Types of AI guidelines, with examples and evidence

Type 1

Institution-wide default rule

A campus rule that applies when an instructor has not set a course policy, so students are never left without guidance.

Stanford University (guidance in force since 2023) Without a clear instructor statement, using generative AI is treated like getting help from another person. Instructors may set their own course policies. Stanford Generative AI Policy Guidance ↗

Evidence

180 / 200 of the world’s top-ranked universities published GenAI guidelines when surveyed in April 2026. Top 200 universities study (2026) ↗

3 nations U.S., Japanese and Chinese guidelines alike cast instructors as gatekeepers working within adaptive policy frameworks. Li et al. (2026) ↗

Type 2

Syllabus statement templates

Ready-made course policy language, usually in several versions from restrictive to permissive, that instructors adapt.

Harvard FAS Faculty must tell students the course AI policy. Draft language is provided for a restrictive policy, a fully encouraging one, and a mixed approach. Harvard FAS Generative AI Guidance ↗

Evidence

40–45% of syllabi in a 31,000-syllabus study still had no explicit AI policy in Fall 2025 (explicit rules reached about 55%), so templates have a gap to fill. Chirikov (2026) ↗

35 syllabi at one university showed institutional guidance only reaches students once instructors translate it into course rules. Wu (2026) ↗

Type 3

Assessment-level scales

Instead of one rule per course, each assignment gets a level that says how much AI is allowed.

AI Assessment Scale A five-level scale from No AI to AI Exploration, now being adapted for specific fields such as English for academic purposes. AI Assessment Scale ↗

Evidence

Task-based Instructors increasingly restrict AI for drafting while permitting it for editing and proofreading: rules are becoming task-specific. Chirikov (2026) ↗

2026 Adaptation of the AI Assessment Scale for English for Academic Purposes, with level-by-level task designs. Roe et al. (2026), TESOL Journal ↗

2024 Foundational: an early pilot at British University Vietnam reported fewer misconduct cases and higher pass rates after adopting the scale. Furze et al. (2024) ↗

Type 4

Disclosure & attribution

Rules for when and how students acknowledge AI use, so permitted use stays transparent.

Stanford University Students should acknowledge any non-incidental AI use and default to disclosing when in doubt. Stanford Generative AI Policy Guidance ↗

Evidence

94% of UK undergraduates use generative AI for assessed work, yet many remain unsure what is acceptable. HEPI (2026) ↗

Vague Students described university AI policies as vague, hard to locate and not specific to their programme or assessment. Sun et al. (2026) ↗

Type 5

AI-detection guidance

Guidance on whether and how to use AI-writing detectors, and what a flag may or may not be used for.

Curtin University (2026) Switched off Turnitin’s AI detection across all campuses from 1 January 2026 while keeping originality checks. Curtin announcement ↗

Dartmouth College (Sept 2026) Authorized one detector under strict conditions: written notice to students and no action on a score alone. Dartmouth guidelines (PDF) ↗

Evidence

3 of 4 detectors had no false positives on human papers, but detection dropped considerably on hybrid and humanized text. Van Vlasselaer et al. (2026) ↗

Type 6

Data privacy & approved tools

Rules on which AI tools are approved and what information, especially student work and records, must never be entered into them.

Harvard University University-wide guidelines focus on protecting confidential data when using generative AI tools. Harvard guidelines for generative AI tools ↗

Evidence

38% of students say their institution provides AI tools, up from 23% in 2025 and 9% in 2024. HEPI (2026) ↗

8 areas A global expert panel placed privacy and equitable access among eight core areas for GenAI policy. Crompton et al. (2026) ↗

Type 7

Assessment redesign guidance

Advice on rewriting assignments so that they still measure learning when AI is available.

EDUCAUSE Horizon Report (2026) Identifies a move toward authentic, process-based demonstrations of learning as AI complicates traditional assessment. 2026 Horizon Report ↗

Evidence

438 faculty and staff who design assessments were surveyed; respondents want to choose when AI is used and called for clearer policies and AI literacy. EDUCAUSE (2026) ↗

Changed Most students say assessment has changed significantly in response to AI. HEPI (2026) ↗

Type 8

Sector-wide principles & AI literacy

Shared principles across many universities, often committing them to build AI literacy for students and staff.

Global Delphi panel (2026) Experts from six continents agreed on eight core areas for university GenAI policy, starting with academic integrity and ethical use. Crompton et al. (2026) ↗

Russell Group (UK, 2023, foundational) Five early principles adopted by 24 universities, including support for AI literacy and equal access. Russell Group principles (PDF) ↗

Evidence

2024→26 Three survey waves show AI moving from novelty to routine, so literacy can no longer be an introductory orientation. From Novelty to Normalisation (2026) ↗