Deep Research for problem fluency – Introducing COSID prompting

In our experience, the most impactful product initiatives are those where we invest meaningful time understanding the problem space before rushing to solutions. This time allows us to build what I call ‘problem fluency’ – a deep understanding that enables teams to navigate complexity. It is essential in creating solutions that genuinely address user needs and deliver business value.

One way to develop problem fluency is through research into complex domains to uncover root causes, user needs, and systemic patterns, leading to a clearer understanding of the problems and opportunities at hand.

Recently, several AI tools have introduced features to support this kind of in-depth exploration, commonly referred to as Deep Research. OpenAI describes Deep Research as ‘multi-step internet research for complex tasks’ and suggests that an agent can do this work for you independently. All you need to do is provide a prompt, and ChatGPT will ‘find, analyse and synthesise hundreds of online sources to create a comprehensive report at the level of a research analyst’.

Isn’t that mind-blowing? Imagine having your own personal research analyst that can process vast amounts of information, identify patterns, acknowledge limitations and synthesise findings! All while properly citing sources and highlighting potential biases. The implications for product teams are transformative, saving time without massive resource investments.

COSID prompting structure for Deep Research

I’ve been experimenting with different prompting techniques, mixing conversational style prompting with the mindset that I’m working with a research analyst who needs a meticulously laid out brief to execute great research.

I started experimenting with ways to structure my prompts more effectively, from the perspective that I’m collaborating with a research partner whose output quality will directly reflect the clarity and rigour of my brief.

This led me to develop a prompting structure I call COSID. The COSID prompting structure encourages users to invest a bit more effort upfront to create comprehensive, well-scoped prompts that consistently yield remarkable results.

The COSID prompting structure organises deep research through five interconnected dimensions:

• Context – Establishes the environment and situational factors influencing the research space
• Objective – Articulates specific goals and outcomes that the research aims to achieve
• Scope – Defines boundaries and parameters for focused investigation
• Interest Areas – Highlights key domains for exploration and hypothesis testing
• Deliverables – Specifies expected outputs and evidence standards

Let me illustrate this using a recent product exploration scenario I’ve been exploring with my AI learning group triad:

Transformative insights through Deep Research

This structured approach completely reframed our understanding of the opportunity space:

• Market gap validation: We confirmed a significant underserved segment, but not exactly the one we initially anticipated! The stronger opportunity existed among women and in tier-2 cities.
• Need refinement: The core problem wasn’t simply “lack of time” but rather “fragmented time” – professionals had small pockets of availability that existing solutions weren’t designed to accommodate.
• Competitive positioning: Deep research identified that localisation went far beyond language – cultural context around fitness goals and social dynamics was inadequately addressed in existing solutions.
• Business model insights: Subscription patterns suggested an opportunity for workplace wellness partnerships rather than purely direct-to-consumer approaches.

One often-overlooked prompt enhancement is defining the structure, writing style, and target audience of the expected deliverables. For example:

• “Make it like an executive report that includes a ‘So What?’ box with exactly three bullet takeaways for executives.”
• “The report should be readable by someone with extensive knowledge of product management.”

Integrating COSID into daily practice

I’ve begun integrating the COSID structure into my regular interactions with LLMs beyond just research projects. The results have been consistently impressive – providing more focused, relevant, and actionable outputs across various contexts.

I’m really excited to see how the principles behind COSID resonate with broader industry thinking on structured prompting approaches. The Prompting Guidance published by Defra in its AI SDLC playbook beautifully illustrates similar patterns and principles. As practitioners across industries experiment with structured prompting approaches, we’re collectively advancing our understanding through shared documentation and real-world application.

Agile India 2025 testimonial

At the Agile India 2025 Conference in March, I demonstrated the transformative power of Deep Research in action. A speaker visiting from the US told me how he had conducted market research manually over several weeks. During our workshop session, we spent about an hour crafting the right COSID prompt, then it took ChatGPT approximately 45 minutes and Perplexity approximately 25 minutes to process the research request.

The results were astonishing! The deep research not only confirmed some of his findings but uncovered additional insights that were missed and which would have cost thousands of dollars to obtain through a commercial research analyst. The additional insights would really support the sales proposal for which he had done the research, but now he has powerful data-driven insights to back it up. All this happened in real time during our collaborative session!

Understanding the power and boundaries of AI research

Deep Research is accelerating traditional desk research to unprecedented speeds and efficiency, yet comes with natural boundaries to consider. Currently, it primarily accesses publicly available information, potentially missing valuable insights behind paywalls, though future iterations may address this limitation through premium features. Since competitors can access identical research capabilities, our competitive advantage ultimately depends on how we uniquely interpret findings and transform them into action.

Let’s connect

Have you been exploring Deep Research in your product practice? Or perhaps you’re curious about how to apply it within your team? At Equal Experts, we’re actively running experiments like these, exploring how AI-powered techniques such as Deep Research can supercharge product thinking, accelerate discovery, and build stronger problem fluency. If this resonates with you, we would love to connect!