Campus Writing and Speaking Program, NC State University, 2026
This workbook supports the AI Policy Lab workshop, a faculty workshop offered through NC State's Campus Writing and Speaking Program. It is a working document. Write in it and argue with it. The annotations in Part II model one reading among several the documents support, and in Part III you challenge, extend, or contradict what you find there.
Framework: Cole, K. K. (2026). AI literacy is not one thing: Conceptual, structural, and operative levels for higher education governance. Pedagogies: An International Journal.Framework page · go.ncsu.edu/cwsp
Part I
How to Use This Workbook
Before you arrive. Bring your current policy language on AI, at whatever level you hold it: a syllabus statement, an assignment rule, a program or department text, or the fact that none exists yet, which is itself a starting point. Have the link to it open, or paste it into Template 2 in Part III.
In short, if you read nothing else first. The Lab reads governance documents with four steps (identify the genre, locate the glitch, read the reveal, name the abstraction) and three questions (what does the policy name, whose problem does it make AI, what judgment does it route and to whom). You will apply both to one case from the census in a small group, then draft three to five sentences of policy at the level you choose, meeting five principles: scope terms by function, date the policy, name the revision trigger, separate durable principles from contingent prohibitions, explain the reasoning.
Each annotated policy entry in Part II applies one four-step analytical procedure and the three-level literacy framework. Apply the same procedure to new documents in Part III.
The Analytical Procedure
Four moves for reading any AI governance document, applied in sequence.
1
Identify the genre
What type of document is this? Who issued it, and in what institutional capacity? Genre sets the authority a document carries and the limits of what it is able to require.
2
Locate the glitch
Where does the document strain, contradict itself, or hedge where it claims to govern? Where two documents conflict, decide whether both are current (a contradiction) or one has replaced the other (a supersession). At the glitch, the document's work becomes visible.
3
Read the reveal
What does the glitch expose about what the institution values, fears, or cannot resolve? What structural condition produced this document?
4
Name the abstraction
What general claim about AI governance, authorship, or power does this document support? What principle would guide better policy at this institution?
The Three-Level Framework
After applying the four-step procedure, characterize the literacy demands the document makes, or fails to make, at three levels. The framework page sets out the levels as the order in which a person reasons a decision (describe, analyze, decide). Here the same three levels read a document: what it asks its readers to recognize, to understand about the system, and to judge.
Conceptual
Recognition and naming: what terms does the policy use, and what do those terms assume? What are students and faculty able to name, and what goes unnamed?
Structural
Power and infrastructure: whose interests does the governance structure serve? Who authored this document, and what institutional pressures shaped it?
Operative
Contextual judgment: what operative choices does the policy ask faculty and students to make? What decision-making capacity does it cultivate or foreclose?
Reading for the Glitch
Every AI governance document produced under institutional pressure reveals something its authors did not intend. We call this the glitch: the moment where the document's language breaks down, hedges, contradicts itself, or produces an unintended meaning.
Treated as an analytical resource, the glitch shows what the institution values and fears, and it marks what the authors were unable to resolve at the time of writing.
Look for: hedging language ("strongly discouraged"), contradictions between sections or between two current documents on one campus, silences where governance should be explicit, borrowed frameworks applied beyond their scope, and temporal artifacts (language accurate when written and no longer true).
Part II
Annotated Policy Corpus
Four annotated entries from the AI Governance Census, version 0.5. Entries 1 and 2 read NC State’s own record, the two OIT pages against each other and the DELTA teaching page against the Code of Student Conduct. Entries 3 and 4 come from land-grant peers, one a contradiction between two current documents (Clemson) and one a supersession with the superseded pages still posted (Michigan State). Each entry applies the four steps and the three questions from Part I.
These annotations model one reading among several. Use them as a starting point for your own. Discussion notes at the end of each entry suggest how to open the annotation to disagreement.
NC State · Entries 1–2
“approved for green (not sensitive) and yellow (moderately sensitive) data only” (OIT, Gemini Chatbot Available to NC State, November 20, 2024). The current Approved AI Solutions list clears Gemini Chat, Microsoft Copilot, and ChatGPT Business or Enterprise for Red (Highly Sensitive) data.
Four-Step Analysis
1
Genre
An announcement and a standing list, both issued by the Office of Information Technology, the list through Vendor Risk and License Management. Between them the two pages carry the binding rules on AI tools which NC State has published, since OIT sets the data classifications every unit works under.
2
Glitch
When OIT announced Gemini in November 2024, it cleared the tool for green and yellow data only. The current list, which carries no date, clears Gemini Chat, Copilot, and ChatGPT Business or Enterprise for Red data, and the announcement stays posted with no note of the change.
3
Reveal
Because the list is undated and the announcement carries no revision note, a reader has no way to tell whether the clearance widened after a review or whether the two pages were never reconciled, so a faculty member deciding whether students are permitted to enter research data reads two rules from one office.
4
Abstraction
Where the operative rule lives in a tool list, the governing question for a course moves from what the course allows to what the data classification allows, and a course policy which names tools without naming the classification inherits whichever page its reader finds first.
Three-Level Literacy Read
Conceptual
OIT names products and versions (free public Gemini rejected, Gemini Chat approved), so the record defines AI by license and version.
Structural
Classification follows the vendor contract. The enterprise license, with its data terms, moves a tool from rejected to cleared for Red data.
Operative
The pages route the decision to whoever handles the data, with no instruction for a course in which students handle research or FERPA-protected records.
Discussion Notes
Decide whether this is a contradiction or a supersession, and name the evidence on either page which would settle it. Then ask what one sentence in a course or unit policy would have to say about data for the policy to hold whichever OIT page is current.
“As of May, 2026, there is no official university policy regarding AI use.” (DELTA). The Code defines cheating to include “using materials, equipment, or assistance in connection with an Academic Exercise that has not been authorized by the Instructor” (8.1.b).
Four-Step Analysis
1
Genre
DELTA’s page is teaching guidance from the instructional technology center, persuasive and without authority to sanction. REG 11.35.05 is a regulation issued under the Vice Chancellor and Dean of Academic and Student Affairs, and it is the text a conduct hearing applies.
2
Glitch
The most restrictive sample statement on the DELTA page tells students that any attempt to use AI tools “will be considered academic misconduct,” while the Code, revised in July 2025, names no AI and defines cheating as assistance the instructor has not authorized.
3
Reveal
Read together, the two pages place the whole of NC State’s academic rule on AI in the instructor’s syllabus, since the Code’s authorization clause takes its content from whatever the course states, and DELTA, in saying there is no university policy, tells instructors as much.
4
Abstraction
Where a conduct code leaves authorization to the instructor and a teaching office supplies the sample sentences, the syllabus becomes the governing instrument for AI at the institution, although no office has decided it should be.
Three-Level Literacy Read
Conceptual
DELTA’s samples name “chatbots, text generators, paraphrasers, summarizers, or solvers,” a list by function. The Code names no technology.
Structural
Authority sits with the Office of Student Conduct and the content with the instructor, and no office between them has published a university rule.
Operative
Each instructor’s statement becomes the rule a hearing applies, so the operative judgment at NC State varies course by course and section by section.
Discussion Notes
Ask whether the unit wants to leave this to each syllabus, to write a shared statement its syllabi cite, or to ask for a university rule, and what each choice costs the faculty who teach the most sections. Minnesota, one of NC State’s peers, closed the same gap by a provost memo applying an existing code clause to AI.
Land-grant cases · Entries 3–4
“The use of generative AI is a form of plagiarism per the Academic Integrity Policy.” (Faculty Cheat Sheet on Academic Integrity). The teaching center’s sample statements, from the same campus, include one permitting fully integrated AI.
Four-Step Analysis
1
Genre
The sample statements come from the teaching center as optional syllabus language. The cheat sheet comes from the academic integrity office, beside a plagiarism page which quotes the undergraduate catalog: graded works “generated by artificial intelligence or ghostwritten (either paid or free) are expressly forbidden.”
2
Glitch
Among the teaching center’s four statements is one permitting fully integrated AI, while the integrity office tells faculty that generative AI use is plagiarism under the policy, so an instructor who adopts the most open sample writes a syllabus which the integrity office’s guidance classes as permitting plagiarism.
3
Reveal
Two offices have each answered the AI question in their own genre, teaching guidance by offering options and integrity guidance by applying the catalog, and neither page refers its reader to the other.
4
Abstraction
Where options and prohibitions are published by different offices for different readers, the contradiction surfaces in a hearing, at the point where a student’s syllabus and the catalog are read together for the first time.
Three-Level Literacy Read
Conceptual
The catalog names “artificial intelligence” without defining it. The sample statements define permitted uses by task.
Structural
The integrity office holds enforcement and the teaching center holds course design, and each writes for its own audience.
Operative
An instructor choosing a sample statement has no way to learn, from the teaching center’s page, that a hearing will read the statement against the catalog.
Discussion Notes
Which document would a hearing officer apply? What would an instructor have to add to the open-use statement for it to hold against the catalog? Then ask whether the same two offices on your own campus have ever been read together.
“These guidelines supersede all previously issued guidance related to the use of generative AI tools at MSU.” (MSU Guidelines, August 12, 2025)
Four-Step Analysis
1
Genre
The 2025 Guidelines are a formal instrument from the provost’s office, the highest-order AI document at MSU. The 2023 interim guidance from IT and the provost’s 2023 materials are earlier guidance from the same two offices.
2
Glitch
The Guidelines supersede all prior guidance and require instructors to state course AI rules in every syllabus. The 2023 IT interim guidance, revised in March 2025, remains posted and names Copilot the only IT-approved generative AI tool, while a February 2026 IT knowledge-base article licenses ChatGPT EDU to departments.
3
Reveal
Because IT revised its interim guidance in March 2025 and posted a licensing article in February 2026, the superseded pages read as current. The provost’s office and IT revise their AI pages on separate schedules, with no shared note of which page governs.
4
Abstraction
Because a supersession clause binds only the readers who reach the new document, the record reads as its oldest posted page reads until the superseded pages are withdrawn or marked.
Three-Level Literacy Read
Conceptual
The Guidelines define generative AI tools by function. The IT pages define approval by product.
Structural
The provost’s office holds the rule and IT holds the tools, and the two publish on separate sites.
Operative
The syllabus requirement routes the course decision to instructors, with the Guidelines’ direction that detector results never serve as the sole basis for a finding.
Discussion Notes
Ask whether a required syllabus statement is something the room wants, and say plainly that Michigan State’s requirement is one instrument among several in the peer set (Minnesota interprets an existing code clause, Texas A&M writes AI into its student rules). Then ask what a unit at NC State would need to withdraw or mark if it adopted a new statement next term.
Part III
Policy Analysis Templates
Two templates cover the workshop: the first for the census case you read in Activity 1 (alone, in a pair, in a group, or as a whole room), the second for the policy language you brought from your own course, program, or unit. Add a third or fourth below if you read more documents. Apply the four steps and the three questions from Part I.
Your responses are saved automatically in this browser.
Template 1: your case from the census
Excerpt or paraphrase. Include a section reference if available.
Four-Step Analysis
Three-Level Literacy Read
Template 2: your own document
Four-Step Analysis
Three-Level Literacy Read
Template 3: another document
Four-Step Analysis
Three-Level Literacy Read
Template 4: another document
Four-Step Analysis
Three-Level Literacy Read
Part IV
Adaptable Policy Language
These templates offer starting language for AI use policies at three levels. Five principles from the census shape them:
1. Scope terms functionally
Define AI by what it does (generates text, produces images, automates tasks). Product names date quickly, and named-tool policies go obsolete in months.
2. Date your policy
Every policy in the census was written in conditions which have since changed. State the term the policy covers and commit to revision.
3. Name the revision trigger
Specify what would prompt an update: tool proliferation, new institutional guidance, or evidence of the policy producing the wrong outcomes.
4. Separate durable principles from contingent prohibitions
State the commitment (what the course or unit develops and assesses) apart from the rule (which functions are barred this term). Different bodies write each, and the first outlasts the second.
5. Explain the reasoning
A policy students understand is easier for them to follow, and a reviewer knows what to keep when the tools change. State why the policy takes its position.
Treat the templates below as starting points to rewrite. Replace everything in [brackets]. Edit directly in each template, and this browser saves your changes automatically. Use the Print or save as PDF button to export your draft.
Course-Level: Syllabus Statement
Course · Syllabus
Bracket guidance: The first bracket asks for a functional definition, without product names. The second and third brackets ask you to decide for your course now. The attribution bracket asks you to name what disclosure looks like in practice. The final sentence commits you to revision, so build the time into your semester-end review. The term check is Module 4's line: the date goes in the policy itself. If your statement mentions detection, Penn State's academic integrity office supplies the model sentence, that detector results "are not reliable enough to use as the basis for an academic decision."
Program-Level: Shared Governance Statement
Program · Handbook / Sequence Governance
Bracket guidance: A program statement leaves some questions open, since its work is to give individual instructors a framework to apply. The decisive move is naming the discipline-specific literacy the program values, because that naming is the governance decision and the rest of the statement follows from it. The data and integrity lines carry two census patterns into the program: data rules outnumber teaching guidance, and integrity codes, NC State's among them, name no AI. Where a unit framework already states both, cite it here and cut the two lines.
Unit-Level: Governance Framework for a Department, School, or College
Unit · Department, school, or college governance
Bracket guidance: A unit framework sets the conditions for course and program policy, and its two load-bearing lines are the authority line, which names who decides at each level, and the review line, which names who checks the framework each term. The commitment and the rule are written apart because different bodies revise them on different schedules, the commitment by the faculty and the rule each term. In the census, campuses whose teaching and integrity offices publish separately are the ones which record contradictions between their own documents, so the data and integrity lines name the university documents the framework depends on.
Your edits are saved automatically in this browser.
Your students need a policy they are able to reason from. A policy that explains its reasoning builds literacy. A policy that withholds its reasoning performs authority.
Part V
Submit Your Work
Submitting sends your Part III templates and Part IV drafts, as they stand, to Kirsti Cole at CWSP. Submissions stay separate from the AI Governance Census and do not enter its corpus. The form collects no name or email address, only tags for discipline, institution type, and engagement. Because submissions carry no name or email, CWSP cannot write back about them. To talk about your policy work, use the CWSP intake form (opens in a new window).
Your analysis and drafts
Two tags are required. The other two help CWSP read submissions across levels and formats.
What is collected and how it is used. Submitting sends your templates, drafts, and tags to Kirsti Cole at CWSP. It collects no name, email address, or affiliation. CWSP uses submissions to track how faculty across disciplines read and write AI policy and to plan programming. Nothing submitted here enters the AI Governance Census. Individual responses are not published in identifiable form. If CWSP uses the submissions as a research dataset, that use will proceed under IRB review, and consent procedures will be updated to match.
Enter a discipline, choose an institution type, and consent to enable the button.
Three places to look. The Institutions tab (opens in a new window) opens each institution with a heading line and a summary, then the documents the search found, each with its issuing office, intended readers, dates, and a quoted passage. The Contradictions tab (opens in a new window) lists every recorded conflict between two documents on one campus, naming both documents and the point where they disagree, and it is the material for Activity 1. The data files (opens in a new window) on GitHub hold the inventory, the institution log, and the sampling frame as CSV, under CC BY 4.0.
Where a document in the census no longer opens, or a page has moved, the census's own correction form takes the report. Analyses and drafts written in this workbook go to CWSP through Part V and stay apart from the census.