AI Toolkit for PhD and Master's Students

Part 2: 10 Prompting Methods for Research and Applications

These techniques are grounded in published prompting research — ELI5 Laddering, Tree-of-Thoughts, Chain-of-Density, Chain-of-Verification, and ReAct are all documented in the literature. Each includes a copy-paste template adapted for graduate research tasks.

1) ELI5 → Ladder Up

What it does: Builds from intuition to rigor. When to use: Learning a new topic, clarifying a messy idea before writing.

Try this:

Explain CRISPR in three passes:
1) ELI5
2) First-year master's level
3) PhD-seminar level with 3 citations to verify keywords
Keep each pass ~120 words.

2) Inverse Prompt

What it does: Surfaces pitfalls by asking for the worst version first, then reverses them. When to use: SOP polish, research statements, cold email drafts.

Try this:

Write a terrible 150-word research-interest paragraph in computational social science.
List why it's terrible.
Now write the optimal version that avoids those issues.

3) Self-Consistency Voting

What it does: Generates multiple solutions and has the model choose the best with reasoning. When to use: Quant problems, study designs, logic puzzles, GRE/GMAT/CASPer prep.

Try this:

Solve this statistics problem with 5 independent solution paths.
Then select the most consistent final answer and justify the choice in 3 sentences.

4) Tree-of-Thoughts Sprint

What it does: Branches ideas, prunes weak paths, doubles down on the strongest. When to use: Research design, grant strategy, literature-review structures.

Try this:

Propose 3 distinct approaches to test hypothesis H.
For each: outline methods → risks → mitigations (2 levels deep).
Prune to the best approach and justify with criteria.

5) ReAct-in-the-Loop

What it does: Mixes reasoning with explicit "actions" (what to check, where to look) to reduce guesswork. When to use: Replication plans, protocols, data-prep checklists.

Try this:

Goal: draft a replication plan for Study X.
Think step-by-step, then list the actions needed to verify each step
(e.g., datasets, parameters, inclusion criteria).
Output: plan + action checklist.

6) Skeleton-of-Thought (SoT)

What it does: Forces an outline first, then expands each bullet for clarity and speed. When to use: SOPs, methods sections, lit reviews, talk outlines.

Try this:

Create a 7-bullet outline for my SOP
(problem, gap, aim, methods, results potential, fit, future work).
Then expand each bullet to 120–150 words. Keep tone confident, evidence-driven.

7) Chain-of-Density Summarization

What it does: Keeps summaries the same length while iteratively adding missing key entities. When to use: Paper triage, seminar prep, advisor briefings.

Try this:

Summarize this paper in 120 words.
Perform 4 density passes: each pass adds the most important missing entities/terms
without increasing length. Mark new entities in **bold**.

8) Chain-of-Verification (CoVe)

What it does: Draft → generate verification questions → answer them → output a corrected final. When to use: Bios, related work, personal websites, grant aims pages.

Try this:

Draft a 200-word research bio.
Generate verification questions for names, dates, grants, and methods.
Answer them.
Produce a corrected bio and include a one-line change log.

9) Socratic Trellis

What it does: Pressure-tests your claim with layered questions and counter-evidence, then refines it. When to use: Thesis statements, argument sections, literature synthesis.

Try this:

Claim: "Technique X improves outcome Y in low-resource settings."
Ask 10 Socratic questions across assumptions, mechanisms, evidence, and counter-examples.
Rewrite the claim and add a testable prediction.

10) Persona + Constraints Matrix

What it does: Locks in role, audience, voice, scope, and a scoring rubric; then self-grades and revises. When to use: Faculty outreach emails, scholarship essays, cover letters.

Try this:

Role: admissions committee member
Audience: faculty reviewers
Voice: confident, specific, collaborative
Constraints: 650 words, 2 research vignettes, no clichés, explicit program fit
Rubric: clarity(30), originality(30), fit(40)
Draft → self-grade → revise to score ≥90.

Quick Combos

Lit Review: Method 6 (SoT) for structure → Method 7 (Chain-of-Density) to pack key entities into summaries.

SOP Polish: Method 2 (Inverse Prompt) to expose clichés → Method 10 (Persona+Constraints) to align with reviewers → Method 8 (CoVe) to verify every proper noun.

Methods Clarity: Method 4 (ToT) to compare designs → Method 3 (Self-Consistency) to cross-check calculations.

Advisor Outreach: Method 1 (ELI5 Ladder) to distill your ask → Method 9 (Socratic Trellis) to anticipate objections and tighten the message.

If you're newer to structured prompting and want a simpler starting point, the SPARK, GUIDE, and FOCUS frameworks in 5 AI Prompts for PhD Students are a good warm-up before working through the techniques above.

---

Part 3: 7 ChatGPT Task Automations

ChatGPT Tasks let you schedule a prompt to run on a recurring schedule — delivered via push or email. This section is about replacing manual recurring work with lightweight background automations.

Setup: In any chat, describe what you want and when ("Every weekday at 9:00 AM ET, summarize..."). Confirm the proposed schedule and the task is saved. Manage it at Profile → Tasks.

Note on document links: ChatGPT Tasks can browse public URLs but cannot authenticate to private files (Google Docs, Notion, etc.). For automations that reference a [doc link], either make the document publicly accessible via link, or paste the relevant content directly into the task prompt instead.

Best-practice setup before you start:

• Be explicit about timing and timezone ("9:00 AM ET")
• Name the output shape ("5 bullets + links," "markdown table," "≤200 words")
• Pin references (links to papers, spreadsheets, syllabi)
• Add guardrails ("include sources," "cap at 120 words")
• Iterate in place after a run ("make it weekly," "add competitor labs")

---

Automation 1: Daily Lit Sweep

What it does: Curates fresh papers from your venues with one-line takeaways and links.

Paste-this script:

Weekdays at 7:30 AM ET, scan major venues in computational biology
(Cell, BioRxiv, ISMB) for 5 noteworthy papers published or updated in
the last 7 days. Output bullets with title | authors | 1-sentence takeaway | link
and a final "Read First" pick.

Why it works: A crisp briefing with sources and a decision cue — you avoid doomscrolling and always have one high-priority read.

---

Automation 2: Advisor 1:1 Prep

What it does: Turns scattered notes into a tight agenda you can send ahead.

Paste-this script:

Fridays 3:00 PM ET, turn this week's chat notes + [lab doc link] into a
≤200-word advisor update: wins, blockers, decisions needed, deadlines next week.
Add 3 agenda questions.

Why it works: Consistent, professional updates. Your advisor knows you're prepared; you stop dreading the meeting.

---

Automation 3: Lab Data Pulse

What it does: Pulls a summary table of metrics and commentary from your experiment log.

Paste-this script:

Daily 6:00 PM ET, read [experiment log link] and output a table with
Runs | Pass% | Anomalies | Notes, followed by 3 bullets: trends, outliers,
suggested next test.

Why it works: Forces signal over noise — perfect for stand-ups and lab notebook hygiene.

---

Automation 4: Writing Sprint Booster

What it does: Generates a structured daily writing plan for your current chapter or paper.

Paste-this script:

Weekdays 9:30 AM ET, generate a 90-minute writing plan for the dissertation
chapter in [doc link]: goal, 3 mini-milestones, a 30-min warm-up
(outline or figures), 2 checkpoints, and a carry-over list.

Why it works: Light structure beats vague intentions. Momentum compounds — a carry-over list means each day starts where the last one ended.

---

Automation 5: Funding & CFP Radar

What it does: Scans for calls and grants inside your application window.

Paste-this script:

1st business day monthly, 9:00 AM ET, list new grants/CFPs relevant to
[discipline] due in 60–120 days. Output a table:
Program | Amount | Deadline | Link | Eligibility | Fit (H/M/L)
and add 3 next-step suggestions.

Why it works: You stop discovering deadlines after they pass. Pair this with Fellowships for Doctoral Degree Programs for a structured overview of the major programs worth tracking.

---

Automation 6: Teaching & Section Plan

What it does: Turns lecture slides and forum threads into a ready-to-run section plan.

Paste-this script:

Sundays 5:00 PM ET, from [slides link] and [forum link], create a 10-minute
section plan: learning goals, 3 discussion prompts, 2 quick checks,
1 extension for advanced students. Keep it ≤150 words.

Why it works: Time-boxed prep with built-in pedagogy. Consistent quality without the Sunday-night scramble.

---

Automation 7: Job-Market Tracker

What it does: Monitors postings and compiles a shortlist with fit tags.

Paste-this script:

Mondays 8:00 AM ET, scan [job boards or dept links] for TT/postdoc roles
in [field]. Output a table:
Institution | Role | Area | Deadline | Link | Fit (H/M/L) | Why,
then 3 tailored pitch angles I could use.

Why it works: Keeps your search cadence steady and applications targeted. You don't miss deadlines and you stop applying to roles that aren't a genuine match.

---

Power-User Layer

• Bundle in Projects: Keep recurring tasks, reference files, and prompts together so each run uses the same context.
• Cross-device: Verify notification permissions on web, desktop, and mobile so results reach you the moment runs complete.
• Troubleshooting: If a task misses, re-enable push/email notifications and re-run from Profile → Tasks. To edit, reply in the original chat thread ("make this Fridays at 4 PM") or use the ✏️ icon on the Tasks page.

---

Where to Start

If you're new to this, don't try to implement everything at once. Pick one item from each layer:

1. Strategy — Run a semantic search on a topic you're currently reviewing and compare what surfaces versus your usual keyword approach.
2. Prompting — Use the Inverse Prompt (Method 2) on something you've already written this week. The gap between the "terrible version" and your draft is useful data.
3. Automation — Set up the Funding & CFP Radar. It takes five minutes and runs itself every month.

Once those three feel natural, the rest of the toolkit slots in around actual needs — not as a system to maintain, but as tools you reach for when the problem fits.

For finding faculty whose active research aligns with your direction, Streamlined AI does that mapping so your outreach starts from a stronger position.

---

Amos Oppong is an entrepreneur leveraging AI to solve problems in academic research and graduate school access. He is the founder of Streamlined AI.