Using AI to Prep for a Negotiation: What Actually Works
Generic AI negotiation advice is genuinely useful and genuinely limited. Here's what Claude and ChatGPT get right before a negotiation, four prompts that work today, and what changes once the AI has a documented framework instead of a generic one.
Ask any AI a generic question about an upcoming negotiation and you'll get a genuinely reasonable answer: know your BATNA, listen more than you talk, don't anchor too aggressively, look for win-win outcomes. None of it is wrong. Almost none of it is specific enough to use. It's the negotiation-advice equivalent of "eat better and exercise more," true in a way that doesn't change what you actually do on Tuesday.
The honest starting point for this guide is that generic AI prep isn't bad, it's incomplete. It's genuinely good at helping you organize what you already half-know and genuinely weak at telling you the thing you didn't know to ask.
What does generic AI negotiation prep actually produce?
Type "help me negotiate a higher salary" into any chatbot and you'll get a competent, structured, forgettable answer. Research your market rate. Let them name a number first if you can. Practice your opening line. Stay calm if they push back. It reads like it was assembled from every negotiation article published in the last decade, because functionally, it was.
The problem isn't inaccuracy. It's that none of it is grounded in your actual situation: your specific manager, your specific leverage (or lack of it), what happened the last time you asked, what you'd genuinely do if the answer is no. Generic prep treats every negotiation as structurally identical. Real negotiations aren't.
Where is AI genuinely good, and where does it fall short, unaided?
Unaided, a general-purpose AI is genuinely useful for two things: organizing information you already have into a structure you can act on, and generating options you hadn't considered, alternate framings, questions you could ask, concessions you could offer instead of the one you were about to default to.
It's weak at three things specifically: holding a consistent negotiating stance across a long conversation (it will happily talk you out of your own position if you push back), knowing when a plausible-sounding answer is actually just plausible-sounding, and predicting the other party's actual behavior with any reliability. Treat its output as a well-organized first draft of your own thinking, not as intelligence about the room you're walking into.
What prompts actually work for negotiation prep today?
These four work in Claude, ChatGPT, or Gemini with no special setup. Use them in sequence.
1. The interest map. "I'm negotiating [situation]. Here's what I want and why, and here's what I know about what they want and why: [details]. List the interests underneath both of our stated positions, not just the positions themselves, and flag any interest of theirs I might be missing." Why it works: negotiation theory since Fisher and Ury has drawn a hard line between positions (what someone says they want) and interests (why they want it), and most unprepared negotiators only ever map the position. Forcing the interest layer surfaces trades neither side would find by arguing the stated numbers.
2. The leverage audit. "Given [situation], list my actual leverage, not what I wish I had, and list theirs. Then tell me honestly whether my walk-away position is credible or if I'm bluffing." Why it works: an AI has no ego investment in your leverage being stronger than it is, which makes it a genuinely useful check against the self-flattering version of your position you'd otherwise walk in with.
3. The objection pre-mortem. "Before I go into this negotiation, list the five most likely objections or pushback points, ranked by how likely each is, and a one-line response to each that doesn't sound defensive." Why it works: most people freeze in a negotiation not because the objection is unanswerable, but because they're hearing it for the first time in the room. Pre-mortem prep converts a live surprise into a rehearsed beat.
4. The roleplay. "Play the other party in this negotiation. Take a realistic hard-but-fair stance based on [their likely position]. I'll make my case, and you respond as they plausibly would, then tell me afterward where my argument was weak." Why it works: rehearsal under mild pressure is a documented way to reduce in-the-moment freezing, and an AI is a patient, always-available sparring partner for it, even if its impression of the other party is necessarily a guess.
Where does this stop working?
Run all four prompts and you'll walk in better prepared than most people who show up unprepared. You still won't have a method. Each negotiation gets a fresh, disconnected set of answers, with no consistent read on when to concede, when to hold, or how to actually get the other side talking instead of positioning. The gap isn't that the prompts are bad; it's that good prompts get you most of the way there, and the missing part is a repeatable diagnostic method, not more prompts.
What changes with a documented framework instead of a generic prompt?
Chris Voss, a former FBI lead hostage negotiator, documented a specific method in Never Split the Difference: tactical empathy, naming the other side's emotional position out loud before making any ask, and calibrated questions, open "how" and "what" questions that make the other party solve your problem for you instead of you asking for concessions directly.
Before, generic prompting: asked to help with a client pushing back on a price increase, a generic AI response suggests "acknowledge their concern, explain your value, offer a small concession if needed." Reasonable. Also exactly what the client expects and has heard from every vendor.
After, Voss's framework applied: instead of defending the price, the framework directs a labeling move first, "It sounds like this increase is landing at a bad time for your budget," said as a flat observation, not a question, which is documented specifically to lower the other party's defensiveness before any ask is made. Then a calibrated question replaces the justification: "How would you like me to handle this?" This hands the constraint back to the client to solve, which the documented method argues surfaces their actual budget reality (the thing you needed to know) rather than provoking a defensive counter-position.
The difference isn't that one prompt is smarter. It's that the framework gives the AI, and you, a consistent sequence to run every time: label the emotion, then ask a calibrated question, in that order, regardless of which negotiation you're in.
How do you actually load a framework into Claude, ChatGPT, or Gemini?
In Claude, a framework installs as a Skill: go to Settings → Skills → Add skill and upload the .zip. Claude reads the skill's description and auto-invokes it when your question matches, or you can force it with /chris-voss-framework at the start of a message. Claude Code users get the same thing via ~/.claude/skills (every project) or a project's .claude/skills folder (that project only).
In ChatGPT, the plain .md file goes into a Custom GPT: Create a GPT → Configure → Instructions, paste the framework's content in. Standard Custom Instructions (the "About me" fields) cap out at 1,500 characters, too short for a full framework, so use a Custom GPT specifically, not Custom Instructions.
In Gemini, paste the .md content directly into the system prompt if the interface you're using supports one, or lead your first message with it: "Here is a thinking framework to apply throughout this conversation. Read it carefully, then answer using this framework's approach," followed by the file's contents.
Where should you go from here?
The Negotiator category collects Chris Voss's documented method along with Fisher, Ury, Heen, Malhotra, and Noesner, as .md files that work in Claude, ChatGPT, or Gemini. If there's a specific negotiation on your calendar rather than a discipline you're building over time, Negotiation Prep is a $99 tool built to run this exact sequence, your leverage, their likely BATNA, and your concession order, before you're in the room.
Frequently asked questions
Is AI negotiation prep actually useful, or is this just a way to sell frameworks?
It's genuinely useful on its own. A few good prompts before a negotiation will surface interests, leverage, and likely objections you hadn't fully mapped, and that alone improves most people's prep. The honest claim here is narrower than 'useless without a framework': generic prep gets you organized, a documented framework gets you a consistent method you can run the same way every time, across different deals, without reinventing the questions each time.
Which AI is best for negotiation prep, Claude, ChatGPT, or Gemini?
For prep specifically, the differences are small. All three can hold the four-prompt sequence in this guide in a single conversation and reason through it competently. The bigger difference shows up in how you carry a framework forward: Claude Skills auto-invoke based on the situation you describe, ChatGPT needs a Custom GPT or a fresh paste each time, and Gemini works best pasted directly into the system prompt or the first message.
Can AI tell me what the other side will actually do?
No, and treat any confident prediction with suspicion. AI can help you generate plausible interests and likely objections based on the situation you describe, which is useful structured guessing. It has no access to the other party's actual private information, so anything that sounds like a firm prediction is pattern-matching on your description, not intelligence about the room. Use its output to widen the range of scenarios you've rehearsed for, not to narrow down to one confident forecast of how the meeting will actually go.
Do I need to tell the AI everything about the deal for this to work?
You need to tell it enough that the questions stop being generic. Vague inputs ("help me negotiate a raise") produce vague outputs. Specific inputs, your actual numbers, their likely constraints, what happened last time you asked, what you'd do if they said no, produce specific, usable prep. The quality of what comes back tracks the quality and specificity of what you put in almost exactly.
What's the single biggest mistake people make prepping with AI?
Accepting the first draft. AI's first answer to a negotiation prompt is usually competent and generic: reasonable talking points, a diplomatic tone, no real edge. The useful move is pushing back at least once, asking it to name your actual walk-away point, or to argue the other side's position as hard as it can. That second pass is where the useful material usually shows up.
Does this replace reading an actual negotiation book?
No, and it isn't trying to. The prompts in this guide compress a documented method into something usable in ten minutes before a call. The books, Chris Voss's Never Split the Difference among them, contain the reasoning behind the method, worked case studies, and the judgment calls a prompt can't fully carry. Use both: the book for the underlying understanding, the prompt for the specific negotiation in front of you this week.
Will this make me sound manipulative?
Not if you use it as intended. Tactical empathy and calibrated questions, the method behind the worked example in this guide, are documented specifically as a way to get to the other side's real position without pressure tactics, which is the opposite of manipulation in the adversarial sense. The prompts here are built to help you understand the other party's actual constraints, not to trick them into a worse deal than they'd knowingly accept.
How is this different from just asking AI to roleplay the other person?
Roleplay is one of the four prompts below, and it's a genuinely useful rehearsal tool. The difference is that roleplay alone tests your talking points against a simulation; the full prep sequence, interests, leverage, objections, and only then roleplay, makes sure you know what you're actually trying to achieve before you rehearse defending it. Skipping straight to roleplay is how people walk in well-rehearsed at defending a position they never actually stress-tested.
Written by Gareth Hoyle. Last updated 26 September 2026. Part of the authority.md guides library.
More guides.
Bringing AI Into a Creative Project Without Ruining It
What generic AI brainstorming actually gives a creative project and what it quietly flattens, four prompts that help today without taking over, and what changes when the AI applies Rick Rubin's documented listening practice instead of generic idea generation.
AI for Hiring Decisions: Where Claude and ChatGPT Actually Help
A grounded look at what generic AI use gets right and wrong when you're deciding on a hire, four prompts that genuinely sharpen the decision, and what changes when the AI applies a documented evidence-based hiring framework instead of generic interview advice.
AI Won't Write Your Best Work, But It Can Improve It
What generic AI editing actually fixes and flattens in a piece of writing, four prompts that sharpen a real draft today, and what changes when the AI applies Joan Didion's documented precision instead of generic style advice.