Policies that say "no AI in this class" and "AI allowed everywhere" both fail. The first is unenforceable. The second abdicates the question entirely.
A more useful approach: design your use of AI by context. Here's how to think about it.
The core distinction: skill-building vs. skill-demonstration
Every assessment falls somewhere on a spectrum from building a skill to demonstrating a skill.
Skill-building context: the student is practicing. Mistakes are expected. Support is appropriate. Feedback is the point.
Skill-demonstration context: the student is showing what they already know. The point is producing reliable evidence of mastery for grading, placement, or promotion.
AI assistance is broadly fine in skill-building contexts. It's broadly not fine in skill-demonstration contexts. Most teacher confusion about AI policy comes from not being explicit about which context a given assignment is in.
Contexts where AI should probably be allowed (or required)
Drafting and revision practice. Students learn to write by writing. AI assistance in iterative drafting — especially when the student critiques and improves AI output — builds revision skills without replacing the student's thinking.
Research support. Helping students quickly understand background material, find key concepts, or navigate unfamiliar topics is a legitimate use of AI. It's comparable to a tutor.
Tool fluency. Future workplaces expect AI fluency. Explicit lessons on prompt design, evaluating AI output, and recognizing AI failure modes are legitimate curriculum.
Accessibility. For students with learning differences, AI can democratize access to ideas that would otherwise be gated by reading or writing speed. This is an accommodation, not cheating.
Make the allowance explicit
"AI use is allowed on this assignment" with a requirement to cite how it was used is dramatically clearer than implicit permission. Students stop having to guess at the norm.
Contexts where AI should be restricted
Summative assessment of individual knowledge. If the goal is to evaluate what this student knows, AI assistance undermines the evidence. In-class, proctored, no-device conditions.
Foundational skill-building. Early in learning a skill, shortcuts undermine the learning. A student who lets AI write their sentences never builds sentence-construction fluency. A student who has AI do their math steps never builds procedural knowledge.
Anything where the struggle is the learning. Some concepts only click when you wrestle with them. AI circumvents the wrestling. This is especially true in math and writing mechanics.
Assessments determining accommodation eligibility, placement, or remediation. The stakes require reliable individual data. AI-assisted work can't provide that.
How to communicate this to students
Students need explicit, per-assignment guidance. Something like:
AI Use Policy for this assignment: ☐ Not permitted — this is a skill-demonstration task ☐ Allowed with citation — document which sections used AI and how ☐ Required with reflection — use AI, then evaluate its output
A rubric slot or a header on the assignment makes the norm clear. Stop relying on implicit understanding.
What AI literacy actually means
"AI literate" students can:
- Evaluate output critically — spot hallucinations, identify missing nuance, recognize confident-but-wrong responses
- Prompt effectively — know how to ask for what they actually want
- Cite appropriately — recognize when AI contribution needs to be acknowledged
- Choose when to use it — understand that AI isn't always faster or better
- Understand its failure modes — know where it breaks down (math, reasoning chains, recent events, specific domain knowledge)
These are teachable skills. They belong in your curriculum the same way information literacy did 15 years ago.
Building an AI-literate classroom
- Label every assignment with an explicit AI policy
- Teach evaluation skills: prompt, output, critique
- Require citation when AI is used
- Reserve no-AI conditions for summative assessment
- Use AI explicitly in skill-building assignments when appropriate
- Discuss AI failures openly when they show up in student work
The near-term reality
Within the next few years, students will be fluent with AI tools the way the previous generation was fluent with Google. Your job isn't to pretend that's not happening — it's to design classroom experiences that build the right skills anyway. That includes both preserving no-AI assessment for valid evidence of individual knowledge, and teaching students to use AI well in the many contexts where it's now a permanent part of how work gets done.
