Last week at the CAKE Conference in Kraków, I found myself surrounded by brilliant minds tackling the intersection of content and AI.
For those of you who don’t know, CAKE is a new conference in Kraków, Poland organized by a vibrant content and writing community that I’ve been working with for years (and a cornerstone to my design thinking study abroad).
It is truly one of the most exciting and welcoming writing communities out there! Come see us next year.
(Kraków is also one of the best cities in Europe!)
While many sessions in this conference reinforced skills I’d seen at technical communication conferences, one theme kept surfacing that I wasn’t exactly expecting: accessibility.
What struck me wasn’t just that accessibility appeared in so many sessions, but how it revealed something fundamental about our relationship with AI that often gets missed.
Technical writers aren’t just writers, we are custodians of information experiences. When you approach content design with accessibility as a guiding principle, you create better information experiences across the board.
We’re approaching AI from the wrong angle entirely if our only focus is on efficiency, productivity, or even ethics broadly speaking.
The Accessibility Revelation
Rather than treating accessibility as a specialized concern or compliance requirement, these presentations helped me think of it as a fundamental lens for understanding how content works—for everyone.
Three speakers approached accessibility from different angles: Anna Dulny-Leszczyńska focused on structural principles for assistive technologies, Sara Grądziel explored automation and AI integration, and Wojtek Kutyła challenged us to see accessibility as inherently political and human.
For me, they demonstrated why accessibility thinking should be central to any content strategy, especially one involving AI.
Structure and Semantics
Anna, a product designer at App Synergy Pros, opened her presentation with a powerful demonstration using the viral dress photo from a few years where different people saw different colors. She asked the audience whether they saw it as white and gold or blue and black. This illustrates how we all process visual information differently. But this kind of diversity extends far beyond perception to language, ways of thinking, and physical abilities.
She outlined three core principles for accessible content design.
Structure: Make content logical for assistive technologies. She noted that 71% of screen reader users navigate by headings, jumping from heading to heading to decide what content to consume. This requires descriptive, unique headings, proper use of lists, and avoiding text embedded in images.
Understandability: Ensure everyone shares the same comprehension by using terms familiar to your specific audience and following plain language guidelines.
Operability: Make actions predictable and clear. She showed how Figma’s “lock/unlock” button creates confusion because users can’t predict what will happen when they click it.
What’s striking is how these same principles also help machines work with content at a scale … Or at least help us think about ways our content can work with diverse machines, not just humans.
Automation Layers
Sara Grądziel’s session posed an equally compelling question: Can accessibility become as simple as spell check? She framed automation as a paradigm shift for accessibility, where we now need to understand how machines play a role in developing accessible content.
Sara outlined three layers of accessibility work.
Automation layer: This involves code checks like alt text presence, color contrast ratios, form label verification, HTML validity, and heading structure. Prime targets for AI agents.
AI-Assisted layer: This requires human collaboration to evaluate alt text accuracy and relevance, discover patterns of repeated issues, assess link text clarity, check heading semantics, and evaluate form error handling. This is where you need human-machine collaboration.
Human-only layer: This layer is about checking tone of auto-generated content, testing keyboard and screen reader navigation, assessing content clarity and cognitive load, and experiencing real assistive technology use. Best leave AI out of this, at least with current technologies.
She emphasized that automation isn’t just about checking stuff, but needs to include practical implementation guidance, contextual review of issues, AI-generated content suggestions, and continuous monitoring rather than one-time reports.
Also known as “human-in-the-loop.” So, yeah, accessibily-first thinking will be key to keeping your job in an AI economy.
The Human Dimension
Wojtek Kutyła’s presentation on accessible web content took a strikingly different approach—personal, political, and unapologetically human.
“Language is ours—and texts are political.”
In other words, content choices have real consequences for people’s lives, and writers have the power to harm, exclude, or help through their language choices. I would argue the same when developing AI content systems.
Wojtek emphasized that content must be easy to understand for actual users, not directors or executives. He compared overly formal insurance language (”Acts of God! Insurrection!”) with clearer alternatives (”natural disasters, war, or civil unrest”). He asked content professionals to make content engaging and human, use headings and sections to break up text, test with real people, and remember that just because you think its good, doesn’t mean it is.
He advocated for an intersectional perspective, acknowledging that we all exist in a matrix of privilege and vulnerability, and our writing choices reflect that. For me, this is also true of AI.
Why This Matters for AI
What struck me most about these three presentations was a realization they didn’t all explicitly state but collectively demonstrated: the same structural principles that help screen reader users also create exactly the kind of systematic, well-organized information that AI systems process most effectively.
When you design content with proper heading structure, logical hierarchy, and descriptive labels for human assistive technologies, you’re simultaneously optimizing for AI retrieval systems.
The semantic richness that makes content accessible to humans makes it processable by machines. And Wojtek’s emphasis on human, political, context-aware writing reminds us that accessibility thinking goes beyond technical compliance to fundamental questions about how we communicate with and respect people.
Technical writers aren’t just writers, we are custodians of information experiences. When you approach content design with accessibility as a guiding principle, you create better information experiences across the board.
This insight connects to something I’ve been exploring in my work on rhetorical theory—the ancient Greek concept of metis, or cunning intelligence. Metis represents the ability to adapt, improvise, and find clever solutions to complex problems. It’s not just theoretical knowledge or technical skill, but practical wisdom applied creatively to overcome constraints.
Accessibility-first thinking is metis in action.
Accessibility as Strategic Intelligence
When content creators approach accessibility strategically rather than as mere compliance, they develop a form of cunning intelligence that naturally improves all aspects of content design.
Consider how accessibility principles mirror the foundational elements of effective AI collaboration:
Clear structure helps everyone. When you organize content with proper headings, consistent navigation, and logical information hierarchy for screen readers, you’re also creating the kind of well-structured information that AI systems can process most effectively.
Plain language serves multiple audiences. Writing that works for users with cognitive differences also works better for non-native speakers, tired readers, and AI systems trying to extract meaning from your content.
Alternative formats expand reach. Creating content in multiple modalities—text, audio, visual—doesn’t just serve users with different abilities; it provides AI systems with richer, more contextual information to work with.
What the CAKE speakers demonstrated repeatedly was that accessibility-first thinking creates content that performs better across every metric that matters: user engagement, search visibility, translation quality, and yes, AI effectiveness.
Now, some might point out a technical limitation here: LLMs process text by breaking it down into tokens and calculating positional relationships between them. In this process, semantic structures like heading tags or ARIA labels get tokenized just like regular text. The model doesn’t inherently “understand” that an <h1> tag signals something different from a <p> tag the way a screen reader does.
This is technically accurate—but it misses the bigger picture about how accessibility thinking shapes AI content systems.
Accessibility-first design influences AI effectiveness at multiple levels beyond just text generation. When you structure content with clear information hierarchy, you improve how retrieval systems find and rank information before any LLM even starts generating text.
When you write in plain language and chunk information logically, you provide better context windows for models to work with.
When you create multiple modalities and build in redundancy, you enable more robust content systems that can adapt to different uses and users.
The real value is in how accessibility principles shape content architecture, information workflows, and system design, all of which influence how well AI tools retrieve, process, and work with information at every stage, not just during generation.
Many AI-assisted content systems rely on structured data, metadata, and semantic layers for retrieval, routing, and orchestration. These elements remain crucial even if they get tokenized away during the generation step itself.
The goal isn’t to make LLMs magically “understand” accessibility features. It’s to recognize that the strategic thinking that produces accessible content also produces the kind of systematic, well-organized information architecture that makes AI systems more reliable and effective.
The Metis Advantage in Content Strategy
This approach to accessibility represents a sophisticated form of metis because it requires seeing beyond immediate constraints to identify opportunities for systemic improvement.
Rather than viewing accessibility requirements as limitations or constraints, this way of thinking recognizes them as design principles that enhance the entire content ecosystem.
A technical writer applying this approach might create documentation that works seamlessly for users with visual impairments, non-native speakers, and AI summarization tools simultaneously, not through separate efforts, but through integrated design decisions that serve all these needs at once.
Instead of “How can we use AI to create more content faster?” perhaps we should ask “How can we create content systems that serve the widest possible range of human needs while also enabling AI to enhance rather than replace human insight?”
A content strategist with this mindset designs information architectures that support assistive technologies while also providing the semantic richness that makes AI tools more effective and reliable.
Most discussions about AI in content creation focus on efficiency, automation, and keeping up with technological change. But the accessibility perspective suggests we’re asking the wrong questions entirely.
Instead of “How can we use AI to create more content faster?” perhaps we should ask “How can we create content systems that serve the widest possible range of human needs while also enabling AI to enhance rather than replace human insight?”
This reframe reveals accessibility not as a specialized concern but as a fundamental lens for understanding content quality in an AI-enabled world. When you optimize for different ways of thinking, different abilities, and different technologies, you naturally create the conditions where human-AI collaboration flourishes most effectively.
The Path Forward
The CAKE Conference reinforced my conviction that accessibility thinking should be central to any serious content strategy, especially one that involves AI tools. This isn’t just about moral imperative (though that matters too)—it’s about strategic advantage.
Organizations that embed this kind of strategic thinking into their content operations will build more resilient, more effective, and more genuinely innovative content systems. They’ll create content that serves diverse human needs while also providing AI systems with the structured, contextual information needed for truly helpful automation.
The ancient Greeks understood that metis, or cunning intelligence, was essential for navigating complex, changing conditions. As we navigate the rapidly changing world of AI-enabled content creation, accessibility thinking offers us exactly this kind of practical wisdom: a way to find clever, adaptive solutions that serve human needs while harnessing technological capabilities.
The most sophisticated AI strategy isn’t about chasing the latest tools or maximizing automation. It’s about developing the cunning intelligence to create content systems that expand access, enhance understanding, and amplify human creativity rather than replacing it.
That’s a form of metis worth cultivating.
What connections do you see between accessibility thinking and your own AI content strategy? I’d love to hear about your experiments with this approach—reply to this email and share your insights.