Our approach

Our Research-Backed Approach
to AI-Powered Mediation

Every design decision traces to published research.

FairMediate isn't a chatbot that gives advice. It's a structured mediation process — private advocates, a neutral mediator, and iterative proposals — built on decades of negotiation and relationship science.

The architecture

Three agents, three roles

Traditional mediation uses three distinct roles: two advocates (or the parties themselves) and one mediator. We replicate this structure with AI — not to replace human judgment, but to make the structure accessible to anyone, anytime, for free.

Your Advocate

Listens, clarifies, surfaces interests

Helps you articulate what you actually need — not just what you're angry about. Uses techniques from relationship psychology to move from positions to interests. Your partner never sees this conversation.

Their Advocate

Same process, independently

The other party gets the same private experience. Their advocate helps them surface their own underlying needs. Neither advocate communicates with the other.

The Mediator

Synthesizes, proposes, iterates

Sees themes and priorities from both sides — never exact words. Identifies common ground, names the real tension, and proposes specific, actionable agreements.

Why three agents, not one? A single AI trying to mediate both sides simultaneously creates the same problem as direct negotiation: each party performs for the other instead of communicating honestly. Separation is the mechanism that makes mediation work — both in research and in practice.

The research

Four frameworks that shaped the design

Each framework addresses a specific failure mode in how people typically try to resolve disagreements.

Interest-based negotiation

Fisher & Ury — Getting to Yes (1981)

The insight

Most disputes aren't about positions ('I want 60%') — they're about underlying interests ('I want my work to be recognized'). When a mediator reframes positions as interests, parties discover solutions neither would have proposed alone.

How we apply it

Each advocate is designed to surface underlying interests — not just record demands. The mediator's proposal addresses interests, not positions.

Source: Harvard Negotiation Project

Private caucusing

Pre-mediation caucus research

The insight

When each party speaks privately to the mediator before the joint session, they communicate more honestly, admit their own role in the conflict, and reveal their true needs. The presence of the other party activates defensiveness that blocks resolution.

How we apply it

The private advocate conversation IS the caucus. Each person speaks freely without the other listening, producing richer and more honest input for the mediator.

Source: Caucus with Care — Conflict Resolution Quarterly

Emotional regulation through structure

Gottman Method & affective neuroscience

The insight

96% of conversations that begin with criticism fail entirely. Emotional flooding — when heart rate exceeds 100 BPM — shuts down the prefrontal cortex, making rational problem-solving physiologically impossible. Structure prevents flooding before it starts.

How we apply it

By separating each party into private conversations, the 'harsh startup' is eliminated. Text-based, asynchronous communication gives the prefrontal cortex time to stay engaged.

Source: Gottman Institute

Structured mediation outperforms direct negotiation

Bogacz, Pun & Klimecki (2020)

The insight

A randomized controlled study published in a Nature portfolio journal found that romantic couples in structured mediation were 1.39x more likely to reach agreement, reported higher satisfaction, and showed lower physiological stress than couples who negotiated directly.

How we apply it

The entire three-agent architecture — advocate, advocate, mediator — replicates the structure that produced these results, adapted for asynchronous text-based use.

Source: Nature Humanities & Social Sciences Communications

Design decisions

Why text-based and asynchronous

Text over voice

Writing engages the prefrontal cortex more than speaking. It forces slower, more deliberate communication — exactly what emotionally charged conversations need. You can't 'yell' in a text box.

Asynchronous over real-time

Parties don't need to be online at the same time. This removes scheduling friction, eliminates time pressure, and gives each person space to think before responding.

Private before shared

Each person processes their own feelings and articulates their needs before any information crosses to the other side. This is the caucusing principle — honesty requires safety.

Proposals over verdicts

The mediator proposes — it doesn't decide. Both parties can accept, suggest changes, or reject. This preserves autonomy and produces agreements people actually follow.

Boundaries

What this tool does not do

Not therapy

Therapy explores root causes over many sessions. This resolves a specific disagreement in one structured process.

Not legal advice

Agreements are between the parties — not legally binding. For legal matters, consult a qualified professional.

Not a replacement for professionals

For domestic abuse, mental health crises, or high-stakes legal disputes, professional help is essential.

Not one-sided advice

Unlike asking ChatGPT for relationship advice, both parties participate equally. The structure prevents validation-seeking.

See the approach in action

Walk through a real mediation example, or see how it applies to couples, business partners, freelancers, or roommates.