Skip to content 
Most researchers think AI is like Google: ask, get an answer.
It’s not.
You need to first find the right colleague— And you need to know how to talk to them.
When it comes to AI, the model you choose matters—but how you prompt it matters even more.
Different models have different strengths, and understanding how to craft your prompts for each can be the difference between getting an “okay” response and one that genuinely accelerates your work.
Over time, I’ve learned that the art of prompting is just as important as the models themselves.
Let me show you how.
1. GPT-4o: The Versatile Workhorse
Prompting Strategy: Megaprompts, Step-by-Step Instructions, and Custom GPTs
GPT-4o is my go-to model for most academic tasks. Why? It’s flexible, fast, and can handle a lot of information at once.
Think of GPT-4o as a reliable research assistant. The kind who thrives when you give them detailed instructions—whether that’s brainstorming ideas, summarizing complex papers, or even helping with coding. It’s the model I reach for when I’m juggling multiple tasks and need a versatile partner.
I) Megaprompts
Brainstorming a Research Idea (Collaborative Approach)
When I was planning a study on the relationship between obesity and psoriatic arthritis, I wasn’t sure where to start. I had a general idea but needed help refining it.
Here’s how I used GPT-4o to brainstorm:
Prompt:
“I’m interested in researching the relationship between obesity and psoriatic arthritis. Help me brainstorm potential research questions. Consider gaps in current literature, possible data sources, and study designs. Let’s focus on generating a few ideas that could be feasible for an observational study.”
The response? A range of thoughtful ideas. Some focused on treatment response, others on disease progression. I didn’t take the suggestions at face value, but they helped me think differently. They sparked ideas I wouldn’t have considered on my own.
Why It Worked:
- Broad Exploration: GPT-4o thrives when given room to explore different angles.
- Collaborative Flow: You’re not expecting a final product—you’re co-creating ideas.
- Multi-Dimensional Thinking: The model can suggest research questions, methodologies, and potential gaps all at once.
Follow-Up Prompts:
- “Can you expand on the idea about treatment response and suggest potential variables to measure?”
- “What would be the strengths and limitations of using electronic health record data for this question?”
This back-and-forth felt less like querying a machine and more like having a brainstorming session with a colleague.
II) Step-by-Step Instructions
Data Analysis
Prompt:
“Your are a top data analysis expert in R. You use tidyverse and best practices for coding in R. Comment on R codes that you provide liberally to explain what each piece does and why it’s written that way with a syntax where possible. Take a deep breath and let’s do this step-by-step:
Step 1: Import a CSV file in R and clean the data (remove missing values, correct data types).
Step 2: Perform descriptive statistics for all numeric variables.
Step 3: Conduct a logistic regression to assess the association between obesity and treatment response.
Step 4: Visualize the results with a bar plot showing odds ratios.”
Why It Works:
- The step-by-step format guides GPT-4o in handling each part of the task methodically.
- It reduces the risk of the model skipping important details.
✅ Best Practices for GPT-4o:
- Use megaprompts for complex tasks—include as much relevant detail as needed.
- Break down tasks with step-by-step instructions for multi-part workflows.
- Define the tone and audience for writing tasks to get tailored responses.
III) Custom GPTs: Supercharge Your Research with Personalized Models
One of GPT-4o’s standout features? You can create Custom GPTs tailored to your academic workflow. This is a game-changer, especially for repetitive tasks.
For example, I created a Custom GPT to assist with manuscript writing. I work on multiple papers, and providing the same instructions every time is tedious.
Instruction Prompt for My Custom GPT:
“You are an academic writing assistant specialized in clinical research. When asked to review a manuscript draft, focus on improving clarity, coherence, and logical flow. Highlight gaps in argumentation, suggest stronger transitions, and provide constructive feedback on the introduction and discussion sections. Use an academic tone suitable for rheumatology journals.”
Why It Works:
- Consistency: The model remembers my preferences for tone and style.
- Efficiency: I don’t have to repeat instructions every time I need feedback.
- Tailored Support: Whether it’s summarizing literature, analyzing data, or refining manuscripts, my Custom GPT is tuned to my specific needs.
This level of customization? Impossible with o1 or o3. They simply don’t support Custom GPTs.
2. o1/o3 Reasoning Models: The Focused Problem-Solvers
While GPT-4o is my generalist, the o1 and o3 models are my specialists. They’re designed for deep reasoning and complex problem-solving.
But here’s the key: they don’t like long, detailed prompts.
In fact, according to OpenAI, complex prompts can confuse o1 and lead to worse results.
Think of o1 as that brilliant colleague who doesn’t need all the background—they just want to know the problem so they can solve it. Quickly.
Example: Debugging R Code (Focused Inquiry)
I was working on an R script for hierarchical clustering. The dataset was large, and the code was painfully slow. I’d tried several tweaks with GPT-4o, but the performance issues persisted.
So, I switched to o1. Here’s what I asked:
Prompt:
“This R code for hierarchical clustering runs slowly on large datasets. Identify performance bottlenecks and suggest optimizations.”
Within seconds, o1 pinpointed the issue. It suggested optimizing the distance matrix calculation—a detail I had completely overlooked. It even provided a more efficient code snippet.
Why It Worked:
- Concise Focus: The prompt zeroed in on the problem without unnecessary details.
- Reasoning-Driven: o1 doesn’t need hand-holding. It jumps straight into problem-solving.
- Iterative Depth: I followed up with, “How does this optimization improve performance?” to deepen my understanding.
How to Prompt o1/o3 Effectively:
- Keep It Short: Focus on one specific problem at a time.
- Be Goal-Oriented: Clearly state what you need—don’t overload with context.
- Iterate Instead of Overloading: Break down complex tasks into smaller parts. Build on responses through follow-ups.
Deep Research: A New Frontier for Academic Exploration
Within the o3 model, OpenAI has recently launched a new feature called Deep Research, designed to elevate how we engage with complex academic content.
Choose o3-mini model and then click on Deep Research.
This feature is currently available for Pro users but is planned to be rolled out to Plus and Free users in the coming weeks.
So, what’s Deep Research?
Think of it as ChatGPT’s ability to perform advanced literature analysis, synthesize large volumes of information, and identify research gaps more effectively than ever before. It’s like having an AI-powered research assistant that doesn’t just summarize papers—it helps you connect the dots across studies.
Example: Conducting a Literature Review with Deep Research
Say I want to write a paper on psoriatic arthritis phenotypes, and I need to synthesize recent findings from multiple studies on the topic as part of my literature review before sitting down to write. Instead of manually going through each paper, I can use Deep Research.
Prompt:
“Analyze the current literature on psoriatic arthritis phenotypes. Identify key trends, controversies, and gaps in the research from studies published in the last five years.”
What I Get:
- A concise synthesis of major findings.
- Identification of areas where evidence was conflicting.
- Suggestions for potential research questions based on identified gaps.
This isn’t just summarization—it’s strategic insight generation.
Why Deep Research Matters:
- Time-Saving: Quickly process large bodies of literature.
- Insightful: Uncovers patterns and gaps that might not be obvious from a single study.
- Focused: Helps researchers identify where to contribute new knowledge.
Side-by-Side Comparison: GPT-4o vs. o1/o3 with Deep Research
| Aspect | GPT-4o (Generalist) | o1/o3-mini (Reasoning Models with Deep Research) |
|---|---|---|
| Prompt Length | Handles long, detailed prompts (megaprompts) | Performs best with short, concise prompts (o1); troubleshooting code (o3-mini-high), supports deep analysis (o3-mini) |
| Task Type | Brainstorming ideas, outlining projects, Custom GPTs | Solving complex problems, refining hypotheses, synthesizing literature (o3-mini) |
| Complexity Handling | Excels at multi-step tasks with detailed instructions | o1 prefers simple tasks; o3-mini handles complex research synthesis via Deep Research |
| Custom GPTs | Can create detailed, personalized Custom GPTs | Cannot be used for Custom GPT creation (yet) |
| Availability | Available to all users | Deep Research currently for Pro users; expanding soon to Plus & Free |
| Example (Research Idea) | “Brainstorm potential research questions on obesity and PsA.” | “Identify limitations of using BMI as an exposure in PsA studies.” (o1) / “Analyze recent trends in obesity-related PsA outcomes.” (o3 with Deep Research) |
Pro Tip: When to Switch Models
Sometimes, even with the perfect prompt, a model hits a wall. Here’s how I handle it:
- Start with GPT-4o for broad brainstorming, outlining, and Custom GPT creation.
- Switch to o1 when facing complex coding problems or data analysis logic issues—keep prompts short and focused.
- Use o3 when I need deep critical thinking, such as evaluating research methodologies or refining hypotheses.
For example, when drafting a grant proposal, I start with GPT-4o to outline the structure. For complex statistical considerations, I switch to o1. And when I need to critically evaluate research gaps, o3 is my go-to.
Final Thought
Prompting isn’t just about crafting the perfect sentence—it’s about thinking like a researcher.
Frame problems clearly, define your goals, and use each model strategically.
With GPT-4o:
- Think big. Use megaprompts, explore ideas broadly, and create Custom GPTs for recurring academic tasks.
With o1/o3:
- Think sharp. Keep it concise, goal-oriented, and iterative for deep reasoning and problem-solving.
Master this, and ChatGPT becomes not just a tool—but an essential part of your academic research workflow.
