terms of how do we
support the creation of a tool like that.
without relying on everybody having expertise in accessibility.
Saif Altalib: An agent skill maybe?
Wendy Reid: Replying to "An agent skill maybe?":
Possibly!
same body.
Sambhavi Chandrashekar: Yeah, so this is of deep interest to me, as you would know.
We in our elements, we do have
an AI component, as well as I'm always interested in being able to prompt and get
more accessible content, but like Lisa and others pointed out,
There has to be a dynamic
audit remediation sort of, you know.
Dynamic feedback chat going on, even about
what is still left to be made accessible, or could we have a
a comprehensive chat prompt set,
that we could build into the system.
to be able to get at least a bare minimum.
required accessibility built into anything that's generated.
I've been issued, like
When you start, you
set the parameters, right? This is what I want. I want you to take this persona, I want you to do this.
Can we tell over there?
I won't know headings and lists and whatever.
Ned Zimmerman: This doesn't this doesn't directly follow on your point, but I I think
Also, something interesting that I came across in the research I was doing was
The difference between
AI-based visual web content creation tools.
and ones that are more LLMs trained on code.
and I'll just quote from one paper I was looking at.
Which points out that
So, visual visual design tools are trained on existing visual UI designs.
But that doesn't include semantics, it doesn't include
ARIA roles, if they're applicable or anything like that.
And so if you're using a prompt-based tool that is
visually laying things out for you.
you're not going to get the same quality that
that an LLM trained on
code that might include accessibility-specific patterns and documentation would. And so there's a conflict there because a lot of the tools that are presenting themselves as, you know, describe what you want your website to look
it may give you visually
a response that matches what you're interested in.
But it might not have any
match with what's needed in terms of the accessibility and semantics of the underlying markup, whereas a code-based tool
could be better at that, depending on what
primitives it was trained on.
So that's a challenge, too, because a lot of the
no-code visual AI prompt-based tools are very focused on that visual appearance.
Wendy Reid: A project like this might be informative to us: https://github.com/Community-Access/accessibility-agents
Evelyn Wightman:👀
Saif Altalib:👍
Saif Altalib: Here is an example front-end ui skill: https://github.com/addyosmani/agent-skills/blob/main/skills/frontend-ui-engineering/SKILL.md
Morgan Murrah: I was going to say that the principle of sensible defaults is a way Id frame this prompt for accessibility issue. If someone has prefers reduced motion on for example, the model should defer to a sensible default of accessible content
Sambhavi Chandrashekar: I was I was thinking more in terms of
providing metadata, like
Not so much web page, but
course content, HTML-based course content.
Evelyn Wightman: Replying to "An agent skill maybe?":
I don’t know if an add-on is the right approach. It’s more tractable, but putting responsibility on the base model for accessible output is stronger.
Wendy Reid: There's, two streams happening at the same time, but I do think,
something interesting to
look at, or
I guess it's kind of an interesting thing, is
Saif mentioned skills, and, this is something that
is being leveraged in a lot of places to kind of
create these, hyper-specific
tasks that you can feed into agents and models
to do very specific things. So, I shared a link in the chat to a project by Taylor Arndt,
called, Accessibility Agents, and I recommend looking at it, because there are a few actually related to document production.
and I mean, this could be
a possible, output for us is, a series of
recommended agents or something, you know, you know, take feel free to take these and do with them what you will.
but this is the place to start.
I think an interesting thing, though, we'll have to consider as well, is
There are,
especially for accessible document production.
You might be so there's kind of,
I think the agreed-upon
accessibility baseline.
Let's say we'll just say WCAG,
meet WCAG, right? You know, your things are labeled, things are good color contrast, all of that.
But then, especially in the excess in the in the world of accessible document production,
There's often highly specialized work happening on
like, optimizing documents for
different, disability needs.
And I wonder if one of the challenges that we face is,
just having, if tools almost go too hard into, automating,
For the baseline,
that it becomes difficult to
optimized, because the tool would constantly be oh, why are you doing
Doing that, you should, you know, especially if in a case where, say, something is less optimal for one,
type of, requirement.
But more optimal for another.
I guess, case in point would be, if you had one
let's say it's, something related to cognitive
accessibility, and so it suggested that using icon images
would make content more understandable, but you're working on a document for someone who is blind
and uses a screen reader, and icon images just would not serve that
purpose, it's much better to put it in text.
Would you end up in a weird fight with your agent, because it's trying to insert icons, and you're like, no, no, no, I just want
I just want text.
that's, I think some of the attention that we might
might have to consider.
Latasha Willis: How about when adding titles to hyperlinks? It seems like it would be pretty straightforward for an LLM to add it if they just grab the H1 from the page it's beign linked to, but how hard do you think training it would be?
Morgan Murrah: Im kind of new around the block of some recent stuff, is MCP the technology where a recommended agent stuff happens. Sensible defaults should apply there too I would think, even if its not my thing directly to use
Sambhavi Chandrashekar: But Wendy, would you say, I just want text? Why would we not go with both?
Saif Altalib: Here is an accessibility checklist md file from Addy Osmani: https://github.com/addyosmani/agent-skills/blob/main/references/accessibility-checklist.md
Wendy Reid: Well, I mean, this is,
I'm I'm noodling more than anything.
Mike Masey: And
So, with ATAG itself, would it make
in terms of the actual documentation that, you know, we'll be updating or tweaking, I know at the moment, we're not focusing on big overhauls or big changes, but
some of you may have started noticing in documentation for other various bits of software, or whatever your
APIs. I started to integrate things like MPC, MCP servers into their documentation.
Is that something that could potentially be utilized for things like ATAG?
to help provide that sort of underlying
connection to whatever type of AI. There's so many different types that they're using are going to be coming out, and sort of giving that underlying
sort of framework for how to get information out of a tag into whatever AI tooling there is.
that could be
sort of used per purpose, because every every, sort of,
Often tools can have slightly different
implementation, right? But it could help
Rather than having one set of baseline rules that tries to cover everything.
It might make it more possible to
please also be a bit more peaceful with it and get just what they need.
I might be going the wrong direction, though, I don't know.
I put it out there.
Wendy Reid: I don't think you're going the wrong direction. I kind of had a similar thought as well
is a theoretical deliverable, something like ATAG MCP.
Ned Zimmerman: Replying to "Here is an accessibility checklist md file from Ad...":
Important to keep this in mind: https://www.deque.com/blog/not-a-checklist-building-accessibility-compliance-into-your-business-processes/
Saif Altalib:❤️
Saif Altalib: Here is an MCP tool that evaluates sites for accessibility: https://github.com/JustasMonkev/mcp-accessibility-scanner
Mike Masey: Yeah.
Jutta Treviranus: Yeah, and we started, we were talking about what did we talk about last week? one of the the
areas that it may be somewhat
Well, it gets us into controversial ground, but also the blurb between a user agent
the end, the
authoring tool, and
the type of personalization that is unfortunately, uh
the basis of, or used, or exploited by,
overlay systems, so
given the huge diversity of what someone might want in terms of how the content
is authored or produced or presented. Is there a possibility of
going further again into
the a user defined
presentation of
the content
and compliance not at the at the page level,
but on the site level, sorry.
Saif Altalib: Replying to "Here is an accessibility checklist md file from Ad...":
It would have to be an accessbility expert that uses these tools properly.
Wendy Reid: This is something that I have
an interesting amount of experience with, because this is a conversation that we have in EPUB world a lot.
Around the line between
the creator intent and user intent
in content display.
Because a lot of
EPUB reading systems have
many, it's almost
a future look into what, theoretically, the web might look like, because most reading systems have options to change the font, to change font size, font spacing,
color, uh
layout, even,
even when a publisher has,
set very clear boundaries.
they get overwritten. And we're even talking about this in the sense of, some of the most strict content out there, which was what we something we call fixed layout.
you know, is it possible? Is there a future where it's possible to break fixed layout down into just text and just images?
and let people consume it as they wish.
But that that concept has not made it as far into, the rest of the web.
Latasha Willis: Yes, and some EPUBs will let you change the background color as well.
Wendy Reid:💯
Jutta Treviranus: So, this actually takes us back to the history of the origins of the web, because
as you as you may or may not know,
there was SGML, then there was XML, and, of course, EPUB came from
SGML and XML, and in choosing what would go into HTML, there was a
Well, I mean, this was a debate back in the 80s, when the HTML
was we were debating what should go into HTML, and at that time, it was right once present
in many, many different formats, which was the SGML. So you write once, but you publish and present in
many, many, many different formats. And if you look at Yuri Rabinski's work,
It, was talking about the possibility of that, and a view to
the web that would act like that, where you would
have a specific content, but it could be reimagined or re-represented in many different ways.
Morgan Murrah: CSS Forced Colors is an interesting one around this area https://developer.mozilla.org/en-US/docs/Web/CSS/Reference/At-rules/@media/forced-colors
Jutta Treviranus: Yeah, and CS Morgan CSS I mean, when we first started discussing CSS, the idea was that there would also be a personal style sheet that you could so if you have
TrueCSS, and it works as it should, then there would be the possibility of having a personal style sheet that you could apply to the content.
as a separate, independent thing.
Morgan Murrah: Yeah, I think Firefox has a limited functionality that allows you to do this.
If I remember right.
Jutta Treviranus: Yeah, unfortunately, a lot of the I mean, CSS
yeah, the restrictions and the,
control of presentation with branding and things of that nature.
started to whittle away at that independence of styling
Wendy Reid: And I mean, I guess there's a part of this conversation that is also, sometimes
Sometimes things like the format or the structure, or the layout
carry meaning as well, and so,
to blow it up is to actually harm the content in some way, even though it's really hard, we're having this discussion in EPUB a lot lately.
Because it is, when is
layout essential.
Like, how do we identify that, the book has to look this way for a reason?
Versus, I just want it to look this way because it's pretty.
Latasha Willis: Replying to "Here is an accessibility checklist md file from Ad...":
Thanks. I just sent this to our web developer. He's currently playing whack-a-mole trying to correct ADA issues with widgets, etc.
Sambhavi Chandrashekar: The experience is cognitive, right?
There are two sites of the same coin, content and presentation.
They build and iterate on each other, so we cannot deny that connection.
Miriam Fukushima: So, for ATAG, we
have to decide in what way we can be prescriptive and descriptive, and in
what way we have to, decide
if we want ATAG to be
well, a bit responsible in shaping it in one or the other direction.
And, thinking if ATAG actually can shape
The direction of how the web goes.
And, if not, then,
we still have to account for
for ATAG, because it's gonna be
should last for some years, that we, account for both tenancies.
That,
Yeah.
Jutta Treviranus: Yeah, be pramatic and have a systemic view of what will be happening.
That's a good point.
Wendy Reid: I guess to take it back to kind of the core thing is,
If we think about this in the con
context of
use cases and requirements.
a use case
I guess, like I described earlier,
I, as a content author,
want to create accessible content.
for a broad audience, and I want my authoring tools to help me do that.
But then the, there's almost a mirrored
user use case of, I, as the
consumer of content want to
adapt it fully to my needs.
Morgan Murrah: I think Providing multiple formats is a good idea too kind of related from WSGs https://www.w3.org/TR/web-sustainability-guidelines/#offer-suitable-alternatives-for-every-format-used
Wendy Reid:
But I need I need the content to be structured in a way that allows me to do that.
What's the what's the requirement that lives between those two?
Yes, providing multiple formats is, uh
It's a good, uh I didn't realize I was in the web sustainability guidelines.
Such a
Morgan Murrah: Yeah, it's got some nice success criteria, I don't know, it's just nice
crisp. It's been edited a lot. Some of these, uh
guidelines, so they're they're kind of much more shortened.
kind of easier to digest.
Like him. Yeah.
Mike Gifford (CivicActions): So nice to have a consolidated edit in the process of creating them.
Wendy Reid: I saw a question in the chat earlier, Saif, about talking to people with disabilities who vibe code or use IDEs, CMS authoring tools. I think that's definitely
should be part of our research discussion.
Morgan Murrah: I have one account that I thought was kind of interesting to share. Basically, there is a man named Jose,
who I saw, talk do a talk at, Linux Fest Northwest.
And he is a blind programmer who uses Visual Studio Code.
And he describes the accessibility features as actually really good,
He just has to operate in a sort of iterative,
Line by line fashion,
And also use, version control really strongly, because
Charles Hall: Providing multiple formats is less important than providing a format that does not remove user agency to change that format.
Lisa Liskovoi, Morgan Murrah:💯
Morgan Murrah: it's not so easy to spot if you move a lot down a big file.
And you make several changes, and then you, you know, just like any programmer, that's a problem for everyone.
If you get into programming at some point, you have that problem, but
yeah, uh
that was just a really interesting, yeah, I got a bunch of commentary about Linux distributions that have problems with
for programming for him, and he's a big Linux
person.
Mike Gifford (CivicActions): I mean, just in terms of discussion around AI, and part of part of the concern that the
web sustainability guidelines group is discussing right now is that the
Anytime you're trying to involve AI in the, to document what is what to put into a guideline, um,
It's just moving too quickly, and I think that they're
There's a challenge with what we can possibly document or do research in, given the fact that
there's so many different ways to implement this, and there's so many different ways that people are using this, and that there's often terms that are mixed up with hype that are
not well understood. So I think that there's there's a even just a
the confusion between, looking at what is AI versus what is automation. there's still people who don't understand the difference between those two.
or if you use AI to create an automation, is that still the same thing as using in AI? It's well,
No, but people are getting confused about what that means, so I think that we're
We need to be able to I would I would
I don't know that we that AI is going to wait until we have this answered for ATAG.
So I wonder if there's a way to
to to somehow, put the AI discussion into a way that does not
stall the development of a of updates to ATAG so that it could actually be something that we can start
using, or encouraging people to use in policymaking and other implementation.
Wendy Reid: That's a good question. I'm trying to think of an example of this, because you're right, it is moving so quickly.
Mike Gifford (CivicActions): So, how do you I mean, given that that
there even, what is, a gen
AI toolkit, or swarms of agents. these are things that people are exploring. They'll be going to a discussion on Monday about the use of open claw, and how people are experimenting with this type of
very much self
organized AI, I just
because the rate of changes is moving so quickly, I think that there's a danger in trying to go off and
have a definitive answer to something that is just that we're never going to catch up to.
Wendy Reid: I mean, I think that's why focusing on the use cases and the requirements and, the outcomes is probably going to serve us a lot better than trying to say,
To do this, use X
thing.
Saif Altalib: You got to.
Jutta Treviranus: Yeah, there's absolutely no way we can be technically prescriptive. It would
break within the day, probably.
Saif Altalib: To Mike's point, I think that's a great way to look at it,
Maybe, you know, what we focus on
is we create the guidelines, and these all these AI tools use use them as, you know,
knowledge or connectors to our guidelines, so
We're not building the skills or the MCP tools and things, but those tools and skills are using our guidelines
they ingest our guidelines into
Into the tool. Or skill.
Morgan Murrah: one interface and format with user agency promotes more inclusion, especially in websites, but also users might choose specific formats, was thinking of the epub example like it would be great to have the original and the remixed version
Ned Zimmerman: one thing that I came across in some research over the last couple weeks that I thought was really interesting was
something that LLMs could potentially be very helpful with.
which the
I can drop a link into this, but
there
there's, uh
someone was describing this as
So there's kind of differences between accessibility issues that can be, identified by automated tools, like Axe or something like that, where, you know, is are alt tags missing?
but then another level is
Are alt tags provided, but not meaningful?
Right? So, you could have something that passes those automated tests by including alt tags, but if all the alt tags just say image, that's not
a meaningful that's still not actually accessible.
And the authors of this particular paper had suggested that that's actually somewhere where LLMs could be really useful, because
their the ability to,
read and interpret the page
through an LLM could help identify
those cement so the people who wrote this paper described the difference between
a syntactic violation, so something where the syntax is wrong, that's a
machine-detectable violation, and a semantic accessibility issue, where
the meaning is there, the information is there, but it doesn't make sense.
and so I think this is, when we start to think about how
AI tools could help provide guidance for improving accessibility. That's an interesting area, that is
Ned Zimmerman: https://dl.acm.org/doi/full/10.1145/3772363.3799364
Mike Gifford (CivicActions): I gave a talk on this at FOSDEM a while back:
https://archive.fosdem.org/2025/schedule/event/fosdem-2025-4709-alternative-text-for-images-how-bad-are-our-alt-text-anyway-/
Ned Zimmerman: Oh, cool, that's great.
Yeah.
Anyway, I thought that was neat, because it's a really good use case for where these tools could actually be really helpful.
In ways that are hard to deal with.
in an automatic way right now.
Sambhavi Chandrashekar: So, Ned,
I have a I think I should bring in here.
Alt text, yes, it's absolutely one of the blockers.
Very important, but I'm just thinking
In our company, we have a severity rubric for WCAG.
for prioritizing defect management.
That also prioritizes human I mean, it goes by human impact.
All that having been said, 1.1.1, 2.1, we'll figure on top. However, I'm just wondering, for the purpose of our
exercise. Should we prioritize in a different way?
Meaning, Arya, for example, or
things that bring in metadata, but are
harder for
people in the normal course.
to be able to bring in.
If there is a way we could figure that out,
And prioritize in that order, we would actually be facilitating.
authoring.
not whole human impact, yes, but
More importantly, what would be the
higher-hanging fruit.
that we could make easier to pick.
Mike Gifford (CivicActions): AI Generated tool on alt text scanning https://mgifford.github.io/alt-text-scan/
Mike Masey: Might there be a way that we could take an approach to provide information / guidelines around ATAG and AI tools in a way that is agnostic to the actual tools?
Latasha Willis, Wendy Reid, Saif Altalib, Lisa Liskovoi:👍
Ned Zimmerman: Like, the hardest things to get right without expert knowledge, right.
Sambhavi Chandrashekar: Which requires expertise.
Also, the ones that are purely
I mean, I would have human automated,
All text definitely is something that requires human intervention.
But things that don't require the first 60-70 person that you can mechanically do.
But requires technical expertise.
If we could build those in, that would be, uh
like, hitting more number of
Or more percentage of accessibility.
I mean, others might have.
alternative, different thoughts that
This just came into my head, so I wanted to share.
Jutta Treviranus: This actually brings up which AI, right? I mean, we're Mike brought up the point of automation versus an LLM would be
very terrible at consistent results and things of that nature, but,
Are we looking back to roles-based and other
AI systems as well, and how much are they actually I mean, if we're looking at authoring tools,
What kinds of AI are being implemented in the authoring tools?
Mike Gifford (CivicActions): I suspect people are worse than AI when it comes to consistency.
Sambhavi Chandrashekar: And again, my brain keeps going back to
There are websites, and there's learning content.
Is there a distinction at all, or isn't there?
Jutta Treviranus: in the original, ATAG, we covered
curriculas well, if it was presented online on the web, but
And LLMs were definitely part of
the the list of authoring tools.
Mike Masey: Thanks. I was wondering how how it sounds like there's
The thing that's almost blocking us a little bit at the moment is just there's too many
too many small details all over the place that
it's hard to consolidate that.
what do we need to do to, sort of, grab all those different
sort of branches, and sort of almost looked down the trunk of that tree, if you'll use that as a metaphor, and find, sort of,
those core different ways that people might use
AI and authoring tools. As a starting point,
So, whether it's for, you know, an education platform, whether it's for a website, whether it's for,
product, and any other there's all these different ways, and whether it's for
code creation, whether it's for content creation,
because those are,
sort of fairly low-level categories of how people might use it, and it gets more detailed from there.
And trying to work out what those are.
I think could help us then figure out how to communicate the information to the people who need it.
Latasha Willis: Replying to "https://dl.acm.org/doi/full/10.1145/3772363.379936...":
Great paper. When it comes to checking for accessibility, you have to dig in and see what's there versus just not getting errors.
Sambhavi Chandrashekar: Right.
So, Lisa, as part of your research, are you planning an annotated bibliography or literature review, or something of
research done so far. I think Abhinav will be interested. He's a PhD student from UMD.
We connected through ATAG, and we're pretty
closely in touch now.
So, some thorough research, so you identify the map and
Go from there.
Jutta Treviranus: We have a few lit reviews and annotated bibliographies of
AI and accessibility, but not specific to well, Lisa can explain
Sambhavi Chandrashekar: Because he's he I just concluded his project with Adobe, but he did some work on their Express
authoring to making it more accessible automatically, and he's working with Professor Jonathan Lazar.
Recently, we supported Carlton University in bringing Dr. Lazar as one of the chairs for a chair, CERC.
chair for bond accessible Documents.
Lisa Liskovoi: We are currently in the process of putting together a literature review specifically to those topics, so I think AI and authoring, and Ned, I don't know if you want to speak about that a little bit more.
Jutta Treviranus: Jonathan is is, uh
focusing largely on
Not so much on web documents, but documents in general. PDF documents and things of that nature, yeah.
Wendy Reid: I mean, I think
we have to we also have to keep in mind that,
basically all authoring nowadays has some kind of web component.
you know, people people create things through online, web-based tools, people publish things online.
So, even if it's a PDF created on your desktop, it ends up on the web at some point.
videos, audio
you know, TikTok's, they're all
kind of fed through the system.
I think in talking about, these different
these different things. What might actually be really helpful for us is to understand
Maybe I can put this out into the into the group, because I don't
necessarily work on, one of these platforms.
is,
is, basically
what do the different, workflows look like for different types of content production?
So, you know, if I am a developer,
What does my workflow look like when I'm
using different tools to create code to write code.
What does
what does it look like if I am a marketer posting ads to my website? What if what does it look like if I'm a social media influencer?
And I'm creating content for Instagram.
having all of those
workflows
And then either,
I think we can, hopefully have the have
have the, workflows annotated with, the moments where AI or LLMs or automation come into play.
Morgan Murrah: I can speak about a Headed CMS and its limitations
Sambhavi Chandrashekar:👍
Sambhavi Chandrashekar: Replying to "I suspect people are worse than AI when it comes t...":
Worse? Inconsistency is the essential trait that makes us irreproducible and facilitates our survival.
Wendy Reid: Where and then it's okay, for this type of content production, we see
this being inserted here.
how do we
How do we account for that?
Because there is, so many different kinds of content production.
There's so many different workflows, there's so many different tools,
but there's probably, I would imagine some commonality in certain phases where,
LLMs creep in.
or creep in, or, are the are, take center stage.
As, the start of the process.
or a key part of the process.
Jutta Treviranus: Yeah, I think what given the space that we're working in, it will need to be a sort of high-level process
values-based set of guidance, so not the
pressure to so the compliance will be determined by the outcome, not by the
by the actual process of getting there, and the outcome will need to be something that is
much more high level than
what has been possible within WCAG?
Lisa Liskovoi: Hey, I have a bit of a suggestion, or maybe just something I would love to float by the group, because I feel like sometimes the it just feels like
too big, and we're trying to think about everything all at once.
And I'm wondering if it would be helpful to actually go through ATAG 2.0 and go through the different sections and requirements and think, okay, how does AI fit into this specific requirement? How does
AI changed this particular aspect of authoring?
Wendy Reid: I like that idea.
Sambhavi Chandrashekar: And I and I want to put out a radically crazy
idea. As AI democratizes.
Content creation, like somebody, Lisa, or someone was no, Wendy was saying, anybody creates
web content.
How about democratizing accessibility?
Instead of making it more normative.
I don't even know what it means.
I don't even know what I mean as I'm talking, but
Morgan Murrah: I could speak just to a quick thought that it made me have, that was an interesting thought, is just that
I work in a CMS at work that is very limited. Basically,
For example, there's all these components and these features that basically I have no influence over, I cannot change.
and I don't really have the power to do that.
I don't know, very optimistically, maybe websites and CMSs are getting better to the point that
you don't have these split roles.
Where you have very limited function control, to me, democracy would be it sounds like ownership, control, being able to feedback into the system.
that maybe, that something should be more accessible than it currently is.
I'm rambling now, but yeah,
Basically, I just feel like I have very limited control. I have a
job role, which is very limited right now.
But I see accessibility things all the time. I have to deal with screenshots, uh
Just yesterday, there was a screenshot with an outline on it, because it was an autofocused you know, tab key, outline, and I had to emphasize that it should remain. I'm not going to remove this outline, it's just a
thing in the screenshots, it's actually should be there. And that was, my first big win for a long time.
basically because I, yeah, I work as a content author
in a system called Adobe Experience Manager, and I could talk a lot about it, but basically, that's what I have to live within, and
yeah, how do we get those
How do we get more feedback from
the content authors, into the systems that they were going to be working in at work and their jobs.
that's my thought.
Jutta Treviranus: https://raisingthefloor.org/our-approach-accessforall/
Latasha Willis: I can relate to the CMS issue. The one we use is mainly for universities with a lot of dotNet customizations.
Jutta Treviranus: https://link.springer.com/chapter/10.1007/3-540-45491-8_123#page-1
Sambhavi Chandrashekar: Replying to "https://link.springer.com/chapter/10.1007/3-540-45...":
Thank you, Jutta!
Sambhavi Chandrashekar: Yeah, no, I was just thinking, optimization instead of
conformancer.
Everybody having the freedom to optimize, and very radically again.
As part of the IAAP, we are so much against overlays.
But I also cannot stop thinking, is overlay the solution?
After 8 years of WebAIM discovery that we're still lingering around 95-96% inaccessible websites.
Should the user be given the power?
to define
Charles Hall: AEM. The top that broke the link element on billions of instances by including the alt attribute which is invalid.
Morgan Murrah:😵
Wendy Reid: This is
This is something
I have wanted us
us, and I say us as in, the accessibility community,
to have, a much more nuanced discussion about.
Because
I think there are elements
of the overlay feature set.
that hold value.
But they're not
the way that they're currently implemented and the overlay model,
is the problem. The features themselves, what it afford what
overlays are currently affording are not the problem. things like being able to add
to adjust font size, or to have focus.
Like, you know, reading things for dyslexia or for ADHD, or, you know, reduce motion and all that. Those features are super valuable.
The problem is that
overlays, there's no, universality, right?
you get one overlay on one website, you can set your settings, you go to a different website,
Maybe they have another they have a different company's overlay. Your settings do not carry over. You've got to do it all over again.
Third website, no overlay.
If anything, these should be built into the browser.
Sambhavi Chandrashekar: More importantly
Morgan Murrah: Like watching TV and having to set captions on for every channel instead of one button
Wendy Reid, Latasha Willis:💯
Sambhavi Chandrashekar: Yeah
More importantly, the same political economy that squashed ATAG is trying to color it.
my line at, and make it more of a
lucrative business.
If we could isolate the benefits and be able to turn it around, I think
democratizing accessibility would be giving the power to every user to define their UI.
Including interactions.
then you don't have to depend on that system that's still not
mending itself.
Mike Gifford (CivicActions): I wrote about this here:
https://almanac.httparchive.org/en/2024/accessibility#user-personalization-widgets-and-overlay-remediation
We don’t talk about it like:
User personalization
Automated overlay remediation
Custom overlay remediation
Jutta Treviranus: Yeah, and of course, this we did this well before overlays, appropriated personalization or individualization. I think Ned is going to speak to this.
Ned Zimmerman: Yeah, and I was just gonna say, I mean, I think I think the thing that I find particularly frustrating about
the way that overlays are presented is that, um,
it's sort of this is the thing that means you don't have to build an accessible website, right?
Jutta Treviranus: Exactly.
Ned Zimmerman: just slap this on top, and you don't need to actually worry about
Knowing how to make your website accessible out of the box.
And what that does is it, as you said, Wendy, there's lots of things that they offer, and that increasingly, user agents offer. I mean,
TextZoom is one that
I remember reading a statistic from, 6 or 7 years ago,
that said more people
more people use
text zoom in their native browser, like the plus or minus keys,
then use Internet Explorer.
And yet, we're still building sites to support Internet Explorer,
But we're still using overlays to enable TechZoom. What makes sense this does not make any sense.
Mike Gifford (CivicActions): Also, on the Overlay Fact Sheet:
https://github.com/karlgroves/overlayfactsheet/issues/36#issuecomment-801263275
Ned Zimmerman: So, yeah, so it's those are features that are valuable,
But the way that they're being marketed is
add these features in our automatic remediation, and you don't need to know how to make an accessible website. And it throws out ends up
there's a bit of a baby in bathwater situation, where it would be great if we could separate and disentangle those things and have personalization be
its own toolset that's not, the
Snake oil, we can just fix your inaccessible website.
product.
Sambhavi Chandrashekar: Yeah, I think Utah, you get to get back there.
The first thing you started was with a floppy, with the USB stick, right?
Jutta Treviranus: Oh, yeah, Web4All. we had so, this was a portable preference file you carried on a
Bell calling card or USB stick, or a SunRing, Java Ring, and you could then modify not just the
the web content, but also
the user interface for the computer, yes, or multi-user workstations.
Charles Hall: Democratization is not measurable.
It is an intent or ideal.
example: vibe coding is not democratizing web content creation because even savvy people (like me) cannot use these tools and processes. It still excludes people.
Miriam Fukushima, Morgan Murrah:👍
Jutta Treviranus: And there, we were very, very clear that you had to it worked the best if you had attended to web accessibility. That was one of the biggest provisos that this is not
it will not trans the site will not transform
Unless you have an accessible site.
Sambhavi Chandrashekar: So, many of us are from the industry. If we can succeed in
disentangling the overlays from the commercial.
vent that it has right now, and reclaiming that into a tag, I think that'll be amazing personalization.
Ned Zimmerman: Yeah, I just I just saw Charles' comment in the chat about
democratization being measurable, and I think it's an interesting point.
And what what you said, Charles, about vibe coding,
not actually democratizing web content creation?
is a good point, and also to be
to zero in on that a little bit more, something that came up in our the lit review that we're working on,
was a post by a
blind user who basically said,
I'm trying to use Claude code to make
software that works for me,
But cloud code is not accessible to me.
or maybe it wasn't clawed code, maybe it was the Claude desktop application or the Claude website.
But I think it's also interesting to think about the challenges that arise when
the LLM tools that we're talking about don't themselves
like, their own interfaces are not accessible.
and so you have this weird situation where this person is,
Latasha Willis: That's definitely a problem
Ned Zimmerman: I can create software that works the way I need
I need it as a not
you know, expert developer using these tools, but the tools themselves pose major barriers for me.
so I just
Jutta Treviranus: Which is part A of ATAG, right? Yeah.
Ned Zimmerman: Exactly, yeah, yeah.
Charles Hall: add a little more context to, the the reason that
that example applies to me.
is is not
what we think of as lack of accessibility of the tool itself via its interface.
It's cognitive, it's understanding.
The challenge that I have
is, I used to be a web developer.
I can write raw HTML and CSS and even JavaScript to some degree.
What I can't do is any of that through a layer of abstraction.
So, all of the tools of the modern web are not accessible to me because the abstraction doesn't make sense.
Sambhavi Chandrashekar: Replying to "Democratization is not measurable. It is an intent...":
Philosophies are drivers. They cannot create systems. But they can orient us.
Jutta Treviranus: the level of adaptability can be measured
Wendy Reid: I longed for the days when I could just write an HTML file in Notepad.
And upload it.
via my FTP, and it just worked.
Charles Hall: Same. Same.
Wendy Reid: And I didn't need, Netlify or something.
This is why I don't currently have a website.
Lisa, I think you raise an excellent point that, we're we're chipping away at an iceberg, but using,
the tiniest fork in the world.
that's a weird visual metaphor.
So I think starting with the next meeting,
what we'll do is we'll
start looking at ATAG 2.0 section by section.
sort of, requirement by requirement.
And we'll look at, okay, does this need to change? Does this need to expand?
and as we do each section, we can also decide, are there is there anything missing?
but we'll start from start from the first and, you know, see how many we get through. So, come prepared, next meeting, having
reviewed at least Part A, because I I highly doubt we get, through more than a few, but, you know.
having an understanding of Part A will probably make it a lot easier.
Morgan Murrah: https://en.wikipedia.org/wiki/Folksonomy - this is an example of something that I would describe as a democratizing thing, tagging systems that allow users to make their own things. Maybe models can be useful for this
