AI Fluency for pK–12 Educators
The 4D framework rebuilt for the classroom, developed with real pK–12 educators and piloted by teachers. You already do the core skills: Description is close kin to lesson planning, and Discernment is close kin to evaluating student work. These skills transfer directly, just with a new collaborator.
- Start from values: what could careless AI use undermine?
- Keep every student-identifying detail out of AI tools
- Weigh the six ethical tensions in real decisions
- Write a personal AI value document and keep it alive
Welcome, and starting with values
This edition begins where teaching begins: with values, not tools. Before touching any AI, write a context document that opens with four questions:
- What should your students become?
- What is your most ambitious classroom goal?
- Which of your goals could careless AI use undermine?
- Where would AI free your time for higher-impact work?
Question three is the one most AI training skips, and it is the one that keeps the technology in service of the teaching rather than the other way around.
No student-identifying details ever enter an AI tool. Not names, not IDs, not IEP or 504 status, not behavioral notes. This rule has no exceptions and appears in every lesson that follows.
- Start from values: what could careless AI use undermine?
- Student-identifying information never enters an AI tool.
The two loops, in teacher language
The Delegation-Diligence loop holds the higher-level decisions: whether a task belongs with AI at all, and how you own what comes out. The Description-Discernment loop is the day-to-day engine: tell the AI what you need, judge what comes back, refine.
And the transfer claim, which pilot teachers consistently confirmed: you already do this. Every sub plan you have written is Description practice, defining outputs, steps, and success criteria for someone who was not in your head. Every stack of essays you have graded is Discernment practice. The competencies are not new; the collaborator is.
- Outer loop: whether and how much. Inner loop: the daily work.
- Sub plans and grading are the skills; AI is just the new counterpart.
The behaviors, in your words
The 24 fluency behaviors (full reference here) sound different in a classroom. A sampler in teacher language:
- Define the audience: "Reading levels in the room range from 3rd to 9th grade."
- Set interaction style: "Act like an instructional coach and push back if my objective is fuzzy." "Be a thought partner, not a yes-machine."
- Recognize what is not for AI: condolence notes, narrative report cards, recommendation letters. The test: does this need my voice, my judgment, or my relationship? If yes, it is mine.
- Privacy in practice: strip names, IDs, and IEP or 504 status before sharing anything.
- Take responsibility: read the entire output as if a parent or your principal were reading it over your shoulder.
- Assess the collaboration: "We've revised this four times; I think the problem is my original ask, not your drafts."
- The voice, judgment, or relationship test decides what stays yours.
- Read outputs as if a parent or principal were watching.
Capabilities and limitations, classroom tests
Run the machine-side probes on your own ground:
- Versatility: ask for the same concept explained for a 2nd grader, a 7th grader, and a new teacher. Did the pedagogy shift or just the words?
- Hallucination: ask it to recommend a curriculum resource, then check the resource exists.
- Staleness: ask about this year's testing window and see whether it flags its knowledge cutoff or answers confidently anyway.
- Reasoning: give it a student's real misconception and ask it to address it. Does it tackle the actual confusion, or just restate the fact?
- Test on your real curriculum, your real misconceptions, your real calendar.
Creating high-quality outputs
Consider a teacher, call her Ms. Okafor, drafting a differentiated science activity. Her first prompt gets a generic worksheet. What turns it into classroom-fit material is not a better trick; it is three habits working together: rich context (her context document plus the specific class), iterative feedback, and pedagogical judgment applied at each round.
- Your first prompt rarely nails it. Refine using what each response teaches you about what you actually wanted.
- Upload past materials you are proud of, so the AI can match your voice and rigor instead of guessing at them.
- Make revisions concrete. Not "make it better," but "shorten the steps in the center directions and add a picture cue for each one."
Run the loop until it is something you would stand behind, not something you would merely hand out.
- Context, iteration, judgment: the three habits of quality.
- Concrete revisions beat vague ones every time.
Ethics and the six tensions
The outer loop drawn for schools: Delegation decides the right task, tool, and data to hand off; Diligence verifies, attributes, and owns. Legal floor first: student-records law applies to AI tools (in the US that is FERPA; in Europe, GDPR). Use only school-approved tools for anything touching student information.
The six ethical tensions
- Dependency. AI replacing thinking students need to do themselves.
- De-socialization. AI displacing human interaction that is the point of school.
- Bias. AI reinforcing stereotypes and inequities.
- Student agency. Who controls how and when AI is used?
- Equity of access. Who gets excluded when tools assume devices, connectivity, or English?
- Environmental impact. The footprint of the infrastructure behind the tools.
Complete three sentences and add them to your context document: "I will not delegate decisions related to ___." "AI is useful when it helps ___ without worsening the tension of ___." "When unsure, I will ___." That is a personal ethical stance, something no tool can produce for you.
A colleague pastes “Jamal, grade 3, IEP for reading” into an AI tool to draft a support plan. What went wrong?
Name, grade, and IEP status are exactly the details that never enter an AI tool. The same draft works with “a third grader with a reading support plan”.
- Six tensions: dependency, de-socialization, bias, agency, access, environment.
- Student-records law applies to AI tools; approved tools only.
Your personal AI value document
Different models have different personalities, and matching the model to the task is part of platform awareness. But the pedagogy is never the model's. AI is not the pedagogy; you are.
Three sections:
- Intentions. What you want AI to do for your teaching.
- Principles. The two or three rules you will hold regardless of convenience.
- Boundaries. The one kind of work you will always do yourself, named specifically.
Keep it alive with one concrete annual practice: reread it each August, and change what no longer fits. Then compare documents with a colleague, not for approval but for contrast; the differences are where the good conversations live.
- Intentions, principles, boundaries, and one named kind of work that stays yours.
- A value document only works if you revisit it.
Owning the process
You belong in the loop. The work is not done when the AI finishes; the skill you are building is owning the process, not the output. And when the 4Ds are used well, they become invisible: you stop naming the competencies and simply work this way.
Produce one real, usable artifact for your classroom with AI, and keep a process log alongside it: what you asked, what came back wrong or thin, what you kept, changed, or threw out and why, and what only you could add. Close the log with one question: would I put my name on this? Then debrief with a colleague, comparing processes, not outputs.
- Own the process; the log is the evidence.
- Fluency done well disappears into how you simply work.
Closure: solve a problem
The framework is a loop, not a line. For your final build, do not complete a task; solve a problem. Start from "I wish ___ existed" or "I spend too much time doing ___" and build the answer with AI, running everything this course taught.
Real examples built by pilot teachers: a career explorer for students, a Fun Friday planner aimed at the week's lowest-attendance day, hip-hop English lessons that met students where they were. None of these came from a template; all came from a teacher's own "I wish."
- Loop, not line. Solve a problem, not a task.
Course quiz
Attribution. Adapted from the AI Fluency courseware developed in collaboration with Anthropic; the original pK-12 edition was built with Teach For America educators. CC BY-NC-SA 4.0. This adaptation © 2026 AI Literacy Foundation, shared under the same license. Want to run this as a workshop for colleagues? See Train the Trainer.