The following steps for undertaking efficient and effective ‘deep research’ with AI are the product (an ‘Operator’s Handbook’) which can be found at the end of a really helpful and well researched article by Torsten Walbaum:
1. Clearly state your goal
The first step in conducting effective AI-driven research is to clearly define your goal. Instead of posing a vague query, you should articulate a precise objective that specifies the desired output. This involves transforming a general interest into a concrete task by stating exactly what you need, such as a summary of best practices, a checklist of actionable steps, a comparison of tools, or a detailed guide. By being explicit about the deliverables, you direct the AI to generate a focused and purposeful response that directly addresses your requirements, rather than a broad and less useful overview.
2. Provide context
The second, and most critical, step is to provide comprehensive context to personalize the research. Without specific details about your situation, the AI will produce a generic report based on broad assumptions. To create a customized and actionable output, you should include key information such as facts about your business (product, model, geography), the intended audience and their familiarity with the topic, any operational constraints, and the ultimate use of the report (e.g., to inform a specific decision). Providing this background transforms the AI into a virtual research analyst, tailoring the findings to your unique circumstances for a much more relevant and impactful result.
3. Specify what you want the output to look like
Third, you must specify the desired structure and format of the final report to ensure it is easy to digest. The default output from AI models can often be dense and poorly organized, so clear instructions are necessary. This includes dictating the content structure, such as asking the AI to follow the “Pyramid Principle” by leading with key takeaways, including specific elements like comparisons or templates, and providing an overview of the sources used. Furthermore, you should specify formatting preferences, such as using bullet points, bolded text, and presenting comparative data in tables rather than lengthy prose, which significantly enhances the report’s clarity and readability.
4. Ask for the Research Plan and Provide Feedback
Finally, to prevent unwanted surprises and ensure the research is on the right track, you should ask the AI for its research plan and provide feedback before it begins. This allows you to review the proposed approach for comprehensiveness, focus, and methodology, and to correct any invalid assumptions the AI may have made. This stage is also the ideal opportunity to guide the AI’s source selection, such as requesting it to prioritize recent data from independent third parties over company websites. By reviewing and refining the plan, you can align the AI’s process with your expectations, ensuring the final output is built on a solid and relevant foundation.
The whole article is long, but really worth reading – the above 4 steps are just a paraphrase of one section of it, and do not do justice to the insights he has provided. . If you are serious about using AI for deep research this article could save you hours of effort.