The AI Read the Room Better Than the Negotiators Did

Dear Negotiation Explorer,

Last Thursday, while issue 47 was landing in your inbox, I was on stage at Sciences Po in Paris. The European Negotiation Conference had put four of us on one panel — AI and Human Interaction in Negotiation — and I had the last slot, presenting the work behind the paper I shared with you last week.

The three presentations before mine were the most interesting hour I've spent this year. Three research teams, three different questions, all testing what happens when AI sits across the negotiation table. And every result pointed the same way.

Let me show you.

Nazli Bhatia, from the University of Pennsylvania, presented work she developed with Zikun Yang and Sudeep Bhatia, asking a question so basic it's surprising nobody had studied it directly: can you tell how satisfied your counterpart is with the deal you just closed?

You'd think so. Satisfaction matters — a counterpart who walks away satisfied comes back, refers you, implements the agreement. And most negotiators believe they can read the other side.

The Penn team had people negotiate over chat, then predict their partner's satisfaction. Then they gave the same transcripts to an AI and asked it to do the same.

The negotiators couldn't do it. Their predictions had essentially no connection to how their partner actually felt. The AI, reading the same conversation, predicted the partner's satisfaction remarkably well.

The mechanism is the humbling part. People anchored on their own feelings: "I'm satisfied, so they must be." And the language analysis showed something sharper — negotiators were listening to their own words when judging their partner, while the signal was sitting in the partner's words all along. That's where the AI was looking.

We assume reading people is the human advantage. In this study, it was the machine's.

And Nazli isn't leaving the finding on the slide. She shared that alongside her core negotiation course at Penn, she's introducing a new course on AI and negotiation — where a tool like this could give students real-time feedback on how their counterpart actually feels. The research and the teaching, moving together.

Leonardo Caporarello, from SDA Bocconi in Milan — a colleague I was particularly glad to share the stage with — asked a different question: does negotiating with an AI change what you feel at the table?

It does. In his experiment, some people negotiated with humans, others with an AI counterpart, fully disclosed. The striking result: negative emotions did less damage when the counterpart was an AI. People regulated themselves differently. As Leonardo put it, describing what he saw in the room: with a machine, people think "even if I answer badly, nothing will change" — and so the spiral that poisons human negotiations never starts.

The table has an emotional climate, and AI changes it. Work in progress, as Leonardo was careful to say — but it opens a door I haven't seen anyone else open. And the presentation was only half of what impressed me. Talking with him privately afterwards, I got a glimpse of the vision he's driving at Bocconi — bringing AI into education across the school, not just into one course. Most of us are still working out what AI means for our own classroom; Leonardo is thinking at the level of an institution.

The third study might be the most elegant of the three. Alexandra Mislin, from American University, presented it — work developed with the legendary Daniel Druckman (George Mason University) and Louise Serre. And I have to say something about the presentation itself: compelling, welcoming, warm. Alexandra made a room full of specialists feel like guests. The findings would have landed on their own; the way she delivered them was a lesson in communication.

Everyone in her study negotiated an apartment lease against the same AI. Half were told the broker was human. Half were told it was AI.

Same counterpart. Same conversation. One belief changed.

The people who believed they were facing a human agreed to pay more — and felt warmer about the relationship, trusted more, described the broker as a partner ("the broker wanted to help me"). The people who believed they were facing an AI negotiated better deals — and described the encounter in the language of manipulation: "it was a program I could manipulate."

There is a human premium in negotiation. And it's a genuine trade-off: believing you face a human buys you a relationship and costs you money.

Sit with what these three findings say together.

Reading the other side's satisfaction — the empathic core of negotiation? The AI did it better. The emotional weather at the table? The AI changes it. The relationship itself? It moves on your beliefs about who you're facing, not on who is actually there.

The parts of negotiation we call "inherently human" are exactly where the research keeps finding us weakest — and the machine strongest.

Then it was my turn.

All three studies tested AI across the table — as counterpart, as observer. My work starts from the other side: the AI as your partner, on your side.

Regular readers know the arc. My 2024 study: executives who prepared with AI under guided instructions improved individual gains by 48.2% as buyers and 40.6% as sellers — and when both sides prepared with AI, joint gains rose 84.4%. The prize is real. But generic AI needed an expert steering it — so I built an expert-level negotiation agent instead. Deepak: built through an eight-phase framework that runs from understanding the domain to real-world iteration, on one cardinal rule — keep what the agent knows (the knowledge base) separate from how it behaves (the instructions). Eight refinement cycles, about 45 hours of collaboration. In my assessment, its preparation now goes beyond what the best human negotiators produce on their own.

And then the newest part — the one you read last week: the memory system that turns the AI from a capable assistant into a thinking partner that gets sharper every session.

One moment from the discussion stays with me. Claude Bruderlein — Harvard, organizer of the AI Negotiation Forum, and a good friend — pushed me, fairly, on whether all this means AI simply replaces the negotiator: if it's that good, why keep the human? My answer: companies have tried automating negotiation, and many are walking it back. Automation keeps failing; augmentation keeps working. The future isn't the machine at the table instead of you. It's the augmented negotiator — negotiation expertise plus AI literacy — in a thinking partnership with the AI. That partnership is the real synergy: human and AI each bringing what the other lacks, getting sharper together every session.

That's also why the three studies before mine made me more convinced, not less. If AI reads satisfaction better than you do — put that reading on your side. If beliefs about the counterpart move outcomes — you'd better understand that before your counterpart does. Everything those teams found across the table becomes an advantage the moment the AI moves to yours.

Someone who had attended the same panel a year ago told me the conversation was unrecognizable — last year we debated whether any of this was coming; this year four presenters showed data on how it already works. That's the pace.

You don't need to build Deepak to act on this. The question to take into your week is simpler: where is AI at your table today? Across from you — a counterpart, a screening tool, a broker you may not even know is a machine? Or beside you — reading the room with you, preparing with you, remembering with you?

The research says the across-the-table version is already better than we are at things we thought were ours. The beside-you version is the one you control.

Both papers behind my talk are free on SSRN — the framework for building the agent, and the collaboration memory that turns it into a partner:

From Preparation Gap to Augmented Expert → https://papers.ssrn.com/sol3/papers.cfm?abstract_id=6308259 Structuring Collaboration Memory for AI-Augmented Negotiation → https://papers.ssrn.com/sol3/papers.cfm?abstract_id=6882538

Where is AI at your negotiation table right now — across from you, beside you, or not in the room yet?

Just hit reply and tell me. I read and answer every one.