Navigating Generative AI Thesis Policy

Before using any AI tool on thesis work, classify the intended use into one of three tiers:

1. Generally Permitted (check department policy): grammar/language editing, translation, formatting citations/equations, coding assistance, brainstorming research questions.
2. Permitted Only With Prior Approval: summarizing literature, refining structure/argumentation feedback, research support tools (e.g., citation discovery).
3. Almost Universally Prohibited: generating original thesis text (arguments, conclusions, results), interpreting findings, fabricating/altering data, writing full chapters or literature summaries verbatim.

If in doubt, default to disclose and ask the supervisor first — this is the one rule nearly every university shares.

Progress:

• Step 1: Identify institution's specific GenAI thesis policy (check graduate school + department level — they can differ)
• Step 2: Classify each intended AI use against the three-tier framework above
• Step 3: Get supervisor/committee approval in writing before using AI for anything beyond basic editing
• Step 4: Use AI only within approved bounds; keep a log of tool, prompt purpose, and output use
• Step 5: Draft the mandatory disclosure statement (tool name, version, purpose, chapter/section affected)
• Step 6: Verify all AI-assisted or AI-suggested facts, citations, and data manually — never trust AI output for accuracy
• Step 7: Submit disclosure alongside thesis per institutional format requirements

If the thesis topic IS generative AI in content creation (research topic, not tool use)

Progress:

• Step 1: Narrow scope — pick an angle: marketing/social media performance, educational content, brand perception, academic integrity, or ethics/regulation
• Step 2: Choose methodology — empirical case study (quantitative metrics like views/followers/production time), qualitative interviews, survey (e.g., Technology Acceptance Model), or policy/document analysis
• Step 3: Ground the thesis in existing literature (content creation theory + AI theory + regulatory frameworks)
• Step 4: Define measurable outcomes if empirical (e.g., engagement %, production time delta, learning curve effect)
• Step 5: Address ethical dimensions explicitly (authorship ambiguity, bias, misinformation, copyright, data privacy)
• Step 6: Draw conclusions tied back to practical/professional implications (e.g., professional identity, workload impact)

Example 1: Input: "Can I use ChatGPT to write the literature review section of my thesis?" Output: Not permitted at most institutions. ChatGPT may be used to find or organize sources and brainstorm themes, but writing the literature review's actual prose is classified as "original thesis text" generation — prohibited under policies like University at Buffalo, Bristol, and UNC. Use AI only to summarize your own notes for personal understanding, then write the section yourself. Disclose any AI assistance used in the research/organization phase.

Example 2: Input: "I want to study how marketing agencies use generative AI tools like Midjourney and ChatGPT for content creation." Output: Frame as an empirical/qualitative mixed study. Research questions: (1) What tasks do professionals delegate to GenAI (brainstorming vs. final output)? (2) How does professional identity affect adoption (Technology Acceptance Model)? (3) Does AI use increase or decrease total workload (accounting for output-verification time)? Methodology: interviews with marketers + survey instrument; measure outcomes like engagement delta and production-time delta, as in comparable published theses (e.g., 8.57% view increase, 15–28% production-time reduction with learning curve).

Example 3: Input: "My advisor hasn't given explicit AI guidance — what should I do?" Output: Do not assume permission. Email the advisor/committee requesting written guidance before using AI for any milestone work (per practices at UT Austin and SFU). Proceed only with editing/formatting-level use until you receive explicit approval, and document the request/response for your records.