NUUBI
Back to Blog

From AI Literacy to AI Fluency: What College Courses Can Teach and Measure

· Nuubi Team

Frequent use of generative AI does not mean a student understands the tools or judges them well. In a July 2025 Student Voice survey by Inside Higher Ed and Generation Lab, 85 percent of 1,047 students from 166 two- and four-year institutions said they had used generative AI for coursework in the past year, and 53 percent said they wanted education on ethical AI use (Inside Higher Ed). Those figures describe one sample at one point in time, yet they raise a practical question for faculty and academic technology leaders: if students already use AI, where do they learn to use it well?

Most of the attention so far has gone to a first step, AI literacy: what AI is, how it works, and where it fails. That step matters. This article argues that courses also need the next one, AI fluency, and it looks at what each asks of faculty and how a course can tell whether students are learning it.

AI Literacy and AI Fluency

AI literacy is understanding. A literate student can explain what a generative model does, why it can sound sure and be wrong, and what social costs come with relying on it. AI fluency is practice. A fluent student uses AI well in real work: decides when to use it, asks for what they need, checks what comes back, and takes responsibility for the result.

The difference shows up in how each is taught and assessed. Literacy can be taught in a lecture or a module and checked with a quiz. Fluency shows up in what a student actually does with AI, over time, on real assignments. Learning about AI use is not the same as being measured, and coached, on your own AI use.

Nuubi's earlier article on what it means to be an AI-fluent student described the habits of skilled AI use. This article turns to the course itself.

The 4D Framework as a Shared Vocabulary for Fluency

One useful description of fluency is the 4D Framework for AI Fluency, developed by Rick Dakan and Joe Feller (AI Fluency, Rick Dakan). It names four competencies. Delegation is setting goals and deciding whether and how to engage with AI. Description is communicating a goal clearly enough to get useful output. Discernment is accurately assessing how useful the output is. Diligence is taking responsibility for what you do with AI and how you do it.

The four terms give faculty and students a common vocabulary for talking about use, not only about tools. A student who gets a poor answer can ask whether the problem was the delegation, the description, the discernment, or the diligence.

The Classroom Gap

Literacy has gotten much of the attention. In a late 2025 AAC&U and Elon University survey of 1,057 faculty, 69 percent said they address AI literacy topics such as bias, hallucinations, and privacy in their teaching, and 87 percent said they had created explicit policies on acceptable and unacceptable AI use (AAC&U). The authors describe the survey as non-scientific, so the figures show a pattern among respondents, not the whole field. Faculty are also worried about the next step: in the Digital Education Council's 2025 global faculty survey, 83 percent of 1,681 faculty expressed concern about students' ability to critically evaluate AI output (Digital Education Council).

Policies tell students what is allowed. Literacy lessons tell them how AI works. The next gap to close is moving from policies and literacy lessons to feedback on each student's actual AI use: looking at what an individual student did with AI in the course, and coaching them based on it.

Digital Promise's Framework as a Starting Point for Literacy

Digital Promise's June 2024 framework describes AI literacy through three connected modes, Understand, Evaluate, and Use, with human judgment and justice at the center (Digital Promise). Understanding is basic knowledge of what AI can do and how it works. Evaluating means critically considering the benefits and costs of AI to individuals, society, and the environment. Using covers interacting, creating, and problem-solving with AI for distinct contexts and purposes.

The framework gives course designers a useful shared vocabulary, though its implementation strategies were written for PK-12 settings. In a university course it works best as a starting point, with faculty translating each mode into outcomes that fit their discipline and level.

Clear Rules for Allowed and Prohibited Use

Students cannot practice responsible use when the boundaries shift from one assignment to the next without explanation. A short use statement on each assignment helps. It can name what is allowed, such as generating practice questions or critiquing a draft outline; what is prohibited, such as producing the final analysis or answering a quiz meant to measure unaided skill; and what students must disclose, such as the tool used and how its output changed their work. When an assignment exists to build a specific skill, the statement should say so, so students understand the reason AI is off limits for that task.

An Illustrative Course Activity

The following activity is an illustration rather than a tested intervention. Faculty can adapt the question and evidence standards to their discipline.

Using an approved tool and no student or sensitive information, the instructor generates one AI response to a question central to the week's material, checks it in advance and shares a saved copy with the class, so every student works from the same output and no one needs a personal subscription. Before starting, students briefly explain one relevant limitation of the system, such as a plausible answer that is not backed by a source. They then complete four steps:

Annotate the response, marking claims that are accurate, inaccurate, unsupported, or incomplete.

Verify two specific claims against course readings or another credible source, citing the source for each.

Identify one social cost or benefit of relying on AI for this kind of question, such as a missing perspective or a privacy concern.

Revise the weakest part of the response and explain in a sentence what changed and why.

The activity touches all three literacy modes and produces written evidence of each student's reasoning. In a writing course the response might summarize an argument; in a statistics course it might interpret a regression output; in a health-professions course it might summarize an assigned guideline, with the exercise kept separate from patient care.

Learning Evidence Beyond Self-Reported Confidence

A survey asking whether students feel more confident with AI does not establish what they can do. Stronger evidence comes from the work itself. Faculty can score annotations against a key of the response's known errors, check whether cited sources actually support the verified claims, and judge whether revisions fix the problem they target. A later task completed without AI, such as finding errors in a new response during class, provides evidence about whether students can apply the skill independently. Comparing early and late attempts on a common rubric offers a modest picture of growth within one course, though it cannot show that any single activity caused the change.

An activity like this still measures literacy. To see fluency, faculty need evidence of how a student actually works with AI across many exchanges, which is harder to gather by hand.

Proportionate Faculty Workload

A course can start with one activity rather than a full redesign. A short rubric can assess four things: explanation of a relevant AI limitation, quality of source verification, reasoning about social impact, and justification of the revision. These are proposed criteria, not a validated literacy scale. Reusing a checked response across sections can reduce repeated preparation; label it with the tool and generation date, and recheck the response and sources before reusing it in another term.

For low-stakes practice, faculty can review a sample in depth or use structured peer review for annotations. Any claim about an individual student's proficiency still needs evidence from that student. Track preparation and feedback time before extending the activity across a course.

At larger universities, a teaching center or instructional design team can maintain a shared bank of activities by discipline. Smaller institutions with lean staff can begin with one or two courses and share what works through a faculty learning community.

Equitable Access Without Personal Subscriptions

Students should not have to buy a personal AI subscription to complete required work. Institution-licensed tools, course tools reviewed by IT, or instructor-generated outputs like the one above avoid that cost and give everyone the same starting point. CIOs and chief AI officers play a direct role by confirming which tools have data agreements in place, what student data they retain, and whether they meet accessibility requirements. Where a tool is not available to every student, the assignment should offer an equivalent path that does not depend on it. A shared output removes the need for students to send their own work to an AI service, but accessible document formats and clear instructions still matter.

How Nuubi Approaches AI Fluency

Literacy gets students started. Fluency is the next step, and it is the one Nuubi is built around. Nuubi is an AI-integrated tool for higher education courses. Students ask questions grounded in course materials, either in a shared class discussion or privately, and Nuubi coaches their AI fluency based on how they actually use AI in the course.

A student can run an AI Fluency analysis of their own posts and comments. It rates their use on each of the 4Ds, Delegation, Description, Discernment, and Diligence, and for each one shows the evidence from their own posts, the reasoning behind the rating, and a suggested next step. The analysis builds from real use, as students post and comment, rather than from a separate lesson about AI. Faculty can open the same analysis for an individual student and see where each student stands.

Nuubi's article on where AI adds real value in a course offers criteria for evaluating any course AI tool, Nuubi included.

For faculty adding AI to a course, a good starting question is this: beyond understanding AI, how will students get feedback on how they actually use it?