Setting AI Rules for Different Assignment Types in One Course

The workable way to set different AI rules per assignment type is a three-tier system — no AI, AI for support, AI permitted — declared once in the syllabus and then stamped visibly on every single assignment. One blanket rule for a whole course always fails in one direction or the other: too strict, and you’ve banned tools your discipline actually uses; too loose, and you’ve gutted the skills the course exists to build. The rule has to follow the learning objective, and learning objectives change assignment by assignment.

That’s the entire framework. The craft is in the labeling, the syllabus language, and the enforcement design — which is where most per-assignment policies quietly fall apart.

Key takeaways

• One course-wide AI rule is either too strict or too loose for half your assignments. Match the rule to each assignment’s learning objective.
• Three tiers cover nearly everything: no AI, AI for support with disclosure, AI permitted with accountability.
• The label at the point of use — on the assignment sheet, in the LMS — matters more than the syllabus paragraph. Students follow rules they can see.
• Enforce by design, not detection: in-class components for Tier 1, disclosure statements for Tier 2, accuracy-weighted grading for Tier 3.
• Treat honest tier confusion as a labeling failure first, misconduct second.

Why blanket rules break

Picture three assignments from one composition course. A timed reflective paragraph, written to build the muscle of producing prose. A researched argument essay, where sources and structure are the point. A revision exercise, where students improve a flawed draft.

A course-wide “no AI” rule makes sense for the first, is debatable for the second, and is actively weird for the third — where having students critique a machine draft might be the best version of the exercise. A course-wide “AI welcome” rule is worse in the other direction: the timed paragraph becomes meaningless.

Blanket rules also collide with what’s happening across campus. A student’s other courses may run on entirely different norms, and the MLA-CCCC task force on AI has pushed in exactly this direction: transparent, assignment-level guidance rather than course-level slogans. Your students aren’t confused because they’re careless. They’re confused because “can I use AI?” genuinely has a different answer in every room, and often within one room.

The three tiers of AI rules per assignment type

Keep the system small enough to remember. Three tiers cover nearly every case:

Tier 1 — No AI (red)

*The assignment measures unassisted skill.* Timed writing, first drafts in a skills sequence, exams, anything where producing the thing is the learning. State the boundary concretely: no generation, no paraphrasing tools, no AI editing — spellcheck and standard grammar underlining are fine. Vague boundaries (“no substantial AI use”) generate the disputes you’re trying to prevent.

Tier 2 — AI for support (yellow)

*The assignment measures the student’s thinking; AI can scaffold but not substitute.* Brainstorming, generating counterarguments, feedback on a draft, grammar cleanup — allowed. Submitted prose must be the student’s own, and use gets disclosed in a two-line statement: what tool, what for. This is where most real-world assignments live, and the disclosure habit is the tier’s real product; it converts AI use from something students hide into something they can talk about. Rubric design for this tier has its own wrinkles — we’ve written up grading assignments where AI is permitted separately.

Tier 3 — AI permitted (green)

*The assignment measures judgment, curation, or output quality.* Students may integrate AI freely, disclose how, and own every claim in the result. The accountability clause does the enforcement for you: unedited AI output is confidently wrong often enough that a grade riding on accuracy punishes laziness without any policing.

Labeling: the part that actually determines success

Here’s the failure mode of most per-assignment policies: the tiers live in an eloquent syllabus paragraph from August, and by week nine nobody — sometimes including the instructor — remembers which rule applies to the annotated bibliography. The fix is unglamorous repetition at the point of use:

1. On the assignment sheet: a colored badge or a single bold line at the top. “AI: Tier 2 — support only, disclose.”
2. In the LMS assignment description: same line, first sentence.
3. Out loud, once: ten seconds when you introduce the assignment.

If you’re drafting the syllabus language itself, our syllabus policy templates for allowed, limited, and banned AI use give you copy-ready wording for each tier.

Enforcement that doesn’t depend on a percentage

Notice what this system doesn’t need: a detector verdict. Each tier enforces itself structurally.

• Tier 1 pairs with process evidence — some of the work happens in class, or drafts and version history come attached. Violations show up as trail anomalies, not scores.
• Tier 2’s disclosure statement changes the violation from a gray area into a false statement. “You didn’t disclose” is a cleaner, fairer conversation than “the software says 87%.”
• Tier 3 self-enforces through accuracy grading.

Detectors can still play a small, honest role: as a screening prompt that sends you to look at the process trail, or as a classroom demo. If you’ve never watched one work, run a sample through a detector and look at the sentence-level heatmap rather than the single number — it recalibrates how much weight any score deserves in a policy dispute.

And when a student honestly follows the wrong tier — it will happen, usually in September — treat the first case as your labeling failing them. Offer a redo under the correct rule and make the label louder. A policy students can navigate without fear is the one they’ll follow all term. More course-policy walkthroughs are in the full library of educator guides.

Frequently asked questions

Why not just have one AI rule for the whole course?

Because assignments do different jobs. A skills-building first draft measures whether the student can produce prose; AI use there defeats the point. A data analysis write-up measures interpretation; AI polishing the prose may be harmless. One blanket rule is either too strict for some assignments — banning tools the discipline actually uses — or too loose for others, gutting the skill you’re trying to build. Per-assignment rules match the rule to the learning objective.

What’s the simplest way to label AI rules per assignment?

A three-tier system, announced in the syllabus and stamped on every assignment sheet: Tier 1, no AI (the work measures unassisted skill); Tier 2, AI for support (brainstorming, feedback, grammar — but the submitted prose is the student’s, with use disclosed); Tier 3, AI permitted (integrated freely, disclosed, with the student accountable for accuracy). Many instructors color-code them red, yellow, green. The label on every assignment matters more than the elegance of the tiers.

How do I stop students from confusing the rules between assignments?

Repetition at the point of use. The tier goes on the assignment sheet itself, in the LMS assignment description, and gets ten seconds of mention when you introduce the task — not just a paragraph in the syllabus nobody rereads in week nine. Confusion thrives when the rule lives in one document from August. It dies when every assignment answers the question on its face.

How do I enforce different tiers without a detector?

Design each tier’s assessment to make violations visible or pointless. Tier 1 work happens partly in class or with drafts and version history attached. Tier 2 includes a short disclosure statement, which turns silent misuse into a documentable lie rather than a gray area. Tier 3 needs no policing at all — the grade rides on accuracy and judgment, which is where unedited AI output falls down on its own.

What happens when a student follows the wrong tier honestly?

Treat the first instance as a communication failure, not misconduct — especially early in the term. Let the student redo or revise under the correct rule, and use the incident to check whether the label was actually visible on the assignment. A policy that produces honest confusion is a policy that needs clearer labeling; punishing students for your syllabus’s ambiguity teaches them the rules are a trap.