When institutions describe their goal as building AI literacy, the phrase often obscures more than it reveals. Literacy in relation to AI is not a single capacity that a learner either has or lacks. It is a layered set of competencies — conceptual, structural, and operative — each requiring the preceding level as its analytical foundation. The three-level framework presented here names those layers precisely, so that faculty development programs, governance bodies, and institutional policy work can target the right level for the right purpose.

The Three-Level AI Literacy Framework

Click any level panel to expand its definition

Each level requires the preceding levels as its analytical foundation
Level 1: Conceptual literacy

Conceptual literacy is the capacity to recognize, name, and describe AI systems with precision, identifying system type, tracing the relationship between training data and model output, and distinguishing among architectures, applications, and classification schemes. Knowledge at this level is primarily technical and definitional. Its operative function is to make the vocabulary of AI legible as a site of decision-making: every term in AI discourse encodes a choice about whose cognitive practices count as intelligence, whose outputs count as correct, and whose contributions to the training set are treated as ground truth. Conceptual literacy is present when a practitioner reads a system description and identifies not only what the system does but whose knowledge the system was built to recognize, so that the definitional question "what is this?" becomes inseparable from the power question "whose knowledge is encoded?"

Level 2: Structural literacy

Structural literacy is the capacity to analyze conditions inherent to AI systems as information infrastructure, including the encoding of dominant-language and dominant-culture norms in training data, the reproduction of existing social hierarchies in model outputs, the concentration of AI development within specific corporate and geopolitical contexts, and the mechanisms by which AI assessment instruments scale linguistic hierarchy as if it were cognitive capacity. Analysis at this level moves from description to diagnosis: not what the system is, but what the system reproduces and for whose benefit. Structural literacy is present when a practitioner traces the chain of production and control behind a system, reads ownership, optimization targets, and data provenance as analytically prior to any question of deployment or use, and treats those conditions as objects of institutional inquiry rather than background facts.

Level 3: Operative literacy

Operative literacy is the capacity to exercise contextual, accountable judgment about AI use in specific situations, including the evaluation of outputs, ethical reasoning about deployment, assessment of institutional governance frameworks, and principled refusal where structural conditions warrant it. Judgment at this level is irreducibly particular: the right action is determined by the specific conditions of a specific situation, analyzed through the structural and conceptual capacities the preceding levels provide. Operative literacy is present when a practitioner articulates not only a decision but the structural analysis grounding it, and accepts institutional accountability for that reasoning. Refusal theory (Simpson, 2014; Campt, 2017; Benjamin, 2019) names what operative literacy looks like at institutional scale: not non-compliance but accountable rejection of conditions that reproduce harm, grounded in analysis rather than preference.

Figure 1. Three-Level AI Literacy Framework (Cole, 2026). Panels show the knowledge character, pedagogical approach, assessment orientation, and central power question for each level. Prerequisite arrows indicate that structural literacy requires conceptual literacy as its analytical foundation, and operative literacy requires structural literacy as its analytical foundation. Click any level panel to expand its full definition.

Cole, K. K. (2026). AI literacy is not one thing: Conceptual, structural, and operative levels for higher education governance. Pedagogies: An International Journal. https://doi.org/[to assign at publication]

The AI Literacies Extension · NC State CWSP

Generative AI has changed what it means to ask students to write. Faculty across disciplines are rethinking assignment design, academic integrity policies, and what authorship requires in their fields. The WOLF AI Literacies extension provides a structured space to work through those questions — not with prescriptive answers, but with frameworks, discussion prompts, and design tools grounded in communication pedagogy.

The extension does not advocate for a single approach to AI in the classroom. It is designed for faculty who want to think carefully about what they actually value in student writing and speaking — and to build assignments that make those values explicit, whether AI tools are permitted, restricted, or somewhere in between. It is built on the same foundations as WOLF: it foregrounds what communication is for, who it is for, and what disciplinary participation requires. It treats AI not as a threat to be managed but as a condition of writing instruction that demands more explicit and intentional pedagogy — which is what WOLF has always argued for.

Who this is for

The AI Literacies extension is appropriate for faculty who are revising course policies in response to generative AI tools, redesigning assignments to better assess disciplinary communication rather than AI-generatable outputs, trying to articulate to students (and themselves) what writing and speaking are actually for in their courses, or interested in teaching AI literacies as a transferable skill within their discipline.

It is available as a standalone engagement or as an extension of an ongoing WOLF partnership. Faculty who complete the extension alongside a WOLF engagement often find that it sharpens the Inquiry and Design phases: the questions AI raises about authorship and assessment make visible assumptions about communication that were previously implicit.

The faculty workbook at the center of the extension guides work through four areas:

Area 1 · Conceptual
Understanding how AI tools produce text — what they are, how they generate output, and what assumptions about language underlie their design.
Area 2 · Relational
Examining the relationship between AI use and the communication skills the course is designed to develop — what AI does and does not displace.
Area 3 · Design
Designing assignments that are meaningful regardless of what AI tools exist — assessment grounded in disciplinary values rather than AI-avoidance.
Area 4 · Policy
Building policy language that is honest and enforceable — grounded in the faculty member's own disciplinary reasoning rather than inherited boilerplate.

Liaisons facilitate the extension using the same model as WOLF: sustained conversation, iterative design work, and documentation of decisions and rationale. The workbook is designed for liaison-facilitated use, but faculty are welcome to work through it independently. Contact Kirsti Cole at CWSP to begin or to connect your work to a full WOLF engagement. More information at cwspwolf.com/ai-literacies.

Governance Application · AI Detection Software

The three-level framework is not only a faculty development tool. Institutions are currently making procurement and policy decisions about AI-based academic integrity software — decisions with direct consequences for student populations most vulnerable to algorithmic misclassification. The framework below applies the three-level analysis to that specific governance question, making visible what a conceptual-literacy-only governance process misses and what a structurally and operatively informed decision requires.

⚠ Conceptual-only governance path — click to see what gets missed
Is a tool available? What does it cost? Budget available? Purchase

Consequence: a governance body with conceptual literacy but not structural literacy distributes the costs of the tool's failure modes — a 61.3% false positive rate for non-native English writers — onto the students least protected from the consequences of a false accusation of academic misconduct.

Governance decision: AI detection software
Level 1 Conceptual Literacy
What is this tool and how does it function?

Before any structural analysis is possible, the governance body must be able to accurately describe what it is being asked to purchase.

  • AI detection software classifies text by measuring perplexity (the predictability of word sequences) and burstiness (variation in sentence complexity)
  • Its output is a probability score estimating the likelihood of AI generation — not a verdict on authorship and not a determination of academic misconduct
  • These tools are trained on text corpora with specific distributional properties that shape their classification behavior
  • Detectors require regular retraining as language model outputs evolve; accuracy at purchase does not guarantee accuracy at deployment
Can the institution accurately describe what it is purchasing and what the output represents?
Yes → proceed
No → halt
Gather conceptual information before proceeding to procurement
Level 2 Structural Literacy
Whose writing does this tool treat equitably?

The structural question is not whether the tool works in general. It is whose writing the tool's design encodes as the standard against which all text is measured.

  • Seven widely-used GPT detectors maintained near-perfect accuracy for US student writing while producing a 61.3% average false positive rate for non-native English speakers (Liang et al., 2023, Patterns)
  • Non-native English writers produce text with characteristics — more predictable vocabulary, simpler syntax, formulaic transitions — that detectors associate with AI output; this is a structural property of what the tools measure, not a calibration error correctable by software update
  • None of 14 tested detection tools reached 80% overall accuracy; accuracy degraded further when text was paraphrased (Weber-Wulff et al., 2023, International Journal for Educational Integrity)
  • A single prompting strategy reduced AI text detection rates from 100% to 13% — the same tools that misidentify non-native English writing are easily circumvented by actual AI use (Liang et al., 2023)
Does this tool encode linguistic hierarchy as a structural property of its design?
Yes → proceed to operative judgment
Uncertain →
Return to structural analysis with institution-specific population data
Level 3 Operative Literacy
Given the structural analysis, what does a principled decision require?

Operative literacy does not prescribe a single answer. It specifies the conditions a principled answer must meet.

  • Are process-based alternatives available? Revision history tools and writing process portfolios assess authorship by making the writing process visible rather than classifying output characteristics — without encoding the structural bias pattern detection software reproduces
  • What accountability structures is the institution prepared to enforce? Disaggregated reporting by student demographic, vendor accountability clauses, independent auditing, student appeal mechanisms, and sunset review clauses are minimum conditions for defensible adoption
  • Are the structural conditions of this tool remediable by policy overlays? If the bias is a property of what the tool measures — not a deployment error — policy cannot correct it
  • What does the institution communicate to multilingual students by proceeding with procurement without addressing structural findings?
Decline
Principled refusal

Warranted when structural conditions reproduce linguistic hierarchy, process-based alternatives exist, and structural conditions are not remediable by policy overlays.

  • Grounds the refusal in structural analysis, not preference or risk aversion
  • Articulates the reasoning in terms accountable to the affected student population
  • Names the alternatives the institution will use instead

This is principled operative refusal in the framework's specific sense: accountable rejection of conditions that reproduce harm.

Conditional
Adopt with conditions

Requires all of the following to be explicit conditions of the vendor contract:

  • Disaggregated score reporting by student demographic group
  • Vendor accountability clauses with performance thresholds
  • Independent third-party auditing at regular intervals
  • Formal institutional review after two academic years
  • Robust student appeal process with non-punitive default

Adoption without these conditions in place is not conditional adoption. It is procurement without structural accountability.

Defer
Incomplete analysis

When the structural analysis is incomplete, defer the decision and return to Level 2.

  • Obtain demographic data on the institution's non-native English writing population
  • Request vendor-provided bias testing specific to your student population
  • Evaluate process-based alternatives before proceeding to procurement

A procurement timeline imposed by institutional urgency is not grounds to skip structural analysis. Urgency distributes costs; it does not eliminate them.

Figure 2. Three-Level AI Literacy Governance Decision Framework applied to AI detection software procurement (Cole, 2026). Click any panel or outcome to expand. The framework applies the three-level definition of AI literacy — conceptual, structural, and operative — to a governance decision currently facing many higher education institutions. The conceptual-only path at the top of the figure shows the governance failure mode: procurement without structural analysis.