Research doesn't follow a straight line — and neither does AI use. The phases below often overlap, repeat, and happen in any order. Use this as a menu, not a sequence.
Research Tasks
Norms vary sharply by field. Many researchers — more so in the humanities — object to using AI to summarize papers at all, and hold their graduate students to that standard. Before this becomes part of your workflow, find out what your department, your collaborators, and your advisor actually accept.
01
Literature Review & Discovery
Orienting yourself in a field
What AI can help with
Summarizing papers and identifying key claims
Finding gaps and contradictions across a body of literature
Interrogating a paper you have uploaded — whether the authors used a particular method, controlled for a given variable, or tested a population with a particular characteristic — where that detail sits in a technical appendix or footnote
Models hallucinate citations — always verify DOIs before use
Training cutoffs mean recent work may be missing entirely
Summaries can flatten nuance or misrepresent methodology
AI-generated search queries may reinforce existing biases
Uploading a published paper to a commercial tool may breach the publisher's licence — check your terms before sharing full texts
Example Prompt
I'm reviewing literature on [topic]. Here are 3 abstracts: [paste].
What are the key claims? Are there contradictions or gaps?
Suggest 3 search queries I haven't tried yet.
02
Research Design & Hypotheses
Structuring your inquiry
What AI can help with
Stress-testing hypotheses and assumptions
Brainstorming confounds and alternative explanations
Comparing study design approaches across fields
Reviewing methods protocols from adjacent disciplines
AI anchors on common designs — niche field-specific norms often missed
Brainstormed confounds may be plausible but domain-wrong
No IRB or ethics awareness — AI cannot substitute for institutional review
Example Prompt
I'm designing a study to test [hypothesis]. My approach: [describe].
Act as a skeptical reviewer: what are the 3 most likely confounds?
What alternative designs would address them?
Norms vary sharply by field. Writing code with AI is routine in some disciplines and contentious in others. Check what your field and your collaborators expect before relying on it, and be especially careful about assuming your students may do the same.
03
Coding & Scripting
Automating and building research tools
What AI can help with
Writing Python / R / STATA scripts from scratch or from a description
Debugging errors and tracing unexpected outputs
Understanding code someone else wrote — asking what a script does and why, rather than reading through it line by line
Translating code between languages (e.g., STATA → Python)
Automating repetitive data collection or file-processing tasks
Setting up reproducible analysis environments and pipelines
Generated code can look correct but produce silently wrong results — always test on known data
AI cannot reason about your specific dataset structure without seeing a sample
Version mismatches and deprecated APIs are common in generated code — check the docs
Do not paste private, IRB-restricted, or personally identifiable data into any commercial tool
Example Prompt
Here is a Python error I'm getting: [paste error + relevant code].
What is causing it? Fix the issue and explain what was wrong
so I understand and can avoid it next time.
Generated code may be syntactically correct but logically wrong — validate all outputs
Do not input sensitive or IRB-restricted data into commercial tools
Statistical interpretations can be confidently wrong — verify with a domain expert
Qualitative coding may miss cultural context and researcher positionality
Example Prompt
Here is the output of my regression in R: [paste].
Explain each coefficient in plain language.
Flag anything that looks unusual or worth investigating.
Norms vary sharply by field. Using AI to structure entire written sections is the use most likely to draw objection, particularly in the humanities, and journals and funders increasingly ask you to disclose it. Confirm what your field, your co-authors, and your target venue permit.
05
Writing & Communication
Shaping and sharing your findings
What AI can help with
Drafting and restructuring sections
Adapting writing for different audiences
Editing for clarity and concision
Drafting abstracts, cover letters, and lay summaries
AI prose tends toward over-polished generic voice — revise to preserve your own style
Factual claims can be smoothly stated but wrong — never trust without verification
Check your institution's and target journal's AI-disclosure policies
AI editing can inadvertently soften claims in ways that change your argument
In Practice
Structuring before drafting
Upload rough notes and ask for several possible organizational structures rather than prose. Evaluating options keeps you in the driver's seat — and you can converse back and forth about narrative structures, hooks, and how to articulate an argument, as you would with a colleague. Asking for a structure means you still have to write it, which also avoids going back and undoing AI voice and errors.
Here are my rough notes on [topic]: [paste].
Give me 3 distinct ways I could structure this as a [paper/essay/report].
For each, describe the narrative arc in 2 sentences and name the trade-off.
Don't write any prose — just the structures.
Ask it to position itself as a critical senior colleague who is territorial about the field, or as a non-specialist reader for a journal with wider readership. You can tune the persona to match the feedback you actually need before submission.
Here is my [section/abstract]: [paste].
Act as a senior reviewer in [field] who is skeptical and protective of disciplinary norms.
Give me your 3 most critical objections. Be direct and blunt — don't soften.
Example Prompt
Here is my methods section: [paste].
You are a reviewer from [adjacent field] with no deep background in my specialty.
Give me detailed feedback on clarity, structure, and accessibility — where did you
lose the thread, what terms need defining, what could be reordered?
Do not rewrite; give specific, actionable suggestions I can act on myself.
06
Grants & Proposals
Making the case for your work
What AI can help with
Tailoring proposals to funder priorities
Checking whether your framing and proposal sections meet call-specific requirements (management plan, timeline, broader impacts, etc.)
Strengthening Significance and Innovation sections
Check funder AI policy first — NEH (see NEH guidance), NSF (SBE, see policy), SSRC, Mellon, Ford, and Guggenheim, among others, may restrict or require disclosure of AI use.
AI-generated grant text can trigger AI-detection flags
Preliminary data sections should never rely on AI-generated figures
Simulated reviewer feedback is generic — it cannot replicate actual study section dynamics
Specialized grant tools go further than general-purpose AI — they search funder databases, track deadlines, and coach you section-by-section through proposal requirements. GrantedAI searches 133K+ foundations and 85K+ grants across all 50 states and 15+ countries, then drafts alongside you. GrantAI focuses on generating compelling narratives with minimal effort and is free to try for 7 days.
Example Prompt
Here is my Specific Aims draft: [paste].
Act as an NIH study section reviewer.
What are the 2–3 weakest points? Rewrite the opening paragraph
to hook reviewers immediately.