Dear Negotiation Explorer,
Last week I wrote about the four-agent workflow we built live in Lightning Lesson 2 — Kraljic and Kahneman feeding into Deepak, the Compiler stitching everything into a one-page negotiation strategy, then Von Neumann stress-testing it against a counterpart who wouldn't move where Morrison expected. The chain works because each agent is grounded. It reasons from specific knowledge, not from whatever the model picked up during training.
That grounding is the layer most negotiators leave out. And it's the biggest miss in how AI gets used for negotiation preparation today.
Most people don't build a knowledge base because they assume the AI already has the expertise. It has read every negotiation book, every Harvard case, every blog post, every article. What more could you possibly add?
That assumption is half right. And dangerously wrong.
What training gives you — and what it doesn't
Training is compression. Trillions of words averaged into general patterns. Every Malhotra chapter sits in the same averaged blur as every LinkedIn post, every Reddit thread, every recycled sales script the AI ever read about negotiation. The result is wide — the AI can speak about negotiation fluently — but thin. It has read about negotiation. It hasn't learned it.
Worse: the AI can't tell what it actually knows from what it's confidently guessing. Hallucinations don't sound like errors. They sound like answers — produced by the same generation mechanism. When the AI tells you "in a deal like this, the typical reservation value is X," you can't tell if that's drawn from a verified principle or extrapolated from thin air. Neither can the model.
What a knowledge base changes
A knowledge base doesn't make the AI smarter. It tells the AI where to look — not across the entire internet, but in specific material you've chosen. That material can be established negotiation expertise the AI draws on with precision, your own thinking and past deals, or both. Three things shift when the AI is working from material like that:
Performance. Answers come from specific, validated material — focused negotiation expertise and your own thinking — instead of the internet average.
Reliability. The same prompt produces the same quality across sessions. The AI isn't pulling from a different corner of its memory each time you open a new conversation.
Hallucinations drop. When the AI is working from your knowledge base, it quotes your principle or admits the knowledge base doesn't cover something. It doesn't have to fabricate to fill a gap.
The experiment running right now
I'm running this exact comparison with a group of practitioners this week. Their first assignment is the same prompt against the same case, run twice — once with no knowledge base loaded, once with the one they build for themselves.
The second analysis is significantly better.
The reason most negotiators never see that gap is they never run the experiment. They load a prompt, accept whatever comes back, and never know how much of the answer was theirs and how much was the internet average.
You don't write a knowledge base. You extract one.
Most people picture building a knowledge base as a blank-page exercise — days of writing, structuring, deciding what belongs. They defer it indefinitely.
Skip that. You don't write a knowledge base. You extract one — from material you already have. Some of it is yours: deal notes, training slides, email threads where you explained your strategy, LinkedIn posts where you laid out a principle. Most of it probably isn't: articles you've saved, excerpts from scholars you keep coming back to, notes from courses you've taken, chapters from books that shaped how you think. Both kinds count. The AI doesn't care who wrote the source. It cares that you've chosen it.
Two or three documents is enough to start.
The extraction prompt has two paths:
Path 1: you upload what you've collected — your own work, material from scholars you trust, or both. The AI reads it, pulls out the principles and patterns, and organizes them.
Path 2: nothing collected yet. The AI drafts a first version grounded in established negotiation expertise. You personalize it through conversation, replacing the generic frames with your own thinking and the sources you actually trust.
Either way, the AI organizes the knowledge into eight areas:
Philosophy — core beliefs about negotiation, ethical boundaries
Preparation & Planning — how to research and approach a deal
Interests & Information — how to uncover what's really at stake, yours and theirs
Value Creation — how to expand the pie, find trades, build mutual gains
Value Claiming — how to handle price, anchoring, concessions, targets
Communication & Relationships — rapport, trust, difficult conversations, listening
Difficult Situations — deadlocks, hardball, emotional moments, power imbalances
Industry-Specific Knowledge — patterns and norms in your field
Why the structure matters as much as the content
That's the content. The structure matters just as much.
The AI doesn't read your knowledge base front to back. It retrieves what each question needs — and how the knowledge base is built determines whether that retrieval actually works.
Chunks. Each concept lives in its own 150–300 word section. Too big — an essay covering five concepts at once — and the AI pulls in irrelevant material alongside the relevant part, blurring the output. Too small — a single sentence — and there isn't enough context to be useful. The sweet spot lets the AI grab one concept, fully explained, and reason from it.
Cross-references. Each chunk links to related ones. BATNA points to reservation value. Value creation points to interests. Difficult situations points back to philosophy. The AI follows the threads instead of treating each chunk as an island — and that's how you get answers that connect the dots across your knowledge.
Map. A top-level index of how the knowledge base is built — the eight areas, what's inside each, how they relate. The AI consults the map to orient — "where do I look for X?" — before diving into specifics.
Chunks, cross-references, map. That's the difference between a long document and an actual knowledge base.
What you can do this weekend
Spend an hour. Pull together two or three documents you already have — proposals, deal summaries, training notes, anything that reflects how you negotiate — and run the extraction prompt. You'll end the weekend with a first-pass knowledge base. Imperfect, but real.
From that point forward, every conversation you have with your AI assistant draws on your expertise instead of the internet average. And every output you build on top of it — including the Augmented Negotiator's Brief, the one-page strategic preparation tool you can drop into ChatGPT, Claude, or Copilot — gets sharper.
The brief is the artifact. The knowledge base is what fills it — with the expertise you trust and the experience you bring.
If you haven't grabbed the Augmented Negotiator's Brief yet, it's at https://negoai.ai/subscribe.
What This Means for You
The AI doesn't need to read more. It needs to read less — but with focus. Specific material. Things that are deep where its training is wide, validated where its training is averaged, and yours where its training is everyone's.
Until you give it that, every answer you get is the internet average dressed up in confident prose. Plausible. Sometimes useful. But never quite what you need.
Build the knowledge base. Every output you put on top of it gets better. And you stop wondering whether the AI is being useful or just being plausible.
This Week's Question
Think back to the last time you used AI to prepare for a negotiation. Where did the answer fall a half-step short — and would your own knowledge base have caught it?
Reply and tell me. I read every response.