Syllabus AI policy language: the three-tier model, with real university sample text

The short answer

Teaching centers have settled on a three-tier template for syllabus AI policies: prohibited, allowed with disclosure, or encouraged within limits — and most publish copy-paste sample language for each. Whichever tier you choose, the statement does three jobs: it tells students where the line is, it becomes the binding document if a dispute arises, and it quietly commits you to an enforcement method. The first two jobs are easy. This article gives you the verbatim sample language for the first two, and an honest look at the third.

A syllabus statement matters more than institutional policy because it is what an integrity panel will actually read. Students are told the same thing from their side — our guide for them, is Grammarly cheating?, opens with "the syllabus is the law."

Tier 1: Prohibited

UT Austin's sample language for the restrictive tier is representative, and notably concrete about consequences — AI-generated submissions "will be treated as plagiarism." UW–Madison's prohibited-tier sample likewise bans generative AI tools on graded work outright.

What this tier commits you to: deciding, in advance, what happens when you believe the policy was violated but the student denies it. A "treated as plagiarism" clause with no evidence plan defaults to detector scores — and Cornell's guidance is blunt about where that leads: "Detection tools claim to identify work as AI-generated, but they cannot provide evidence for that claim." If you choose this tier, write down your evidence standard in the same paragraph as the prohibition.

One scope decision to make explicitly: whether grammar tools count. Notre Dame's policy folds Grammarly into any gen-AI prohibition unless the instructor says otherwise; Birmingham's draws the line at generative features. Students genuinely cannot guess which reading you intend — say which you mean.

Tier 2: Allowed with disclosure

The middle tier — UW–Madison publishes sample language permitting AI use "with appropriate citation and disclosure" — is where most courses are landing, and the mechanics matter more than the permission. Tufts' CELT samples are useful because they specify how to disclose: a note accompanying the submission describing which tool was used, for what, and where its output appears in the work.

What this tier commits you to: defining disclosure precisely enough to grade it. "Disclose your AI use" invites a one-line footnote; "append the prompts you used and identify which passages began as AI output" produces something you can actually evaluate. The MLA-CCCC task force's version of this — making the AI transcript part of the submission — turns disclosure from a confession into a document.

Tier 3: Encouraged within limits

UT Austin's permissive sample treats AI as a course tool: students may use generative AI, with the expectation that they remain responsible for accuracy and originality of the final product. Courses that choose this tier usually pair it with assessment redesign so the grade attaches to what the student contributed — the design patterns are in designing writing assignments when detectors fail.

What this tier commits you to: grading judgment and integration rather than production, and saying so, since students will otherwise assume the old rubric still applies.

The clause every tier needs: evidence

Here is the uncomfortable symmetry. The prohibited tier needs evidence when a student denies AI use. The disclosure tier needs evidence when a disclosure looks incomplete. The permissive tier needs evidence when "responsible use" is contested. Every tier eventually arrives at the same question — what actually happened while this was written? — and no syllabus sentence answers it.

A policy is only as enforceable as its evidence plan, and the honest options are: detector scores (which the teaching centers themselves warn cannot provide evidence), staged drafts (self-reported, launderable), version history (records states, not authorship), or a provenance record created while the student wrote. If your statement invites or requires process evidence, consider naming what will count — in writing, in the syllabus — before the first assignment. Students who keep a sealed writing-provenance record can attach proof to any tier: the record shows typed, pasted, and AI-assisted text as it arrived, it belongs to the student, verification is open to anyone without an account, and it is free for individuals. A syllabus line as short as "process records are accepted as evidence in any integrity question" costs nothing and gives honest students a way to be believed — which, after the last three years of detector disputes, is the part of the policy they will remember.