Write Better AI Prompts for Deep Research

I created an 8-part checklist that consistently gets better deep research from AI tools.

Getting Better Results from AI Deep Research: the “RGOSPACE” Framework

Sometimes we want AI to give us quick answers. Other times, we need to examine a topic in depth, and the Deep Research functionality found in many common AI tools (ChatGPT, Gemini, etc) is fantastic.

For these deep research tasks, how we ask the AI to do the work matters a lot, and when you need comprehensive research it pays to spend a bit more time on your prompt.

The prompt is the text of the request you give the AI to ask it to do something. I teach a very basic prompting framework called GOAL for many tasks, and it helps with straightforward requests.

But for complex research, I found the need for more structure. Building on GOAL, I developed a checklist or framework called “RGOSPACE” to help me think through the research request. It has some of the elements of GOAL but adds more detail for these bigger, more important research tasks.

Clear instructions lead to clear, useful output from AI.

Or as we’ve said in computing for decades “Garbage In, Garbage Out”.

Why a Structured Prompt Matters for Deep Research

Deep research tools are powerful because they combine reasoning language models with search engines to give you comprehensive analysis – often searching hundreds of websites and reading many research papers. In fact because of this depth of research, many of these tools limit how often you can use them.

When you have limited uses, every prompt needs to count. A weak prompt gives you poor results and wastes a valuable research session.

When you do deep research, you are asking the AI to act like a research assistant. A good assistant needs a good brief, but vague requests give vague results. This framework helps you create a clear brief, which means the AI can give you better, more focused information. You get what you need, faster.

What is “RGOSPACE”?

RGOSPACE helps you remember eight parts of a strong research prompt. I suggest you start with the specific research Role first, then your Goal, then your Output.

The other parts – Scope, Purpose, Audience, Context, and Essentials – add further clarity to make your request bulletproof.

Let’s look at each part.

1. R – Role:

• What it is: Tell the AI what ‘hat’ to wear. What job or expertise should it have?
• Why it matters: This sets the viewpoint for the AI. The AI will use knowledge related to that role, and its answer will reflect that role’s way of thinking.
• Examples:
  • “You are a historian specialising in medieval Ireland.”
  • “Act as a financial analyst looking at company performance.”
  • “You are a science teacher explaining a concept to students.”

2. G – Goal:

• What it is: Say what you want to achieve. What is the main question you want answered?
• Why it matters: This is the main task for the AI. A clear goal keeps the AI focused on what actually matters.
• Examples:
  • “My goal is to understand the causes of the 1798 rebellion in Ireland.”
  • “I want to compare the features of three different software products.”
  • “The objective is to identify future trends in renewable energy.”

3. O – Output:

• What it is: Describe what you want the AI to give you. Think about format, style, structure, and length. For deep research, you often want a structured document – a report, detailed summary, literature review, or analysis. You might also want a comparison table, structured findings, or presentation script.
• Why it matters: This tells the AI how to present the information so you get it in a form you can use directly.
• Examples:
  • “Write a 1500-word report with clear sections: Introduction, Findings, Conclusion. Include a bibliography.”
  • “Create a comparison table with columns for key features, strengths, and weaknesses. Follow with a 300-word summary.”
  • “Produce an 800-word literature review summary in academic but clear style. Highlight key themes and research gaps.”
• Pro tip: I like adding this to most outputs: “At the top, include an Executive Summary that explains the key findings and insights, written for a non-technical audience.” This works even for technical topics because it gives you a clear overview you can share with anyone.

4. S – Scope:

• What it is: Set the limits for the research. What should be included? What should be left out? Are there time limits or specific places to focus on?
• Why it matters: This stops the AI from giving too much, or irrelevant, information. It keeps the research tight and focused on what you actually need.
• Examples:
  • “Focus only on the years 1900 to 1950.”
  • “Examine the impact in Europe, not globally.”
  • “Do not include information about political factors; only economic ones.”

5. P – Purpose:

• What it is: Explain why you need this research. How will you use the information? What decisions will it help you make?
• Why it matters: When the AI knows the purpose, it can make the information more relevant to your specific needs.
• Examples:
  • “I will use this information for a presentation to my team.”
  • “This research will help me decide which supplier to choose.”
  • “The findings will be part of a chapter in my book.”

6. A – Audience:

• What it is: Who will read or use this research? Are they experts? Are they new to the topic?
• Why it matters: This helps the AI choose the right words, level of detail, and complexity for your specific readers.
• Examples:
  • “The audience is a group of university students.”
  • “This is for a manager who needs a quick overview.”
  • “Write for people who have no prior knowledge of this subject.”

The next two, Context and Essentials, make the difference between good and excellent results.

7. C – Context:

• What it is: Give the AI background information. What does it already need to know to understand your request? Include data or examples if you have them.
• Why it matters: This gives the AI a starting point and saves it from making wrong assumptions about your situation.
• Examples:
  • “We have already tried solutions X and Y, and they did not work.”
  • “The current market share is 20%, up from 15% last year.”
  • “See the attached document for last year’s report on this topic.”

8. E – Essentials:

• What it is: List anything that must be in the answer. Are there specific points to cover, questions to answer, or data to include?
• Why it matters: This makes sure you get all the key information you need. It acts as a checklist for the AI so nothing important gets missed.
• Examples:
  • “You must include a section on ethical considerations.”
  • “Compare the costs of option A versus option B.”
  • “Ensure you mention the work of Dr. Jane Smith on this topic.”

How to Use This

RGOSPACE is a guide for your thinking that helps you prepare before you write your prompt. The order of RGOSPACE is a good order for building your prompt, and I recommend starting with the Role at the beginning of your written prompt.

Here’s how to use it: Go through each letter of RGOSPACE and write down your points for each letter. Then write your full prompt to the AI, using your points. You can write it as a conversation or a set of instructions – whatever feels natural.

Here’s a quick template to copy-paste:

“You are a [ROLE]. My goal is to [GOAL]. I need [OUTPUT]. Focus on [SCOPE]. This will help me [PURPOSE]. Write for [AUDIENCE]. Background: [CONTEXT]. You must include: [ESSENTIALS].”

Don’t just list facts for each RGOSPACE point; explain what you mean and be clear about your expectations.

Your first prompt using RGOSPACE might not be perfect, and that is normal. Examine the AI’s answer and see what you can improve in your prompt for next time. This is how we learn to work better with AI.

Full Examples

Here are three examples showing how to use RGOSPACE.

Example 1: The Future of Peatlands in Ireland

“Act as an environmental scientist specialising in peatland ecosystems and conservation policy. My goal is to understand the main challenges and opportunities for Irish peatlands over the next 20 years (2025-2045).

I need a comprehensive 1000-word report structured as follows: Executive Summary (written for a non-technical audience), Introduction, Challenges section, Opportunities section, and Conclusion with recommendations. Use a neutral, informative style appropriate for a community group with strong environmental interest but varied levels of scientific knowledge.

Focus specifically on peatlands within the Republic of Ireland. Examine environmental, economic, and social aspects. Do not cover peat extraction for horticulture in detail – focus more on conservation and alternative uses.

This research will inform a local community group’s response to government consultations on land use policy. Irish peatlands have a long history of traditional use, particularly turf cutting for fuel. However, current EU and national policies are shifting towards conservation, rewetting, and carbon sequestration. There’s ongoing debate about balancing traditional rights with environmental imperatives.

The report must include: 1) The role of peatlands in carbon storage and climate change mitigation, 2) At least two major challenges facing Irish peatlands (such as drainage issues or impacts of past extraction), 3) At least three viable opportunities for sustainable peatland management (such as paludiculture, eco-tourism, or renewable energy development on degraded sites), and 4) Discussion of how these changes impact local communities, both positively and negatively.

Please ensure the Executive Summary can be understood by someone without scientific training, as this may be shared widely within the community.”

Example 2: Comparing AI Chatbot Technologies for Small Businesses

“You are a technology consultant with expertise in AI-powered customer service solutions for small to medium-sized enterprises. I want to compare the top three current AI chatbot technologies that would be suitable for my small e-commerce business.

I need you to create a detailed comparison table highlighting key features, pricing models, ease of integration, and scalability options. Follow this with a 300-word summary recommending one specific option with clear reasons why. Start with an Executive Summary that a non-technical person can understand.

Focus only on AI chatbots available in the Ireland market as of early 2025. My business has around 500 monthly customer interactions, and my budget is under €100 per month. I need features relevant to e-commerce: order tracking, FAQ handling, product recommendations, and basic customer support.

I will use this comparison to choose and implement an AI chatbot for my online shop to improve customer response times. Currently, we handle all customer queries via email, which is becoming too slow as we grow.

I run a business selling handmade crafts and use Shopify as our e-commerce platform. I have some technical understanding but am not an AI expert – I need solutions I can implement without hiring developers.

Your comparison must include: 1) At least one open-source option if it’s viable for my needs and budget, 2) Specific details about integration capabilities with Shopify, including any plugins or APIs required, and 3) Clear assessment of how easy each chatbot is for a non-developer to customise and maintain.

Please be practical and honest about which option would work best for a small business owner who needs to implement this themselves.”

Example 3: (Real example) Understand the concept of AI orchestration, agents, and LLMs

“You are an AI research engineer with hands-on experience in LLM agent orchestration and toolchains like Roocode and VSCode.

My goal is to develop a deep understanding of how orchestration, agents, and LLMs fit together — both conceptually and in terms of practical experimentation.

I need a comprehensive 1200–1500 word document structured as follows:
– An Executive Summary written for a technically literate reader new to this domain
– A high-level conceptual overview explaining:
1. What LLMs, agents, and orchestration mean
2. How these elements work together
– A landscape section comparing at least three agent orchestration tools (e.g., LangChain, CrewAI, Roocode), highlighting key capabilities and trade-offs
– A practical hands-on guide to using Roocode within VSCode to build, modify, and experiment with agent workflows. Focus on showing what’s possible, not basic installation.
– A final section with recommendations, pitfalls to avoid, and next steps for further learning.

Focus specifically on using Roocode+VSCode for experimentation. Do not include installation instructions or LLM fundamentals.

This research will help me become confident in using agent workflows in my own AI experiments and projects.

Write for a developer or analyst who understands code and AI, but is new to the concept of orchestration and multi-agent systems.

Background: I already have Roocode and VSCode installed and working. I want to move beyond the basics and start building useful workflows. I’m familiar with concepts like prompts and LLM fine-tuning, but I’ve never used orchestration or agent-based tools.

You must include:
– A clear analogy or mental model to help understand how orchestration and agents interact
– A comparison of at least three tools, including Roocode
– A walkthrough of how to create and test a basic agent workflow using Roocode
– Tips on common beginner mistakes or misunderstandings”

Here’s the output of that research so you can see what it produced.

Structured Prompts Will Make Your AI Research Better

RGOSPACE is a tool, and like any tool, it takes practice to use well. Start with a research topic you know and try to build a prompt using these eight parts. See what difference it makes to the AI’s answer. You will find it encourages clearer requests from you and better, more useful research from the AI.

The framework works because it makes you think through what you actually need before you ask for it. Most people ask vague questions and get vague answers back. RGOSPACE stops that happening by turning you into a better research partner for AI.