To prepare a negotiation plan with AI, organize verified facts, objectives, issues, authority, alternatives, questions, and conditional trades in a controlled worksheet. AI can expose gaps and help rehearse scenarios, but it must not guess the other side’s position, reveal confidential limits, interpret law, communicate externally, or make a commitment.

The general guide to using AI provides the source and review foundation. This workflow covers preparation for a commercial or operational negotiation, not autonomous negotiation or legal advice.

Key Takeaways

Define scope, participants, decision rights, and confidentiality before prompting.

Separate facts, assumptions, questions, objectives, and authority limits.

Evaluate your alternatives with evidence instead of inventing leverage.

Make concessions conditional and record who may approve each one.

Treat the other side’s interests and limits as hypotheses to test respectfully.

Use AI for challenge and rehearsal, never for external commitments.

What should you define before using AI?

Write a short control brief. Name the negotiation, business purpose, counterpart organization, participants, decision owner, authorized negotiator, confidentiality classification, topics in scope, and topics excluded. Do not include personal data or confidential limits unless the approved tool and policy permit that exact use.

Control field	Record
Negotiation ID/version	Stable identifier and revision
Purpose	Business outcome the discussion may support
Scope	Products, services, markets, dates, and issues included
Exclusions	Matters the team will not negotiate
Participants	Roles, not unnecessary personal details
Authority	Who may discuss, recommend, approve, and sign
Confidentiality	Data allowed in the AI workspace
Decision deadline	Real deadline and source
Escalation owner	Person to contact when limits are reached

The Harvard Program on Negotiation’s preparation checklist covers goals, interests, alternatives, reservation points, timing, relationships, and multiple issues.［1］ Use such questions to structure preparation, but answer them from your evidence and authority—not from a model’s confidence.

If the negotiation starts from a proposed solution or commercial case, keep the approved business proposal as a separate source. The negotiation plan records how issues may be discussed; it does not rewrite the proposal or change its approval status.

How do you build an issue and objective matrix?

List every issue separately so a headline objective cannot hide trade-offs. Distinguish what you want, why it matters, what evidence supports it, the allowed range, and who can approve a change.

Issue ID	Objective	Priority	Evidence	Authority	Unknowns
`I-01`	Delivery window	High	Operations plan `O-7`	Lead may discuss; director approves	Counterpart capacity
`I-02`	Support hours	Medium	Service need `S-3`	Lead may trade within range	Cost of extension
`I-03`	Reporting format	Low	Audit requirement `A-2`	Fixed requirement	Accepted equivalent

Use clear statuses: fact, assumption, question, proposal, authorized limit, and decision. Ask AI to preserve these labels. A discussion point is not automatically an approved offer, and an internal preference is not a contractual requirement.

For financial or operational comparisons, complete a controlled cost-benefit analysis outside the language model. Link its version and conclusions rather than asking AI to calculate or recreate them from prose.

How do you evaluate alternatives and reservation points?

Your best alternative to a negotiated agreement is what you can realistically do if no agreement is reached. It is not a threat, a wish, or an estimate of the other side’s weakness. Record the alternative, owner, availability date, dependencies, cost, evidence, and confidence.

Alternative field	Question
Description	What action can actually be taken?
Evidence	Which approved analysis supports it?
Availability	When and under what conditions is it available?
Consequences	What costs, delays, risks, and benefits follow?
Owner	Who can activate or reject it?
Confidence	What could make it unavailable?

Set target and reservation points only within authorized policy and analysis. A reservation point is an internal boundary for the team; do not paste it into an unapproved external tool or share it automatically. If the boundary depends on legal interpretation, financial authority, employment policy, or regulated procurement, obtain qualified review.

The GOV.UK Sourcing Playbook emphasizes early planning, market engagement, clear outcomes, risk allocation, and structured commercial strategy in public sourcing.［2］ It is a procurement-specific reference, not a universal rule for private negotiations. Its useful lesson here is to connect a negotiating position to documented objectives and risks.

How should you handle the other side’s interests?

Write known statements as facts with sources. Write inferred interests, constraints, authority, alternatives, and reservation points as hypotheses. The model may generate questions that test those hypotheses, but it cannot know another party’s hidden motives.

Use a hypothesis table:

Hypothesis	Basis	Confidence	Respectful question	Disconfirming signal
Schedule may matter more than format	Prior written request	Low	“Which delivery constraint is most important?”	They prioritize audit format
Approval may require another role	Meeting invitation	Low	“Who else needs to review a change?”	Named attendee has full authority

Avoid personality labels such as “aggressive,” “irrational,” or “easy to pressure.” Describe observable behavior and ask clarifying questions. Do not ask AI for manipulation, deception, impersonation, threats, or ways to exploit a suspected vulnerability.

Separate information from influence. A statement made during negotiation may be true, incomplete, mistaken, or strategic. Record it as a claim to verify rather than accepting or dismissing it based on tone.

How do you plan concessions and conditional trades?

A concession is not a free item to distribute when conversation becomes uncomfortable. Record what may move, the permitted range, what the team needs in return, the approval level, and the effect on other issues.

Trade ID	We may offer	Only if we receive	Limit	Approver	Dependencies
`T-01`	Later reporting date	Earlier source-data delivery	Approved window	Operations director	`I-01`
`T-02`	Longer term	Defined service-level protection	Approved range	Commercial owner	Legal review

Ask AI to detect one-sided concessions, circular dependencies, inconsistent units, or trades that exceed recorded authority. It may suggest questions, but it must not invent an approval or recommend hiding a material condition.

The U.S. Federal Acquisition Regulation requires documented prenegotiation objectives for covered federal pricing actions and ties them to analysis and pertinent issues.［3］ That rule applies to its procurement context. Use it as an example of why objectives, evidence, authority, and documentation should be explicit, not as legal instruction for another setting.

Order issues deliberately. Start with shared facts and process, then explore interests and lower-risk questions before decisions that require approval. Keep related issues visible so the team can evaluate conditional trades instead of conceding each point in isolation.

How do you prepare a negotiation plan with AI through challenge and rehearsal?

Give AI a redacted version of the control brief, issue matrix, permitted positions, and public or synthetic counterpart information. Ask it to challenge missing evidence, inconsistent priorities, ambiguous wording, dependency conflicts, and unanswered questions.

For rehearsal, define a role and boundaries: “Act as a skeptical procurement manager using only the stated concerns. Ask one question at a time. Do not invent confidential facts or make binding offers.” After each answer, ask the model to identify unclear claims, unsupported statements, accidental disclosures, and moments when escalation was required.

NIST’s Generative AI Profile identifies risks such as confabulation, privacy loss, harmful bias, and overreliance.［4］ Mitigate them by using synthetic or approved inputs, marking every scenario as a simulation, and reviewing the rehearsal transcript. A plausible response from the model is not evidence of how the real counterpart will behave.

Prepare failure paths:

a requested term is outside authority;
new information contradicts the plan;
the counterpart introduces an excluded issue;
a calculation cannot be verified;
a participant discloses sensitive information;
time pressure prevents qualified review;
no acceptable agreement is available.

For each path, write the pause phrase, owner, information needed, and escalation route. “I need to confirm authority before responding” is often safer than improvising.

How do you run the meeting and record decisions?

Create an agenda with purpose, issue order, participants, time boxes, questions, note owner, and decision points. Share only the external agenda, not internal reservation points, alternatives, or concession limits.

During the meeting, record proposals, questions, evidence requests, tentative language, owners, and deadlines. Label any tentative point as subject to approval. Do not let an AI note taker convert discussion into agreement or send a summary without authorized review.

After the meeting, reconcile notes against the plan. List agreed facts, open issues, proposed terms, approvals required, and the exact next communication owner. If contract wording is involved, the appropriate legal or procurement owner reviews the source document; a contract comparison can identify text changes but cannot determine legal effect.

Version the plan and decision record. Keep a trace of who changed an objective, limit, or trade and why. If conditions change materially, rerun the relevant scenario analysis instead of quietly adjusting assumptions.

Summary

A negotiation plan separates evidence, objectives, authority, alternatives, hypotheses, and decisions.
Reservation points and confidential limits remain in approved systems.
The other side’s interests are hypotheses to test, not facts for AI to invent.
Conditional trades need limits, dependencies, and named approvers.
AI can challenge and rehearse a plan but cannot communicate or commit.

FAQ

What is the minimum input for an AI negotiation plan?

Provide scope, roles, authority, issues, objectives, evidence, known constraints, alternatives, questions, confidentiality rules, and escalation owners. Leave unavailable information marked unknown.

What is a BATNA in negotiation?

A BATNA is the best realistic alternative available if the negotiation does not produce an agreement. Evaluate it with evidence, dependencies, cost, timing, ownership, and uncertainty rather than treating it as a slogan.

Can AI calculate my reservation point?

AI may organize inputs, but the authorized team must calculate and approve the boundary using controlled analysis. Legal, financial, procurement, employment, or regulated constraints require the appropriate qualified owner.

Should AI estimate the other side’s BATNA?

It may help list hypotheses and questions based on known information, but it cannot know a hidden alternative. Keep any estimate labeled uncertain and never treat it as a fact or basis for deception.

How should concessions be planned?

Record each possible concession as conditional: what may move, what must be received, the limit, dependencies, and approver. Never let the model invent authority or turn a discussion idea into an offer.

Can I upload confidential terms to an AI tool?

Only if the data classification, tool approval, access controls, and organizational policy allow that exact use. Otherwise redact the material or use synthetic ranges and keep sensitive limits in the approved system.

Is role-play with AI reliable?

Role-play is useful for testing clarity and rehearsing questions, not predicting the counterpart. The simulation must stay within supplied facts and be reviewed for invented information and accidental disclosures.

Who approves the final negotiation plan?

The negotiation or decision owner coordinates commercial, operational, financial, legal, procurement, HR, or other required reviewers. Approval must identify the exact version and authority granted.

Disclaimer: This article provides general preparation information, not legal, procurement, employment, financial, or professional advice. Requirements and authority vary by organization, transaction, and jurisdiction.

Sources

Program on Negotiation at Harvard Law School — A Negotiation Preparation Checklist — https://www.pon.harvard.edu/daily/negotiation-skills-daily/negotiation-preparation-checklist/
GOV.UK — The Sourcing Playbook — https://www.gov.uk/government/publications/the-sourcing-and-consultancy-playbooks/the-sourcing-playbook-html
Acquisition.gov — Federal Acquisition Regulation Part 15, Contracting by Negotiation — https://www.acquisition.gov/far/part-15
NIST — Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile — https://www.nist.gov/publications/artificial-intelligence-risk-management-framework-generative-artificial-intelligence

Sources checked 4 September 2026.

How to Prepare a Negotiation Plan with AI