DEV Community

Jakub
Jakub

Posted on

What a framework-based AI conflict mediator does differently than a general chatbot

We built Verdict Buddy because we kept watching the same pattern: someone pastes a relationship argument into a chatbot, gets a response that mostly agrees with them, and walks away thinking they were right all along.

The problem is structural. A chatbot responds to the input it receives. If one side of a conflict writes the prompt, the model optimizes for that person's framing. The other side never gets a turn.

After 290 ratings averaging 4.9 out of 5, we think the fix is mechanical.

What a general chatbot actually does with a conflict

We ran the same argument through four general-purpose AI assistants. A recurring disagreement about splitting household tasks. Every single one:

  • Validated the person who wrote the prompt
  • Offered generic advice ("try having a calm conversation")
  • Avoided taking a position

This doesn't resolve anything. The argument happens again next week.

A chatbot doesn't know about Gottman's Four Horsemen (criticism, contempt, stonewalling, defensiveness). It doesn't check whether the conflict is solvable or perpetual. According to the Gottman Institute, 69% of relationship conflicts are perpetual. It doesn't apply Nonviolent Communication (NVC) structure or Emotionally Focused Therapy (EFT) attachment patterns.

It just responds.

How Verdict Buddy approaches the same input

Verdict Buddy is an AI conflict mediator that gives an unbiased, framework-based verdict (Gottman, EFT, NVC) on relationship, work, family or roommate conflicts in minutes.

Instead of a single prompt-response cycle, it runs a structured process:

  1. Both sides speak. Solo mode collects both perspectives from one person. Couple mode gives each person a separate input. Group mode handles multi-party conflicts.

  2. Pattern classification. The system classifies the conflict against established frameworks. Is this a Gottman Horseman pattern? An unmet attachment need (EFT)? A boundary expressed as blame (NVC)?

  3. Verdict with reasoning. The output is a specific verdict: who has the stronger position, why, and what each side can do next. Three to five actionable steps, grounded in the framework that best fits the conflict type.

The whole thing takes under five minutes.

Side-by-side comparison

General chatbot Verdict Buddy
Input structure One person writes everything Both sides get a turn
Framework None (general language model) Gottman, EFT, NVC, CBT
Bias check None Structured to weigh both sides
Output Generic advice Specific verdict + reasoning + next steps
Conflict classification None Solvable vs. perpetual, pattern type
Modes Single chat Solo / Couple / Group

How we wired four frameworks into one prompt

The temptation when you have four psychological frameworks is to run them sequentially. A checklist where every conflict gets the same four-step audit. That produces thorough but unreadable output.

What we did instead: the first pass classifies the conflict type and selects the primary framework. A workplace authority dispute routes to NVC (needs vs. strategies). A recurring couple argument about the same topic routes to Gottman (perpetual vs. solvable). A fight where one person shuts down routes to EFT (attachment patterns).

The secondary frameworks still inform the verdict, but the structure follows whichever one fits best. The result reads like a mediator's assessment, not a textbook.

The numbers behind the problem

85% of employees deal with conflict at work. They spend an average of 2.8 hours per week on it. That adds up to $359 billion in annual costs in the US alone (CPP Global / SHRM). Most of that time goes to unstructured venting, not resolution.

A five-minute structured verdict from Verdict Buddy won't replace a mediator or a therapist, but it can break the loop where the same fight keeps happening and nobody knows who has a point.

What we're tracking

This is the first Verdict Buddy post on Dev.to. We at Inithouse are measuring reactions, saves, and referral traffic to verdictbuddy.com. Two Medium posts went live earlier this week covering different angles. The Dev.to audience skews more technical, so we leaned into the prompt architecture side here.

If you've tried using a chatbot to settle an argument and felt like it just told you what you wanted to hear, try Verdict Buddy. The verdict might not be in your favor.

Top comments (1)

Collapse
 
marcusykim profile image
Marcus Kim

Solo mode still gets both perspectives from one narrator, which leaves some of the framing problem you set out to fix. Couple mode's separate inputs seem like the more meaningful change; I'd test whether reversing their order or rewriting one account more fluently changes who gets the stronger position. An explicit "not enough information" outcome would also help avoid forcing a winner from incomplete accounts. Alongside the 4.9/5 ratings, tracking whether both participants found their next conversation more useful could help distinguish a satisfying verdict from actual progress.