You wouldn’t email your collaborator with zero background.

So why do that to ChatGPT?

Most researchers still treat ChatGPT like a search bar:

Type a question. Hit enter. Hope for the best.

But ChatGPT isn’t a search engine.

It’s a conversation partner.

And like any good conversation, the quality of the answer depends on the context you provide.

Context Isn’t Fluff. It’s the Secret to Getting Useful Output.

When you skip context, you get responses that are vague, generic, or just off-base.

But when you include just a bit of background—your role, your audience, your goal—you get outputs that are tailored, relevant, and far closer to what you actually need.

It’s the difference between:

❌ “Write a summary of this paper.”

✅ “Summarize this paper for the audience who specializes in epidemiology but not AI. The study explores the use of LLMs in predicting flares in inflammatory arthritis.”

What Kind of Context Helps Most?

There are 3 broad types of context. And using them wisely can dramatically improve AI responses:

1️⃣ System Prompting:

This sets the overall purpose and scope of the task.

Examples:

• “Your task is to summarize clinical trial results for an IRB application.”
• “You are helping prepare a plain-language summary for a patient-facing website.”

It tells ChatGPT what kind of work it’s doing and frames how it should respond.

2️⃣ Contextual Prompting:

This gives specific, task-relevant information.

Examples:

• “This paper builds on our K23 grant and focuses on clustering PsA phenotypes using EHR data.”
• “The intended audience is a group of internal medicine residents with limited rheumatology exposure.”

This helps ChatGPT understand the immediate situation, so its answers are accurate, coherent, and on-point.

3️⃣ Role Prompting:

This assigns a persona or tone for the AI to adopt.

Examples:

• “You are a biostatistician reviewing a manuscript for a clinical journal.”
• “Act like a patient educator explaining the randomized trial at the 6th-grade level.”

Role prompting tunes the voice, depth, and style of the response.

Combine All Three: A Simple Prompt Upgrade Formula

Here’s how you can combine these three types of context into one powerful prompt:

“I’m a physician-scientist preparing an NIH grant resubmission on inflammatory arthritis phenotypes. These are the preliminary aims: [paste aims]. Please suggest a stronger rationale for Aim 2. Act like a scientific review panel of rheumatologists and data scientists and write a response geared towards them.”

This short prompt includes:

• System-level framing: “preparing an NIH grant resubmission”
• Contextual info: “inflammatory arthritis phenotypes,” “these are the aims”
• Role definition: “scientific review panel of rheumatologists and data scientists”

That’s all it takes to dramatically boost the quality of ChatGPT’s output.

How to Format Your Prompt So It’s Easy to Understand

Here are four practical structures to separate background from task:

✅ Triple Dash Format

I’m drafting a plain-language summary for a patient advisory board. Here’s the background:
---
This study explores the use of AI models in detecting early psoriatic arthritis flares using EHR data.
...
Please summarize this in 100 words at an 8th-grade reading level.

✅ XML Tags

<background>
This is for a lay summary in a patient engagement report. We studied how SGLT2 inhibitors affect non-diabetic heart failure outcomes.
</background>
<task>
Write a 150-word plain-language summary of our findings.
</task>

✅ Markdown Headers

## Context
This is a resubmission of an R01 grant. We are testing treatment prediction models in inflammatory arthritis.

## Task
Provide me stronger arguments for the “Significance” section emphasizing why current models underperform in heterogeneous diseases like psoriatic arthritis.

✅ Mix and Match

## Background
This is an abstract for a machine learning paper on predicting cardiovascular events using UK Biobank data.

<audience>
Cardiology researchers with minimal ML experience.
</audience>

## Task
Rewrite the abstract to focus on clinical relevance, not modeling technique.

Better Context In = Better Output Out

If your prompt wouldn’t make sense to a human collaborator, it won’t make sense to ChatGPT either.

When you give it the right context—what you’re working on, who you are, who it’s for, and what you need—it stops acting like a guessing machine and starts becoming the collaborator you hoped for.

P.S. If You’re Tired of Figuring Out Prompts From Scratch…

This is exactly the kind of friction I’ve worked to remove inside Research Boost—a platform I’m building for clinical researchers like you for almost a year now.

It bakes in all these prompt strategies under the hood. So instead of wrestling with formatting or instructions, you just provide the key information for any section of a manuscript (abstract, intro, methods, results, or discussion) and get high-quality drafts grounded in your actual research and your own perspective.

No fancy prompting. No AI guesswork.

If you found this post helpful, you’ll love what we’re building.

👉 Join the waitlist for Research Boost here (early access with free trial, beta pricing on Pro).