We all do research for our work in one form or another. Staying current, preparing for a meeting, understanding a market, evaluating a decision. Deep research tools have genuinely compressed that work. What used to take half a day can now take an hour.
But faster research is not the same as better decisions. The problem with AI research is not the speed. It is the confidence.
I have been playing with a structure to close that gap. The idea is simple: question the assumptions. Ask what AI cannot tell you. Then ask what you should do differently.
Three stages. Three prompts. Here is how it works.
🎓 Coming up: AI Fluency for Senior Leaders
Developing AI fluency is not optional anymore. The best way to learn it is to be hands on with others who are on the same path. My goal is to share my own systems and workflows so it’s all practical and usable right away. AI transformation begins at fluency because without it we would not know what works and what does not. Don’t wait too long to build yours! Join me!
🎯 The Three-stage System
Stage 1: Gather
Before asking AI to find anything, give it context: your question, what you already know, and the specific decision you are trying to make. Use the AI prompt below to run this stage.
If you are using a deep research tool, skip that prompt and use the deep research prompt instead, give the tool your question and context, let it run, and bring the report into Stage 2. The deep research tool does Stage 1 for you.
AI prompt:
> Here is my research question: [paste question]. Here is the context I already have: [paste links, notes, or background]. Please tell me: (1) what you can answer confidently from these sources, (2) what you can address only in general terms, and (3) where I need a primary source or human expert for a reliable answer.
Deep research tool prompt:
> Research question: [paste question]. Context I have: [paste any links, notes, or background]. I am trying to decide: [paste the specific decision]. Please research this thoroughly, cite your sources, and flag where evidence is limited or conflicting.
Stage 2: Synthesize
Ask explicitly for synthesis. A summary is not the same as synthesis. This prompt forces AI to tell you what to believe, what it cannot tell you, and what assumption you need to test, the three things a summary almost never surfaces. If you ran a deep research tool in Stage 1, paste its output here and run this prompt on it.
> Based on the material gathered above, please synthesize into: (1) the three things I should believe based on this research, (2) the two things this research cannot tell me, and (3) one assumption embedded in this research that I should test before acting on it.
Stage 3: Extract decisions
This is the stage that is easy to skip when you are running short on time. However, a beautiful research summary does not change behavior. A direct question does: given everything above, what should I do differently? Ask it every time.
> Based on the synthesis above, what are three concrete things I should do differently in [this project / this client engagement / this decision]? If the research does not support a concrete recommendation for any of these, say that rather than inventing one.
🔁 Make it Repeatable
You don't need to find this issue every time you want to run the workflow. There are three ways to make it permanent, regardless of which AI platform you use:
Make it a skill - save the block below as a named skill you can call from any conversation.
Create a project and add as custom instructions - every conversation in that project runs the workflow from the first message.
Create a custom agent, gem, or GPT - depending on the platform you are in, build a dedicated agent with the block below and share it with your team.
Copy this and paste it wherever you want the workflow to live:
> You are my research assistant. When I give you a research question, run three stages in sequence.
> Stage 1 — Gather. Start by asking me: do I want to run this through a deep research tool, or work through it here? If deep research: give me this prompt to copy and run in my tool of choice — "Research question: [paste question]. Context I have: [paste any links, notes, or background]. I am trying to decide: [paste the specific decision]. Please research this thoroughly, cite your sources, and flag where evidence is limited or conflicting." — then ask me to come back with the report and we will start at Stage 2. If working here: ask me what context I already have — links, notes, prior meetings, the specific decision I am trying to make — then tell me (1) what you can answer confidently from those sources, (2) what you can address only in general terms, and (3) where I need a primary source or human expert.
> Stage 2 — Synthesize. After gathering, synthesize into: (1) the three things I should believe based on this research, (2) the two things this research cannot tell me, and (3) one assumption embedded in this research that I should test before acting.
> Stage 3 — Extract decisions. After synthesizing, give me three concrete things I should do differently in my specific context. If the research does not support a concrete recommendation, say that rather than inventing one.
> Always start by asking me whether I want to use a deep research tool or work through it here.
💬 Your Feedback Matters
A big THANKS to those who filled out the survey!!
If you have not responded to the Clarity Prompts survey yet, I would love your input. Two minutes, and it shapes what comes next.
Found this useful? Forward it to one person who needs to hear it. Getting this for the first time? Subscribe free to receive this weekly newsletter here.
↗︎ Think Better. Execute Faster.
See you next week — Purti
Clarity Prompts The weekly AI operating system for sharper leaders.
