Increasingly, workers, job seekers, and students are turning to artificial intelligence (AI) tools such as ChatGPT for negotiation advice. AI negotiation tools can offer guidance on how to navigate specific negotiation situations, such as salary negotiation. It can also describe more general negotiation strategies such as how to create or claim value.
However, AI’s responses can be unreliable. Chatbots are prone to “hallucinations”—that is, they sometimes fabricate information—and their responses may be based on incomplete, outdated, or biased knowledge. And when offering negotiation advice, chatbots have been found to be “pushovers,” conceding too much too early.
Claim your FREE copy: Negotiation Skills
Build powerful negotiation skills and become a better dealmaker and leader. Download our FREE special report, Negotiation Skills: Negotiation Strategies and Negotiation Techniques to Help You Become a Better Negotiator, from the Program on Negotiation at Harvard Law School.
Yet negotiation experts have found that when they create their own AI negotiation bots, tailored to a specific learning environment, these bots can be highly effective negotiation coaches. For their multiparty negotiations courses at MIT, Harvard Negotiation Project instructor Samuel Dinnar and MIT professor Lawrence Susskind, aided by MIT computer scientist Leroy K. Sibanda and MIT graduate Ololade Olaleye, created AI negotiation bots that have helped students learn more from complex dealmaking simulations.
Building AI Negotiation Bots for Deeper Learning
For these MIT negotiation classes, the AI negotiation bots were designed to encourage each student to develop “their own personal theory of practice,” rather than to teach them to negotiate uniformly, said Dinnar at the Program on Negotiation’s March 2025 AI Negotiation Summit. (Videos of Summit presentations, including Dinnar’s, can be found here.)
In Harborco, one of the complex multiparty negotiation simulations used in the MIT classes, students are assigned to represent one of six organizations (a port developer, a labor union, an environmental coalition, other regional ports, the governor’s office, and the federal department of coastal resources). The students prepare by studying background materials and by answering questions from “coaching bots.” In small groups, the students attempt to negotiate the construction of a new deepwater port. After the negotiation, a “debriefing bot” helps students reflect on their performance and underscores lessons learned.
Back-Table Bots
The MIT teaching team also created a “back-table bot” specifically trained to role-play members of the other side’s so-called back table, or behind-the-scenes stakeholders, such as colleagues, constituents, and supervisors. These back-table parties often profoundly shape the proposals and agreements that parties consider yet can be difficult to include in negotiation practice in the classroom.
The back-table bot enables students to interact with various stakeholders to assess their interests and positions. For example, a student playing the environmental group’s representative might engage in a chat conversation with the bot playing the role of the local mayor’s chief of staff to assess how to get the mayor on board with a proposal.
“By engaging with the other sides’ back-table stakeholders, these behind-the-scenes negotiators can sometimes explore ‘out-of-the-box’ trade-offs and creative solutions in a protected setting,” the MIT team writes in an open-access article for the Negotiation Journal.
After participating in the simulation with the help of the three bots (preparation, back-table, and debriefing bots), 82% of students reported that the bots had helped them feel better prepared for the negotiation. In addition, 77% said the bots helped them better identify the other side’s interests.
Engagement with the back-table bot, in particular, helped refine students’ “personal theory of practice.” In fact, “the majority . . . reported that interactions with the bots prompted them to rethink and refine their negotiation strategies and theoretical approaches,” the MIT teaching team reports.
Designing Effective AI Negotiation Bots
In their article, Dinnar, Susskind, and their team offer a seven-step process to help instructors design AI negotiation bots that can enhance role-play simulations in the classroom:
- Pedagogy process: Carefully choose the case to teach based on your pedagogical objectives and decide where it fits within the class syllabus. For example, a multiparty negotiation with opportunities for a back-table AI negotiation bot might teach concepts such as agreeing on ground rules, building coalitions, and creating value through trade-offs.
- Pedagogy objectives: Define tasks and patterns of engagement for AI negotiation bots that reflect teaching objectives specific to the case and to each role within the case. For example, a preparation AI negotiation bot might be designed to clarify each role’s priorities and anticipate the other side’s negotiation strategies.
- User experience design: Describe how you would like students to interact with and learn from the AI negotiation bot. Draft introductory statements, possible exchanges, and the end of each chat session. Give back-table negotiation bots a distinct communication style or “personality.”
- Integration planning: Plan to integrate and test AI negotiation bots based on a thorough understanding of the requirements and constraints of current AI tools. For example, be aware of restrictions on message and session length, and the amount of material that can be used in a prompt.
- Prompt design: Develop and test each bot’s AI prompt (the specific instruction given to the bot) by engaging with a chosen large language model, such as ChatGPT. The MIT team designed prompts by focusing on the substance and style of conversations, communication session dynamics, and the bot’s “memory” of past conversations.
- Integration debugging: Integrate your tested prompts into the learning system’s interface and perform integration testing.
- Refinement: Evaluate and adapt your team’s earlier choices of prompts through further testing. Repeat until each prompt produces the desired effect.
In the classroom, AI negotiation bots don’t substitute for human instruction, negotiation theory, and student interactions, but rather underscore them by increasing opportunities for practice and inquiry.
What other uses do you see for AI negotiation bots in negotiation training and instruction?
Claim your FREE copy: Negotiation Skills
Build powerful negotiation skills and become a better dealmaker and leader. Download our FREE special report, Negotiation Skills: Negotiation Strategies and Negotiation Techniques to Help You Become a Better Negotiator, from the Program on Negotiation at Harvard Law School.
