The Shadow Literacy Gap: 92% Adoption, 36% Training

Here's a number that should alarm anyone working in education, policy, or technology design: 92% of students now use AI tools. Only 36% have received any training on how to use them. That's not a digital divide in the traditional sense — everyone has access. It's something worse: a shadow literacy gapwhere the tool is everywhere but the fluency to wield it effectively is not.

Access Is Not Fluency

A February 2026 report from Genio paints a stark picture. Higher-income students use AI for deep research synthesis — constructing arguments, cross-referencing sources, building knowledge architectures. Lower-income students use the same tools for basic summaries. Same ChatGPT. Same interface. Radically different outcomes.

This pattern is exactly what we'd predict from the ALC Stratification framework. Access to the application layer — having a login, being able to type a prompt — is table stakes. What determines outcomes is fluency: the ability to navigate system affordances, construct effective queries, interpret outputs critically, and iterate on results. That's not a skill you absorb by proximity. It's learned, practiced, and — crucially — unevenly distributed.

Algorithm Awareness ≠ Algorithm Literacy

The research community is catching up to this distinction, but slowly. Most studies still measure what I'd call schema-level knowledge: can you identify that an algorithm is shaping your feed? Do you know TikTok uses recommendation systems? That's awareness. It's necessary. It's not sufficient.

Felaco's 2025 study of TikTok users is instructive here. They found that users could articulate that algorithms influence what they see — high awareness scores. But this awareness didn't translate into different behavior, better critical evaluation of content, or any measurable shift in how they interacted with the platform. Knowing the algorithm exists and knowing how to converse with it are fundamentally different competencies.

In ALC terms: awareness is recognizing the schema. Literacy is participating in the conversation. The gap between those two is where stratification compounds.

Three Dimensions of the AI Divide

A new Frontiers in Computer Science paper (February 2026) maps what they call three dimensions of the AI divide: access, skills, and outcomes. This tracks with decades of digital divide research from Hargittai and van Dijk, but applied specifically to AI systems. Their framework reinforces what the Genio data shows empirically — device access alone is insufficient. Without the skills dimension, access produces divergent outcomes along existing socioeconomic lines.

What's missing from their framework, and what ALC adds, is the mechanism. Why do skills diverge? Because the application layer isn't neutral. Every interface embeds assumptions about what the user knows, what they want, and how they should interact. Those assumptions are designed by people with high ALC fluency for other people with high ALC fluency. The result is a feedback loop: fluent users get more value, which builds more fluency, which extracts more value. Everyone else gets basic summaries.

Hong Kong's Lampposts and the Limits of Design

The most vivid example of ALC fluency under pressure comes from Hong Kong. During the 2019-2020 protests, citizens didn't just worry abstractly about surveillance — they physically dissected smart lampposts to understand the data infrastructure inside them. They traced data flows from interface to backend. They mapped which sensors connected to which government databases.

Ting's 2026 HICSS paper on folk theorization and data disobedience documents this as what he calls "infrastructure investigation" — citizens developing their own theories about how systems work, then testing those theories through direct engagement. That's not digital literacy in any conventional sense. That's application layer fluency under political duress. It demonstrates that ALC isn't an abstract academic framework — it's a survival skill.

What This Means for Design

If 92% of students have access and only 36% have training, the response can't be "more training." Training doesn't scale to match adoption velocity. The response has to include design accountability: building systems that don't require high ALC fluency to produce meaningful outcomes.

This is where the consulting work gets concrete. An ALC audit examines a system's interface and asks: what assumptions does this make about the user? Where are the stratification points — the moments where a fluent user diverges from a non-fluent one? What would it take to flatten those divergence points without dumbing down the tool?

The shadow literacy gap isn't inevitable. It's a design choice masquerading as a skills problem. And the first step to closing it is measuring it correctly — not as awareness, but as fluency. Not as access, but as outcomes.