The brainstorm-to-research-prompt pipeline

# The brainstorm-to-research-prompt pipeline Here's what most people do before running a deep research: they type a vague question into an AI and hit enter. The output is vague in proportion to the input. This workshop gives you a 5-step pipeline that takes you from "I have some fuzzy thinking about a topic" to a structured research prompt that produces sharp, usable output. You'll run all five steps right now and submit the final prompt. ## What you need - A topic you actually want to research. Pick something real — a market you're exploring, a technology you're evaluating, a question your business needs answered. - A voice input tool (Whisper Flow from the previous workshop, or just type if you haven't set that up yet) - An AI chat open — [claude.ai](https://claude.ai) or [chatgpt.com](https://chatgpt.com) ## Step 1: Voice-brainstorm with no structure Open a new chat. Don't think too hard. Hold your hotkey and talk for 60–90 seconds about the topic: - What you already know - What confuses you - What you think the answer might be - What you'd do with the information if you had it Don't organize it. Don't edit it. Get the messy, unfiltered version out. It'll look like a stream of consciousness — that's the point. If you're typing instead of speaking, set a 90-second timer and write without editing. Resist the urge to clean it up. ## Step 2: Ask AI to extract themes and questions After your dump, send this message exactly: ``` Read that brainstorm. Extract: 1. The core themes (3–5 bullets) 2. The most important unanswered questions (3–5 bullets) 3. Any assumptions I seem to be making that might need checking ``` Read the output. It'll surface things you were circling without naming directly. Sometimes the AI's framing of your own thinking is more useful than the thinking itself. ## Step 3: Ask AI to draft a research prompt scaffold Now send this: ``` Based on those themes and questions, draft a Deep Research prompt scaffold. The prompt should: - Focus on the most important question - Be written for a research AI that will cite sources - Include a section for specifying the output format - Leave placeholders where I need to add date ranges and specific constraints ``` What you'll get back is a skeleton — a structured prompt with gaps you still need to fill. Don't use it as-is. That's the next step. ## Step 4: Edit the scaffold This is the step most people skip and why their research output is mediocre. Open the scaffold and add four things: **Date anchor.** Research without a time constraint produces a mix of current and outdated information. Add something like: "Focus on sources from 2023 to present" or "Current as of Q1 2026." Pick a window that matches your question. **Citation requirements.** Tell the AI what you expect. "Cite at least 5 primary sources" or "Include a works cited section with links" or "Prioritize peer-reviewed sources over blog posts." Without this, you get confident assertions with no way to verify them. **Output format.** Specify what you want back. "A structured report with an executive summary, key findings, and a limitations section" produces a different — and usually better — output than an open-ended request. Match the format to how you'll actually use the research. **Scope constraints.** If the question is too broad, narrow it. Geographic scope, industry scope, company-size scope — any constraint that makes the answer more actionable for your specific situation. ## Step 5: Save the final prompt Copy the edited prompt somewhere you'll find it. A Notion page, a notes doc, a dedicated research prompt file — whatever you'll actually use. Name it after the question, not the topic. Bad: "AI in healthcare research." Better: "What are the most defensible moats for vertical AI SaaS in the US healthcare market as of 2025–2026?" You now have a deployable research prompt. The next three workshops cover where to run it. ## Your submission Paste your final research prompt below. Not the brainstorm, not the scaffold — the edited, ready-to-deploy version with all four elements present. --- Now you can apply this to: - Any market research question before committing budget to a new direction - Technical evaluation when you're choosing between tools or approaches - Competitive landscape work before a pitch or product decision