AI in Your Classroom: A Practical Guide for Academic Educators in 2026

AI didn't break your course — it changed it. Here's how to redesign your assessments, update your AI policy, and actually teach students to use these tools well.

2026-02-20 6 MIN READ PILLAR: EDUCATORS

Academic classroom with students working alongside AI tools on their laptops

If you’ve been teaching for more than two years, you’ve watched your existing essay assessments become partially obsolete, your AI policy document become outdated before you finished writing it, and your students arrive with wildly variable AI literacy — some using it for everything, others avoiding it entirely.

This guide won’t tell you whether AI is good or bad for education. That’s not a useful question. It will give you:

  1. A practical framework for updating your existing assessments
  2. An AI policy template you can actually use
  3. Strategies for teaching AI literacy within your discipline
  4. Ways to use AI to reduce your own marking and admin burden

The Assessment Problem

The honest framing: many traditional assessments that asked students to produce text have become tests of whether students will submit AI-generated text rather than tests of the learning objectives those assessments were meant to measure.

This is only a problem if you conflate the assessment task (writing an essay) with the learning objective (understanding X, being able to analyse Y, arguing Z).

The repair process:

  1. List your actual learning objectives for each assessment
  2. Ask: does the current task assess those objectives — or does it assume that writing the task is the process of developing those skills?
  3. If the latter, redesign toward process visibility

Assessment Formats That Hold Up in the AI Era

Process portfolios: Students submit evidence of their thinking over time — drafts, annotated sources, decision logs, AI conversation histories showing how they refined their thinking.

Oral examinations: 15-minute conversations about student work. Impossible to outsource. Diagnostic for understanding in a way written work never was.

Applied problems in a constrained context: “Here is a real dataset from our discipline. In the next 90 minutes, answer these three questions.” Timed in-session work returns value to demonstrated performance.

Reflective annotations: Students explain their choices — why they structured an argument a particular way, why they chose those sources, what they’d do differently. AI cannot write this convincingly because it requires genuine experience of having made the decisions.

AI-integrated tasks made explicit: “Use AI to generate a first draft of this argument. Your submission must include the AI output, your critique of it, and your revised version.” This is arguably harder than writing from scratch and teaches exactly the skill they’ll need.

The single most practical change

Add a 5-minute oral attestation to any written submission. Ask students three questions about their work. This is not primarily a detection mechanism — most students will engage honestly if asked directly. It’s a signal that understanding matters more than output.

Writing an AI Policy That Actually Works

Policies that simply ban AI use fail because:

  1. They’re unenforceable at scale
  2. AI detection tools have unacceptable false positive rates
  3. They put faculty in adversarial relationship with students
  4. They don’t prepare students for work contexts where AI use is expected

Here is a more workable template:

[Course Name] — AI Use Policy

AI tools are a part of the landscape you’ll work in as a [researcher / professional / practitioner]. This course is partly about learning to use them well.

Permitted without declaration:

  • Using AI to understand concepts covered in readings (you are responsible for accuracy)
  • Using grammar/spelling tools
  • Using AI to brainstorm structures before writing

Permitted with declaration:

  • Using AI to generate draft text that you significantly revise (declare what you used and how in a brief note at the end of your submission)
  • Using AI to summarise sources you have read yourself

Not permitted:

  • Submitting AI-generated text as your own without revision or declaration
  • Using AI to complete tasks specifically designated as in-class or individual
  • Using AI to generate references without independent verification

What declaration looks like: “I used Claude to generate an initial outline for this essay, which I then substantially revised. I used Elicit to surface candidate sources, which I verified and read directly.”

I will not penalise declared AI use unless the work demonstrates that the learning objectives were not met.

Adapt the specifics to your discipline. The key principle: transparency is required, not prohibition.

Teaching AI Literacy as Disciplinary Practice

The most valuable thing you can do for your students is teach them to evaluate AI outputs within your specific discipline.

This looks different in each field:

Sciences/Medicine: Where do AI systems fail on scientific reasoning? Have students identify hallucinated citations, evaluate whether AI-generated experimental designs are plausible, spot statistical errors.

Humanities/Social Sciences: Where does AI reproduce mainstream cultural assumptions? Have students analyse AI outputs for whose perspectives are centred, what is erased, what interpretations are treated as neutral.

Law/Professional fields: Where does AI produce plausible but incorrect legal/procedural claims? The practitioner’s skill of recognising pattern vs. law applies here directly.

All fields: What is the provenance of AI-generated claims? Can students trace assertions to verifiable sources? This is just traditional source evaluation applied to a new type of source.

One class activity that works well

Ask AI to answer a question you know the field answer to. Print the AI output. Have students annotate it in groups: what’s accurate, what’s oversimplified, what’s missing, what’s wrong. The point is not that AI failed — the point is that disciplinary expertise is what lets you know.

Using AI to Reduce Your Own Administrative Load

Ethically and practically, academics can use AI for:

Feedback generation support: Draft feedback comments for common error patterns in student work, then personalise. This doesn’t replace your judgment — it reduces the time spent typing.

Reading summaries: Feed dense administrative documents, draft policies, committee reports through an LLM to get navigable summaries before deciding whether to read in full.

Syllabus and rubric drafting: Use AI to generate a first draft of a rubric based on your learning objectives. You’ll revise it, but the scaffolding time is reduced.

Recommendation letter drafting: Provide key facts about a student and ask for a draft structure. Revise substantially. This is faster than starting from scratch for your 12th letter of the year.

Email responses: For repetitive student queries, draft a response template via AI and personalise.

The principle: use AI for tasks where your professional judgment shapes the output, not for tasks where the output is the product of your scholarly expertise.

Resources for Going Deeper

  • AI for Educators — UNESCO’s practical guide (free PDF)
  • The PAIR Guidebook (Google) — Human-AI collaboration patterns
  • Vanderbilt Center for Teaching — Running practical AI integration workshops
  • The Augmented Scholar YouTube channel covers tool-specific demonstrations

Next in this series: Designing AI-Resistant Assessments for Specific Disciplines →