Original Insight
“This sentence alone was the one thing that made the biggest difference: execute multiple search passes… I’m having it go through 5 different passes. That’s why it’s finding as much information… Because I’m saying, don’t just look for the generic searches — also look for the specific terms, the related issues, the edge cases.” — Lou
Expanded Synthesis
A single search returns what is obvious. Five searches return what is true.
Lou’s August 7 session revealed the architectural secret behind the dramatic performance jump in his legal AI system. The improvement wasn’t just a better model — it was a better retrieval strategy. Specifically: instead of performing one semantic search against the database, the system now performs multiple coordinated passes, each approaching the same question from a different angle.
The passes included:
- Exact term search — what the user literally asked for
- Semantic variation — synonyms and related meanings
- Opposing perspectives — what the other side might argue
- Temporal variations — how positions changed over time
- Edge cases and related issues — what the user might not know they need
The result: a jump from 16-18 retrieved document chunks to 29-31, with dramatically higher relevance scores (from 27% to 80%+ in some cases). But more importantly, the answers shifted from generic summaries to specific, actionable, context-aware analysis.
The principle here transcends AI systems. The most valuable thinkers — the consultants, coaches, and advisors high-performers pay for — don’t answer the question you asked. They interpret the question from multiple angles, anticipate what you haven’t yet thought to ask, and retrieve insight from the full scope of their experience before synthesizing a response. This is multi-pass retrieval in human form.
The blind spot for high-performers: most people search their own knowledge the same way a naive RAG system does. They ask themselves one version of the question, pull the first thing that surfaces, and treat it as the answer. The result is confident reasoning built on a partial evidence base. The person who asks “what’s my best move here?” is doing one-pass retrieval. The person who asks “what’s my best move, what would the opposition do, what am I not seeing, what will matter in six months, and what’s the edge case that could break this?” is doing multi-pass retrieval.
For coaches, this is a direct framework for improving client decision quality. When a client presents a problem or decision, the coaching conversation itself is a retrieval mechanism. The quality of the questions you ask determines the quality of the material surfaced for reasoning. Weak coaching questions retrieve obvious content. Powerful coaching questions execute multiple passes — temporal, relational, oppositional, emotional, strategic — and bring genuinely useful raw material into the client’s awareness.
Lou explicitly noted he didn’t have to write this logic himself — he told the AI what he was trying to achieve, gave it context about the user and the database, and the AI generated the multi-pass strategy. This is a meta-level insight: you can instruct any intelligence (AI or human assistant) to search more comprehensively by explaining the why behind the question, not just the what.
The “clarification protocol” Lou built in — where the system asks for clarification when ambiguity would cause it to miss critical information — is worth highlighting for coaches. The instinct to answer questions immediately is often the enemy of depth. Building in a pause, a clarification, a “what are you really trying to accomplish?” is not slowing the process. It is the process.
Practical Application for PowerUp Clients
The Five-Pass Decision Framework
When facing any significant decision, run it through five passes before acting:
- The Literal Pass: What is the specific question I’m trying to answer?
- The Semantic Pass: What are three different ways to frame this same question that might surface different information?
- The Opposition Pass: What would someone who disagrees with my likely answer say? What evidence supports their view?
- The Temporal Pass: How does this look differently if I think about it from 6 months ago, right now, and 2 years from now?
- The Edge Case Pass: What’s the scenario where my current thinking completely fails? What am I assuming that might not be true?
For Coaching Sessions: Use this as a structured debrief tool with clients before major decisions. Run through each pass together. The coach’s job is to hold the client to all five passes — not to let them escape into action after pass one.
Journal Prompt: “What question am I currently answering with only one search pass? What would I find if I asked it four more ways?”
For AI Tool Users: When building a custom GPT, chatbot, or AI assistant with access to your knowledge base, add this to your system prompt: “Before retrieving information to answer any question, identify 5 different angles from which the user might need information related to their query, and search for all of them before generating a response.”
Additional Resources
- The Art of the Good Life by Rolf Dobelli — on the value of expanding your information search before deciding
- Decisive by Chip and Dan Heath — on widening your options as a core decision-making strategy
- Insight - Codify Your Judgment Into Skills, Not Just Prompts — on turning your methodology into replicable AI behavior
- Insight - Teach One Era Ahead of Your Audience, Not Eight — on the difference between surfacing existing knowledge and generating new frameworks
Evolution Across Sessions
The Aug 21 session discussed “trust before automation” and “build the business model that matches your energy.” This insight operates one layer deeper: once you’ve automated, the quality of how the automation searches determines what it finds. The Aug 14 session extends this further by asking: what happens when multi-pass retrieval still isn’t enough — and when should you switch from a retrieval architecture to a relational one (knowledge graphs)?
Next Actions
- For me (Lou): Develop a client-facing “Five-Pass Decision” worksheet based on this framework — applicable to AI prompt design AND high-stakes business decisions.
- For clients: Introduce the five-pass framing in the next decision-making session. Challenge clients to stay with all five passes before moving to action.