🤖 AI-augmented negotiation
AI & decision support
AI-assisted negotiation: research on negotiating agents (N. Brown et al., CICERO, Science 2022), decision support and bias detection; preparation and debriefing with large language models.
Full detail in the “Origin & history” section below.
Author of the library
Negotiation lecturer at KEDGE Business School
A rare threefold background serving negotiation: teaching, commercial justice and audit, backed by experience as a company director in Bordeaux.
Prompt engineering applied to negotiation consists in methodically formulating the instructions given to a language model (Claude, ChatGPT, Gemini) in order to obtain a strategy, a meeting script, an objection matrix or counter-arguments that are genuinely usable. The quality of the output depends directly on the structure of the prompt: assigning a role, precise factual context, explicit constraints, expected response format. Well practised, the technique turns the AI into a preparation sparring partner, able to simulate the opponent, stress-test an offer and generate variants in seconds. Badly practised (vague prompts, no context), it produces generic platitudes and a false sense of mastery.
At a glance
Vigilance: moderate (4.0/10) · Preparation required: 8/10
Indicative profile: it situates the “Négociation augmentée par l'IA” family as the AI-augmented negotiation school practises it, not this technique taken in isolation. Techniques from the same family and school therefore share the same profile. NEGOCOACH editorial rating out of 10, non-experimental · the higher the “relational risk” value, the more costly the technique is to the relationship.
NEGOCOACH assessment
Tactical potential 5.7/10 (effectiveness, impact, discretion) and vigilance moderate (relational and ethical risk): two distinct readings, deliberately never merged into a single score that would reward risk. NEGOCOACH editorial rating calibrated from the “Négociation augmentée par l'IA” family and the “AI-augmented negotiation” school. Each criterion is rated out of 10; click to understand what it measures.
How far the technique can carry the negotiation in the intended direction when it is well executed.
Strength of the effect produced on the counterpart's perceptions, emotions and decisions.
How hard it is for the other party to notice the technique is being used. A high value = very discreet.
The information, analysis and rehearsal required upfront to use it effectively.
Potential cost to the relationship and to trust if the technique is spotted, refused or fails. A high value = riskier.
Moral acceptability: fairness, transparency and respect for the counterpart's autonomy. A high value = more defensible.
Documented school
The school this technique stems from is documented by recognised work and established practice, without experimental consensus. This indicator qualifies the school, not this technique taken in isolation.
Indicative NEGOCOACH editorial rating, for teaching purposes. For “Relational risk”, a high value signals a cost to the relationship, not a quality.
The notion of a "prompt" as a natural-language prefix guiding a generative model emerges with GPT-2 (2019) then GPT-3 (Brown et al., OpenAI, 2020). The term "prompt engineering" was formalised in 2021, notably with Laria Reynolds and Kyle McDonell's paper "Prompt Programming for Large Language Models" (2021). The rise of structured techniques follows with the "chain-of-thought" of Jason Wei et al. (Google, NeurIPS 2022) and becomes systematised in "The Prompt Report" (Schulhoff et al., 2024). The specific application to negotiation has no single founding author: it results from the transposition of these generic techniques by sales, procurement and training practitioners since 2023.
Prompt engineering for negotiating is the discipline of designing structured instructions addressed to a language model in order to produce negotiation deliverables. An effective prompt articulates four building blocks: (1) a ROLE assigned to the model ("You are a senior car-industry buyer"), (2) a dense factual CONTEXT (stakes, history, balance of power, known BATNA), (3) explicit CONSTRAINTS (objective, red line, tone, length), and (4) a usable output FORMAT (objection/response table, three-step script, ranked list of concessions). Iteration, rephrasing, refining, contradicting the first response, is an integral part of the method.
Application by context
A salesperson gives the model the role of "sceptical, time-pressed head of procurement", the context of their offer and their floor price, then asks for the list of the ten most likely objections, each with a costed response.
A buyer prompts the AI by giving it the supplier's quote, comparable market prices and their target discount, and has it draft a counter-proposal email that is firm but preserves the relationship.
An HR director describes the state of a pay dispute (union demands, budget leeway, red lines) and asks the model for a grid of graduated concessions with the associated session-opening message.
A crisis negotiator simulates, via a prompt playing the role of an agitated counterpart, several escalation replies in order to test their calming and time-gaining techniques in advance.
An adviser describes the positions of two camps on a text and asks the model for compromise wordings "that let each party claim a victory" to prepare the mediation.
A buyer gives the asking price, the defects observed and their maximum budget, then has a price-reduction argument generated, backed by factual points and a proposal for a justified low anchor.
An executive prepares a negotiation in Japan by prompting the model about the other side's codes (importance of consensus, indirect communication) to adapt the pace and wording of their offer.
In a tense inheritance, a person describes the interests of each heir and asks the model for an agenda and open questions to defuse positions and bring out the real needs.
Counter-techniques
Negotiation is also played on defence. Here is how to recognise this technique when it is used against you, and turn it around.
The signals that give it away
The counters that defuse it
Turn it into an advantage
Name the manoeuvre: said out loud, a technique loses most of its power.
Reacting emotionally instead of coming back to the facts.
Strengths: speed of production (several hours of preparation condensed into minutes); the ability to simulate the other side and to shift the negotiator away from their blind spots; the generation of variants and creative options; the democratisation of methodical preparation, including for lightly trained profiles; permanent availability as a sparring partner. Weaknesses: it depends entirely on the quality and honesty of the context provided ("garbage in, garbage out"); a risk of hallucinations (invented figures, legal precedents or clauses); generic output if the prompt is vague; the model's sycophancy bias, which tends to validate the user; the confidentiality of the data entered; a false sense of mastery that can substitute for critical analysis.
Particularly useful in the PREPARATION phase: structuring a strategy, mapping objections, drafting and calibrating a script, simulating the opponent, generating counter-arguments. Relevant when time is short, when you want to break out of your reflexes, or to debrief afterwards. To be kept offline: it is a backstage tool, not a pilot in session. Poorly suited to situations where the context is too sensitive to be entered (confidential data, defence secrets) or when the human relationship and intuition take precedence over a prepared argument.
Sales · The objection table before a key-account meeting, A sales manager at a software SME prepares a renewal negotiation with a client threatening to switch to the competition. They write a structured prompt: role ("You are this client's CFO, reputed to be tough and ROI-focused"), context (contract value, history of incidents, competitor's price), and ask for a three-column table, likely objection, hidden intention behind the objection, recommended response. The model produces twelve objections, three of which the salesperson had not thought of, notably on migration costs. In session, they precisely head off these three points and keep the account. A representative case of common usage, not attributed to an identified person.
Business · Simulating the union before a mandatory annual negotiation, An HR director prepares the annual pay round (NAO). She prompts the model, having it play in turn an intransigent union delegate then a moderate delegate, with the real context of the demands and the available budget envelope. She thus tests several opening wordings and spots which one triggers the least escalation, before refining her grid of graduated concessions. A representative scenario illustrating opponent simulation, without attribution to a verified real case.
Faced with an opponent likely to use AI: their arguments may be exhaustive and well calibrated but sometimes too smooth, generic or disconnected from the specifics of your file. To defend yourself, ask very concrete, contextual questions that only real field knowledge can handle; introduce unexpected elements that were not in their prompt; test the depth by digging into a point of detail. An AI-prepared argument gives itself away by its overly perfect structure and its difficulty improvising on an unexpected angle. Recognise your own dependence too: if you cannot defend an argument without your sheet, you have not mastered it.
Limits: the model does not know the real human relationship, the non-verbal cues, the emotional history or the weak signals of the session; it can hallucinate facts; its outputs reflect the biases of its training data. Ethics: transparency about the use of AI in certain settings (labour relations, mediation); confidentiality and GDPR (RGPD) on the data entered (do not disclose information covered by a secret or third parties' personal data); a risk of manipulation if the tool is used to fabricate misleading arguments or false precedents. AI remains a decision aid: the ethical and legal responsibility stays entirely with the human negotiator.
Related techniques: "role prompting" (assigning an expert identity to the model); "chain-of-thought" (asking for step-by-step reasoning); "few-shot prompting" (giving examples of expected outputs); AI role-play / role simulation (having it embody the opponent); "red-teaming" your own offer (asking the AI to attack your position); AI-assisted debriefing after a negotiation. These building blocks combine with the classic preparation techniques (Harvard method, concession matrix, BATNA/MESORE preparation).
Quick exercise
Answer in your head, then reveal the solution. Memory is built through active recall.
Frequently asked questions
Prompt engineering applied to negotiation consists in methodically formulating the instructions given to a language model (Claude, ChatGPT, Gemini) in order to obtain a strategy, a meeting script, an objection matrix or counter-arguments that are genuinely usable. The quality of the output depends directly on the structure of the prompt: assigning a role, precise factual context, explicit constraints, expected response format. Well practised, the technique turns the AI into a preparation sparring partner, able to simulate the opponent, stress-test an offer and generate variants in seconds. Badly practised (vague prompts, no context), it produces generic platitudes and a false sense of mastery.
It sits on the line: effective, but it can tip into manipulation if it exploits an information asymmetry. Use it with measure and without deliberate deceit.
Reacting emotionally instead of coming back to the facts. The right reflex: slow down and reformulate.
NEGOCOACH does not assess the experimental validation of this technique in isolation. What we document is the grounding of its source school (AI-augmented negotiation): documented school. Full detail is in the "At a glance" section of this page.
Practise with AI
Copy, paste into your assistant, replace the [brackets]. Works with ChatGPT, Claude, Gemini, Mistral, Perplexity.
Build your plan before the meeting
You are an expert negotiation coach. Help me prepare to use the "Prompt Engineering for Negotiation" technique in the following situation: [describe your situation]. Give me: the conditions for success, a 3-step script, my counterpart's likely objections and how to answer them.
Rehearse against an AI counterpart
Play the role of my counterpart in a negotiation. I am going to test the "Prompt Engineering for Negotiation" technique. React realistically and with resistance, do not give in too quickly, then at the end analyse my performance and suggest 3 concrete improvements.
Analyse a past negotiation
Here is how my negotiation went: [paste the exchanges]. Analyse whether the "Prompt Engineering for Negotiation" technique was used well, what worked, the mistakes made, and spell out precisely what I could have done better.
References
Founding works of the 🤖 AI-augmented negotiation school this technique belongs to.
Human-level play in the game of Diplomacy (CICERO), Science
ArticleFAIR (Meta) - N. Brown et al. · 2022
Noise: A Flaw in Human Judgment
BookD. Kahneman, O. Sibony & C. Sunstein · 2021
Co-Intelligence: Living and Working with AI
BookE. Mollick · 2024
AI-assisted negotiation: research on negotiating agents (N. Brown et al., CICERO, Science 2022), decision support and bias detection; preparation and debriefing with large language models.
On video
Videos to picture Prompt Engineering for Negotiation and anchor it through examples.
A verified video selection is being enriched; the search above already surfaces the best videos on the topic.
Technique map
Every technique sits within a network: what it draws on, what it combines with, where it applies, and how to defend against it.
Spot its signals, neutralise it and turn it around with the defensive playbook on this page.
See the counter-techniquesPrompt engineering applied to negotiation consists in methodically formulating the instructions given to a language model (Claude, ChatGPT, Gemini) in order to obtain a strategy, a meeting script, an objection matrix or counter-arguments that are genuinely usable. The quality of the output depends directly on the structure of the prompt: assigning a role, precise factual context, explicit constraints, expected response format. Well practised, the technique turns the AI into a preparation sparring partner, able to simulate the opponent, stress-test an offer and generate variants in seconds. Badly practised (vague prompts, no context), it produces generic platitudes and a false sense of mastery.
Name the manoeuvre: said out loud, a technique loses most of its power.
Reacting emotionally instead of coming back to the facts.
Our programmes turn theory into a concrete advantage.