STRAND C · DEEPEN — YOUR ASSIGNMENTS AND ASSESSMENTS

Hand students the AI version

Generate a piece of work from your own real task, hand it to your students, and have them grade it against your rubric and mark what's missing or wrong. It is the best free AI-literacy lesson available — no policy, no accounts, no permission — and it works even if you never touch AI yourself. Students are far more ruthless about a machine's work than their own, and for once you do not have to be the expert in the room. This is the rung above Grade the AI against your rubric, where you did the grading; here they do.

A STARTING POINT — AND IT NEEDS YOUR JUDGMENT

This page is a draft, and I’m still iterating on it. It wasn’t written by an expert in your field — it is here to start a pathway you then shape to fit you: your subject, your level, your students, your department. Your expertise runs far deeper than anything here, and none of this is finished until your judgment has been on it.

That is the whole point. Judgment is the most important part of this — yours, and the judgment you are building in your students. These pages exist to help you develop and exercise it, never to replace it. Take what is useful, change what is not, and tell me where I have gotten it wrong — that is how these get better.

THE MOVE

Students score AI-generated work against your rubric, mark what it's missing or got wrong, and justify every mark — then turn the same lens on their own work.

WHY IT'S DEEPEN

Evaluation sits at the top of the taxonomy. Judging work against a standard is a harder, more transferable skill than producing it — and this is the cheapest way to make students practise it.

THE CATCH

It only works above the point where students can actually judge the output. Below that line it teaches the opposite lesson — that the machine is right. Naming that line is most of the work.

The whole idea in one line. The tool's weakness becomes the lesson. AI produces work that is fluent, confident, and quietly wrong in ways your rubric was built to catch — so you stop fighting that and hand it to students as the thing to be caught. The machine stops being the source of the answer and becomes the thing they argue with. You are not teaching them to use AI; you are teaching them to outgrade it, which is the same skill as knowing whether their own work is any good.

WHY THIS ONE MOVE DOES SO MUCH

Most AI-in-the-classroom ideas cost you something — accounts to provision, a policy to defend, equity headaches when access runs from paid subscriptions at home to nothing. This one costs almost nothing and pays four ways at once:

IT TEACHES EVALUATION

Scoring a piece of work against a standard and justifying the score is the top of the cognitive taxonomy — harder and more durable than producing the work. Students rarely get to practise it, because they rarely see finished work that is wrong. AI manufactures exactly that, on demand, from your own task.

NO POLICY, NO ACCOUNTS, NO PERMISSION

You run the tool once, before class, and print the output. Students never log in and never need to. There is no policy question to settle because no student is using AI — they are grading a handout. It is equitable by construction: everyone works from the same sheet of paper.

IT WORKS EVEN IF YOU ABSTAIN

You do not have to like AI, use it in your own work, or allow it in your room to run this. If you have decided AI has no place in your classroom, this is still yours — arguably it matters to you more, because it is how students learn to distrust the thing you have banned. Bring me the task and I will generate the output for you.

THEY'RE RUTHLESS ABOUT A MACHINE'S WORK

A student who will defend every comma of their own essay will tear a machine's paragraph apart without hesitation. There is no ego in the room and nothing to protect. That critical distance is precisely what you want them to eventually turn on their own work — and grading the machine is the low-stakes rehearsal for it.

And the part teachers underestimate: you do not have to be the expert in the room. You are not lecturing on how AI works or defending a position on it. You are handing out a flawed artefact and running the discussion you already know how to run — the one about what good work looks like. The AI just supplied the flawed draft, which is the one thing that was always hard to produce on purpose.

HOW TO RUN IT

The base version is five steps and one class period. Use a task you already assign, not one invented for the occasion — the point is to grade the machine on your real work.

Generate the output from your real task. Give the tool the actual prompt you give students and take what it produces. Read it once yourself first — both to know what's in it and to verify any facts it asserts (see the honesty note below).
Hand it out with the rubric. The same rubric you'd grade a student on. No preamble about AI being bad; just “here is a submission, grade it.”
They score it and mark what's missing or wrong. A number in each rubric row, plus annotations in the margin — where it's thin, where it's confidently incorrect, where it hit the form but missed the thinking.
They justify every mark. This is the assignment. “I took a point here because it asserts the claim but never shows why the evidence supports it.” The justification is the evaluation skill made visible — and a better diagnostic of what they understand than their own draft would give you.
They turn the same lens on their own work. Now grade your own last piece against that rubric, the way you just graded the machine. The critical distance they found for the machine is the thing you are trying to transfer — this step is where it transfers.

Three variants, once the base move is familiar:

The error-hunt

Seed the output with a known set of errors — or ask the tool to — and tell students how many are hidden in it. They hunt. Same noticing skill, but the count turns it into a game and guarantees there is something to find. This is also the safe version below the threshold (see the next section).

“Is this an A?”

A calibration exercise. Hand out the machine's work and ask for a letter grade and a defence of it, then reveal the spread of grades the class gave. Where they disagree, the rubric language is ambiguous — and they discover for themselves that “sounds good” and “is good” are different measurements.

Redraft the machine

Marking what's wrong is diagnosis; fixing it is the harder half. Students rewrite the weakest section to actually meet the rubric. It forces the evaluation to cash out in production — you cannot fake having noticed the problem if you have to repair it.

THE THRESHOLD — WHERE THIS FLIPS FROM USEFUL TO HARMFUL

This is the load-bearing caveat, and skipping it is the single most likely way the move goes wrong. Grading the machine only works above the point where your students can actually judge the output. Below that point, a student looking at a fluent, confident piece of AI work has no basis for finding its faults — so they will assume it is right, adopt it, and you will have taught them, with your own handout, that the machine is right. That is the exact opposite of the lesson.

Where in my sequence does a student become able to look at machine output and say “that's wrong, and here's why”? Below that line, every grade-the-machine task is harmful. Above it, the same task is one of the best things you can assign.

The line is real and it is specific to your discipline. A novice can be handed a machine translation and will simply trust it; a student two years further on can say “it used the past tense there and I don't think it should have.” Same task, opposite outcome, and the only thing that changed is whether the student had the expertise to contest what they were looking at. The World Languages page works this out in detail for translation; every department has its own version of the line, and its own version of what a student at that threshold can reliably catch versus what sails past them.

So name where the line falls in your own sequence, and below it, use the error-hunt instead. When you seed the errors yourself and tell students they are there, the machine is no longer an authority to be trusted or doubted — it is just a passage with known mistakes in it, and the task is the same skill of noticing with none of the machine's borrowed credibility in the room. You get the evaluation practice without teaching a student who can't yet judge that the machine's answer is the standard.

DON'T TEACH THAT THE MACHINE IS ALWAYS WRONG

The whole move relies on the AI output being flawed, so it is tempting to let students walk away with “AI work is bad.” Resist it — that is its own false lesson, and students will catch you in it the first time the machine produces something genuinely good.

Sometimes the output is good, and that is also the lesson. When the machine gets it right, “why is this actually good?” is the same evaluation skill run the other direction — students naming what a strong piece does, against the rubric, is exactly the calibration you want. Grade the work in front of you on its merits, not on where it came from.
Verify anything the AI asserts as fact yourself, first. These tools fabricate — confidently, and in the register of something true. Before you hand out a passage, check any date, quotation, citation, or factual claim in it. Handing students invented facts and calling them a rubric exercise teaches the wrong thing and undoes your credibility. If you would not put it on the board yourself, do not put it in their hands unread.
The target is discernment, not disdain. You want students who can tell good work from bad regardless of who or what produced it — not students who reflexively distrust one source and reflexively trust another. The machine is a foil for teaching judgment, not the villain of the lesson.

WHERE THIS SITS

This is one rung on one strand. The moves next to it:

The rung below — Grade the AI against your rubric. There you run your task through AI and grade the result, to find out what your assessment actually measures now. Same tool, same rubric; the difference is who holds the pen. Do that one first — it tells you whether the output is even worth handing out.
The rung above — Building a Contested task. Instead of a one-off handout, the tool itself is configured to withhold the answer, argue, or seed errors a student has to catch — the move built into the assignment rather than run as a lesson. It fails silently if the setup is skipped, which is why it is the top rung.
The assessment-design cousin — Write an assessment worth trusting. Its Innovate rung makes judging AI output the assessed skill as a validity move — widening the task back out to include the thinking that the machine can't fake. Same idea from the assessment side rather than the AI side.
The discipline-specific version — your subject page. The generic move is worth about what it costs. Where the threshold actually falls, and what a student at it can catch, is a departmental answer: Classical & Modern Languages is written; English, Math, Science, Social Studies, Theology and the arts each get their own. Start there for the version that fits your room.

Bring me one task you already assign and we'll generate the version to hand out — the output, the rubric to grade it against, and a read on whether your students are above the threshold where this teaches judgment rather than the opposite. Twenty minutes, and more useful than another document.

Jason Beyer · Educational Technology · Regis Jesuit High School · A.M.D.G.

HelpDesk@regisjesuit.com · 303.269.8080 · LAST VERIFIED — AUGUST 2026