---
title: "How to Prep a Negotiation Using AI"
description: "Prep a negotiation with AI: an interests and tradeables grid, a rehearsal prompt that argues back, and the concession pricing a model will never do for you."
canonical_url: "https://builderscamp.com/guides/tools/ai-for-negotiation-prep"
date_published: "2026-09-18"
date_modified: "2026-09-18"
author: "Andre Albuquerque, Inês Lourenço"
publisher: "Builders Camp"
guide_class: "tools"
---

# AI for negotiation prep: rehearsing positions before the room decides them

**TL;DR:** Negotiation prep with a model is two artefacts: a grid of interests and priced tradeables, and a rehearsal where the model holds the other side's position and refuses to concede. The grid takes 20 minutes and the rehearsal takes 10, and the part neither produces is your walk-away number, which has to be decided in writing before either. Rehearsal also has a cost worth naming: fluency in your own argument can make you a worse listener.

## What part of a negotiation is actually preparable?

Not the tactics. The part you can prepare is the map: what you want, what you need underneath what you want, what they want, what they might need, what you can trade, and what you do if there is no agreement. Builders Camp's negotiation material puts this first for a reason, opening on interests, fallbacks and constraints before any technique, and most PMs walk into scope conversations having prepared only the first item on that list.

A model helps with two specific parts of it. It generates the other side's likely position faster than you would alone, which is useful and dangerous in the same breath. And it will hold that position under pressure while you practise, which is the thing you cannot do with a colleague without burning an hour of their time and some of your credibility.

## The grid, in eight rows

Fill this before you prompt anything. Twenty minutes, in a document you keep.

1. **Your ask,** with a number and a date.
2. **Your interest.** The thing the ask is a proxy for. "Six weeks of two engineers" is usually a proxy for "the renewal cycle does not blow up in Q1".
3. **Their ask,** in their words, as far as you know them.
4. **Their interest, marked as a guess.** With a confidence label and the evidence behind it.
5. **Your fallback if there is no agreement.** The glossary entry on [BATNA](https://builderscamp.com/guides/glossary/batna) covers the definition and how a fallback becomes a reservation value; this grid assumes you have one and that it is real rather than a bluff.
6. **Their fallback, guessed.** Usually the most valuable line on the page and the one most often left blank.
7. **Tradeables, priced twice.** Each item with a cost to you from 1 to 5 and a guessed value to them from 1 to 5.
8. **Opening, target, walk-away.** Three numbers, written before any rehearsal.

Row 7 is the engine. A trade only exists where something scores low for you and high for them: a two-week delay you can absorb against a date they have already promised a customer. Written as a single list of "things I could give", those pairings stay invisible. Written as two columns, they jump out.

## The prompt for the grid, and the label that fixes it

> Here are the facts of the situation and everything I know about the other side. Fill rows 3, 4, 6 and 7 of this grid. For every interest you attribute to them, label it stated, inferred, or guessed, and name the evidence. For guessed items, say what one question in the first five minutes would confirm or kill the guess.

That last clause is the one worth keeping. A model asked to infer someone's motives will produce a confident, coherent, plausible profile with no marker anywhere distinguishing the line you told it from the line it made up. Asking for a confirming question per guess converts the whole thing from a prediction into a checklist you can run live, and it costs nothing.

Expect the guesses to be roughly half right. The useful ones are usually the constraints rather than the motives: a model reasons well about what a sales leader's quarter looks like and badly about what this particular sales leader will do when cornered.

## The rehearsal prompt

Role-play is where a model earns its place, and the instructions have to be unhelpful on purpose.

> You are the VP of Engineering. I am going to ask you to move two engineers to the admin console for six weeks, which delays the billing revamp. Your position: you have already committed those people, and you think my request is the third reprioritisation this quarter. Hold that position. Do not concede unless I give you something concrete, and do not soften your language to be polite. Push back on vague answers. After ten exchanges, stop and tell me which of my answers was weakest and why.

Three things make this work. Naming the role gives the model a stable position instead of a general objection generator. Forbidding concession stops the default cooperative drift where it agrees with you by exchange four. And the debrief is the actual output: a model asked to grade your weakest answer will name it accurately, and it is usually the one where you changed the subject.

Run it twice with different temperaments. Once with the counterpart who argues on the merits, once with the one who says "this is the third time this quarter" and refuses to engage with your numbers. The second version is the harder rehearsal and the more realistic one.

## What a model will not do for you

- **Price your concessions.** It will list what you could give. It has no idea which of those costs you a week of goodwill and which costs you a quarter of roadmap, and getting that wrong is how a negotiation ends in agreement and regret.
- **Mark its own guesses.** Unless instructed, inferred motives arrive in the same confident register as the facts you supplied.
- **Set your walk-away.** Deciding that number under pressure means the other side sets it. Write it before you rehearse, and treat rehearsal as testing it rather than revising it.
- **Delete the trade you cannot authorise.** A model will happily propose giving away something that belongs to another team. Cross those out before the conversation, not during it.

## The cost of rehearsing

Rehearsal makes you fluent, and fluency has a failure mode. After three rounds against a simulated counterpart you have a template of how this person responds, and templates are sticky: you start hearing the beginning of an objection and answering the version you practised rather than the one being made. Negotiations are lost that way more often than by being underprepared, because the other side can tell when they are being answered from a script.

Two corrections, both cheap. Write down one question you will ask and genuinely not answer, so at least one minute of the conversation is spent listening rather than positioning. And after the real conversation, note which of the model's guesses about their interests were wrong, because that list is the calibration that makes the next grid better. If the same guess is wrong twice, stop asking the model about that person.

## Where this sits in the rest of the work

Negotiation prep shares most of its machinery with decision-meeting prep, and if the conversation happens in a room with several people rather than one, [AI for meeting prep](https://builderscamp.com/guides/tools/ai-for-meeting-prep) covers the objection table and the red team pass that generalises this to a group. For the setup work, keeping the grid and the rehearsal instructions as reusable prompts rather than rewriting them each time, [ChatGPT for product managers](https://builderscamp.com/guides/tools/chatgpt-for-product-managers) covers the mechanics.

Negotiation for Product Managers is the direct match: 1 week, 2 live sessions and 9 microlessons, opening on interests, fallbacks and constraints, then framing, stakeholder pressure, trades and concessions, and conflict without avoidance. Product Storytelling is the neighbouring skill, because the framing that moves a conversation from positions to outcomes is the same framing that makes a narrative land. If you are negotiating on behalf of a team rather than a project, Head of Product covers the stakeholder and portfolio layer where these conversations repeat every quarter with the same three people.

[See the Negotiation for Product Managers bootcamp](https://builderscamp.com/bootcamps/negotiation-for-product-managers?utm_source=guide&utm_medium=organic&utm_campaign=ai-for-negotiation-prep)

One thing the grid does that nothing else on this page does: it survives. Keep the filled version after the conversation with the outcome written at the bottom, and by the fourth negotiation with the same counterpart you will have a record of which trades they accepted and which they never will. That record beats any rehearsal.

## Frequently asked questions

### What can a model prepare that a checklist cannot?

An opponent. The part of preparation that is hard to do alone is hearing your own argument answered by someone who is not trying to help you, and a model instructed to hold a position and refuse to concede does that in ten minutes for a conversation you get one attempt at.

### How do you stop the model inventing the other side's motives?

Make it label confidence. Ask for each inferred interest to be marked as stated, inferred from evidence you supplied, or guessed, and to name the evidence. Models will comply, and the guesses stop looking like findings once they are labelled as guesses.

### What is the single most useful prep artefact?

The tradeables list with two prices per item: what it costs you, and what you think it is worth to them. Trades only work when something cheap for you is valuable to them, and that pairing is invisible until both numbers are written down side by side.

### Should you set your walk-away number before the conversation?

Yes, in writing, before any rehearsal. A number decided under pressure is the number the other side asked for. The glossary entry on BATNA covers how a fallback turns into a reservation value you can actually hold.

### Can rehearsal make you worse?

It can. Rehearsing against a simulated counterpart builds fluency in your own argument and a template of how the other person will respond, and you can end up answering the template instead of the person in front of you. Going in with one question you will ask and not answer is the cheapest correction.

### Does this apply to internal negotiations?

Most PM negotiation is internal: scope, deadlines, headcount, sequencing. The mechanics are identical, with one difference that matters. You keep working with this person on Monday, which prices relationship cost into every concession and rules out tactics that win the hour and cost the quarter.

### Which Builders Camp bootcamp covers this?

Negotiation for Product Managers, a 1 week bootcamp with 2 live sessions and 9 microlessons, opens on preparation: interests, fallbacks and constraints, then covers framing, stakeholder pressure, trades and concessions, and difficult conversations.

## Sources

- [Builders Camp: Negotiation for Product Managers bootcamp](https://builderscamp.com/bootcamps/negotiation-for-product-managers)
- [Harvard Program on Negotiation: What Is BATNA?](https://www.pon.harvard.edu/daily/batna/batna-negotiators-how-you-can-avoid-striking-out-and-create-mutual-gains-in-your-next-business-negotiation/)

## How this guide was made

Researched from Builders Camp's bootcamp, track and masterclass material and the sources listed on this page, drafted with AI, and fact-checked against every source cited.