A Columbia Prompt Library for Effective Academic AI Use
Introduction
Use this prompt library to get practical, consistent outputs for academic work from large language models (LLMs). We encourage you to make a copy of this document for your own purposes – it’s designed for quick adaptation, and you should cut and add content as you find useful.
This collection centers on one composite template, followed by a library of prompts for individual use cases categorized by teaching, research, and administration.
AI Prompting Techniques and Principles
Our prompt library was built from a handful of AI prompting techniques and principles that you should incorporate into your work:
- Be explicit about the task, audience, and constraints
- Include relevant context or short examples
- Specify the output format so it’s paste-ready; add guardrails (“no fabrication,” allow “Unknown,” cite real sources)
- Treat every output as a draft: verify facts, keep sources, and adapt language for your discipline and audience.
- Check for bias, privacy, and policy compliance (including the Provost’s Office AI Policy and any other AI policies that may be specific to your work)
Sources that inspired these principles can be found below. Explore our AI Prompt Design and Advanced Prompt Engineering articles for deeper technique.
If there’s one takeaway from this library, it’s this: use AI to help you write the prompt itself.
Treat the model as a co-author. Ask it to draft a first version, point out what’s missing (audience, length, format, constraints, etc.), and then refine together. This is faster than hand-crafting the “perfect prompt” and it surfaces blind spots you’d miss on your own.
Use AI in two simple ways: polish the output when it’s close, and fix the instructions when the model keeps doing the wrong thing.
When the output is almost there, use:
Tighten the output by fixing these gaps: [list]. Keep length and format constraints. Improve clarity for [audience].
When the model’s behavior misses the mark, use:
Here's a prompt: [PROMPT]
The desired behavior from this prompt is for the agent to [DO DESIRED BEHAVIOR], but instead it [DOES UNDESIRED BEHAVIOR].
While keeping as much of the existing prompt intact as possible, what are some edits/additions that you would make to encourage the agent to more consistently address these shortcomings?
One Composite Template
ROLE: You are a [expertise + experience] who [approach/values].
OBJECTIVE: Produce [deliverable] that [measurable/observable goal] for [audience]
SCENARIO: [Relevant context constraints, stakeholders, resources].
EXPECTED SOLUTION:Return [format: table | JSON | outline | prose] including [must‑have elements] and excluding [off‑limits].
STEPS: [short process the model should follow].
GUARDRAILS: No fabrication; cite only real sources; say "Unknown" if missing; flag assumptions and time‑sensitive claims.
QUALITY BAR: End with 3 risks, 3 things to verify, and confidence 0–1.
This composite template is adapted from Mindsteam’s ROSES structure. We extended it for academic use by: (a) baking in anti-fabrication guardrails and uncertainty handling, (b) requiring explicit output formats (table/JSON/prose) so results are paste-ready, and (c) adding a lightweight self-check to surface risks and verification needs.
Prompts by Academic Category
This prompt library is broken down into three categories: (A) Teaching & Learning, (B) Research, and (C) Administration. These categories reflect real use cases and consults that the Emerging Technologies team sees and supports across Columbia. We prioritized patterns that recur, are high-impact, and benefit most from structure and guardrails.
A) Teaching & Learning
ROLE: Experienced instructor in [discipline].
OBJECTIVE: Create a lecture plan on [topic] for [course level].
SCENARIO: Students have [prereqs]; class size [N].
EXPECTED SOLUTION: Sections: learning objectives (Bloom level), timed agenda, demo/activity, 3 assessment questions + keys, accessibility tips (UDL), 2 misconceptions.
STEPS: Draft → check timing → add misconceptions → append checklist.
GUARDRAILS & QUALITY BAR as above.
ROLE: Experienced assignment designer in [discipline].
OBJECTIVE: Design an assignment where students [produce X].
SCENARIO: Due in [weeks]; collaboration policy [policy].
EXPECTED SOLUTION: Goals (3–5); instructions; milestones; 4‑level analytic rubric (criteria + descriptors); "AI usage policy" for this task.
STEPS: Align to goals → rubric → integrity notes.
GUARDRAILS & QUALITY BAR.
ROLE: Experienced assessment designer in [discipline].
OBJECTIVE: Generate 10 items on [topic] at [difficulty].
EXPECTED SOLUTION (TABLE): Item#, Question, Correct Answer/Key, Learning Objective (Bloom), Common Distractors.
GUARDRAILS: Avoid ambiguity; one best answer for multiple-choice questions
ROLE: Experienced instructor in [discipline].
OBJECTIVE: Provide formative feedback.
SCENARIO: Rubric: [paste/attach]; Draft: [anonymized excerpt].
EXPECTED SOLUTION: 3 strengths, 3 prioritized improvements, 2 next actions, 1–2 sentence summary
B) Research
ROLE: Expert research specialist.
OBJECTIVE: Propose search strategy on [topic].
EXPECTED SOLUTION: Questions, inclusion/exclusion, databases, screening plan, keywords. No invented sources (return quality confidence)
ROLE: Expert reviewer.
OBJECTIVE: Create screening summaries.
EXPECTED SOLUTION (JSON only):{"title","year","study_type","population","intervention_or_focus","main_finding","limitations","relevance":"high|medium|low","notes"}. Use null for unknowns.
ROLE: Expert methodologist.
OBJECTIVE: Compare [Method A] vs [Method B] for [goal].
EXPECTED SOLUTION: Decision table covering assumptions, strengths/limits, data needs, failure modes, cost, interpretability; 150‑word recommendation for [context].
QUALITY BAR.
ROLE: Expert research ethicist.
OBJECTIVE: Draft IRB pre‑planning checklist for [study].
EXPECTED SOLUTION: Participant risks, data handling, consent language notes, de‑identification, retention schedule. Plain language.
ROLE: Experienced privacy reviewer.
OBJECTIVE: Scan text for potential privacy issues.
EXPECTED SOLUTION: Findings by risk level with safer rewrites.
QUALITY BAR.
C) Administration
ROLE: Expert operations manager.
OBJECTIVE: Draft a standard operating procedure for [process].
EXPECTED SOLUTION: Scope, roles, step‑by‑step, decision points, risks, and templates appendix. Mark compliance requirements.
QUALITY BAR.
ROLE: Expert facilitator in [subject].
OBJECTIVE: Turn goal list into timed agenda + minutes template.
EXPECTED SOLUTION: Agenda with owners/outcomes; minutes template with decisions, action items, deadlines.
ROLE: Experienced event planner.
OBJECTIVE: Create brief for [audience/size/purpose].
EXPECTED SOLUTION: Objectives, program outline, speaker criteria, comms plan, budget ranges, accessibility plan, risk register + milestone table.
ROLE: Expert communications lead.
OBJECTIVE: Draft an email to [recipient] about [issue].
EXPECTED SOLUTION: Subject, preview text, 150–200 words, clear ask, deadline, next steps; firm and empathetic tone.
QUALITY BAR.
General Purpose Prompts
Act as a skeptical reviewer for [venue/audience]. List the 5 strongest counter‑arguments or failure modes. Revise the draft to address the top 3.
End every output with: 3 biggest risks or likely errors, 3 items to verify externally, Confidence (0–1)
Identify ways this prompt/output could cause harm, bias, or policy violations in an academic setting. Propose safer wording and guardrails.
Read More
Prompt Engineering
CUIT - Advanced Prompt Engineering