Responding to a suspected case of undisclosed AI use in student work

Context

A university or post-secondary instructor who has already received a piece of work and observes specific signs that it was generated entirely by AI without disclosure. Used when there are concrete suspicions and the instructor must decide whether to have a preliminary conversation first or open a formal process, how to document the case, and how to give the student formative feedback even if a sanction is applied.

Paste first

Before opening the model, have ready:

• The text of the work, with the student's name and student ID removed (replace with "Student A" or similar).
• The articles of your institution's academic integrity policy that apply to this case, copied exactly.
• The AI use clause from the syllabus, if any, copied verbatim; if there is none, state that explicitly.
• The original assignment prompt and the rubric criteria.
• The specific signs you observed, written in detail: what has changed compared to the student's previous writing style, what is missing, what seems unusual, and whether you used an automatic detector and what it returned.

Privacy rule: Never paste the student's full name, institutional email, student ID, or disciplinary history into the model. Use a pseudonym ("Student A"). If you describe academic history, do so in terms of patterns ("the student's two previous essays had recurring subject-verb agreement errors"), not as grades or data that could identify them. Applicable data protection laws — never share students' personally identifiable information with a third-party LLM; if in doubt, consult your institution's student data privacy policy before pasting any material.

Act as an academic integrity advisor with experience in cases of undisclosed AI use in higher education. I have received a piece of work that shows specific signs that it may have been generated entirely by AI without disclosure, and I need guidance to proceed fairly, with documentation, and in a way that is formationally valuable.

**Institutional context:**
- Institution and type: {{institution name}} — {{university / community college / technical institute — public or private}}
- Course and level: {{course name}} — {{level, e.g. "third year History Education"}}
- Type of work assessed: {{type of work, e.g. "1,500-word argumentative essay"}}
- Academic integrity policy applicable: {{name or articles of the policy, or "we have no written policy"}}
- AI use policy declared in the syllabus: {{paste the exact syllabus clause, or "no clause exists"}}

**Signs I observe in the work (no student identifiers):**
{{list the signs in specific detail — for example: "uniformly correct prose with no agreement errors, unlike previous submissions which had recurrent errors; no references to local or situated examples despite the prompt explicitly requiring them; no personal voice or reflection even though the rubric valued it; identical paragraph structure across all sections; automatic detector X returned Y%"}}

**What I need:**

1. An honest assessment of the signs: are they sufficient to open a formal process, or do they only justify a preliminary conversation? Include alternative explanations that do not imply AI use for each sign, so I can reason carefully before acting.

2. A conversation script: what to say, what to ask, how to document the student's responses without pressuring them or suggesting what to answer.

3. A written record template for the institutional file: what data to include, in what order and format, so the record is procedurally valid if the case escalates.

4. Formative feedback addressed to the student, ready to send, useful regardless of the process outcome: what to learn from this situation about appropriate AI use in professional training.

5. The minimum procedural steps under your institution's academic integrity policy, with an explicit warning about what not to do to avoid exposing the student or the institution to an invalid process through procedural error.

Do not treat the signs as definitive proof. Make the reasoning explicit before each recommendation.

Expected output

**Assessment of the signs:** none of these signs, individually or together, constitutes proof of AI use. The abrupt change in error profile carries high weight: it requires explaining not only the new quality but also the disappearance of specific, previously recurrent patterns. The absence of references to the teaching practicum the prompt explicitly required carries very high weight: a language model has no access to the student's real experience and produces exactly this pattern. The Turnitin result (87%) is a statistical indicator, not expert evidence — do not use it as the sole basis for a formal complaint.

The combined picture justifies a **preliminary conversation**, not a formal complaint. If the student cannot explain the writing process in detail, the evidence is sufficient to escalate the case.

**Conversation script — open questions about the writing process (block 1):**
- "Can you walk me through how you organized your time to write this essay — when you started and how you worked on it?"
- "Do you have any drafts — in a document, handwritten notes, anything? Could you share one?"

The direct question about AI is asked only if responses to block 1 and block 2 (about the content of the text) produce inconsistencies.

**Formative feedback (excerpt):** "Generative AI tools can be valuable partners in academic work: finding references, reviewing drafts, organizing ideas. This course has one rule: any use must be disclosed. That rule is not arbitrary — this essay exists so you can build your own argument, not to produce correct-sounding text."

Also included: a full institutional record template and minimum procedural steps with explicit warnings about errors that invalidate the process.

Watch out for

• Automatic AI detectors (Turnitin AI, GPTZero, ZeroGPT) have significant false-positive rates, particularly with formal academic writing. A high percentage is one more sign, not proof. If the model gives heavy weight to the detector result, correct it: ask it to evaluate the signs without that data and see whether the conclusion changes.
• The model may propose procedural steps that sound reasonable but do not match your institution's specific policy. Verify every step against the exact text of your institution's academic integrity policy before acting; a procedural error can invalidate the entire process.
• The model does not have the student's work unless you paste it. If you did not include the text in the prompt and the suggested conversation questions do not fit the type of work, run again including the text (with all identifying information removed).

Suggested iteration

If after the conversation the case remains unresolved, ask: "The student provided this explanation: [describe in general terms, no identifying information]. Does this change your assessment of the signs? Should I proceed with a formal process or close with a note in the file?"

If you need to draft the formal notice of process opening, ask: "Draft the notification letter to the student according to the section of the policy I provided on [topic], using institutional tone and without anticipating the outcome of the process."