Dear Negotiation Explorer,

Welcome to Week 21 of our NegoAI series.

Last week, we explored the structured prompt framework (Role → Context → Instructions → Criteria → Output) and how voice prompting transforms AI from a search tool into a thinking partner.

But even with strong prompting, a crucial question remains:

How do you know whether the AI’s negotiation analysis is any good?

Because negotiation decisions rely on solid reasoning, we need a structured way to evaluate the quality of the AI’s analysis.

This is where a rigorous Quality Assessment (QA) system becomes indispensable.

AI agents are specialized systems designed to perform specific tasks autonomously within defined parameters. Unlike general LLMs, agents focus on particular functions and can work together in sequences to solve complex problems.

Research shows that effective negotiation preparation involves understanding counterparts' behavioral tendencies and communication preferences. Traditional approaches rely on subjective assessment, but AI agents can now systematically analyze digital footprints to generate evidence-based behavioral profiles.

Key components of functional AI agents include:

LLMs have several structural limitations, and the three we are tackling at this stage are the ones that most directly affect the quality and reliability of negotiation analysis.

LLMs often produce different results even when given the same prompt and materials.

In negotiation preparation — where alignment and repeatability matter — this inconsistency is a real risk.

LLMs may generate incorrect, invented, or unjustified statements while sounding fully confident.

In negotiation, hallucinations can distort:

LLMs do not reveal the reasoning behind their conclusions.

You cannot see:

Because these constraints are structural, you should never rely on the first AI output without evaluating it.

Quality Assessment is what turns AI into a reliable analytical partner.

A four-criterion system to evaluate any AI-generated negotiation analysis

Before diving into the detailed scoring rules, here is a simple overview table of the four criteria and their weights:

Quality Assessment Criteria & Weights

Now let’s break down each criterion in detail:

Weight: 30%

Evaluate whether the AI captures the full spectrum of interests — not just positions.

Score 5 if:

Score 1 if:

Weight: 25%

Assess whether alternatives and consequences are analyzed properly.

Score 5 if:

Score 1 if:

Weight: 25%

Determine whether the output generates multiple, distinct, realistic scenarios for the counterpart.

Score 5 if:

Score 1 if:

Weight: 20%

Evaluate whether the AI proposes options beyond price.

Score 5 if:

Score 1 if:

Final Score = (Interests × 0.30) + (BATNA × 0.25) + (Scenarios × 0.25) + (Value Creation × 0.20)

Interpretation:

4.5–5.0 → Expert-level analysis

4.0–4.4 → Strong and reliable

3.5–3.9 → Acceptable but requires refinement

3.0–3.4 → Major iteration required

<3.0 → Reject the analysis and re-prompt

Want to see this in action?

On November 20, I'm running a free 30-minute session where I’ll voice-prompt a full negotiation case study and demo the no-code workflow — built on negotiation and AI expertise, not AI hype.

Register here

Know someone who’d benefit? Tag them below.

Feel free to copy, save, or adapt this prompt for your next negotiation. It’s designed to be reused across cases and contexts.

Evaluate the quality of the negotiation analysis you just generated using the following four criteria.

For each criterion, assign a score from 1–5 and explain the rationale in 2–3 lines.

Criteria

Interest Analysis (30%) – Have you identified a full spectrum of interests (motivations, fears, constraints) for both parties, clearly ranked and going beyond price?

BATNA Analysis (25%) – Have you identified explicit and inferred BATNAs, evaluated consequences of no agreement, and distinguished between strong and weak alternatives?

Scenario Building (25%) – Have you created multiple, distinct, realistic scenarios for the counterpart, each with ranked interests, BATNA, and diagnostic questions?

Value-Creation Insight (20%) – Have you proposed feasible, multi-issue options that expand the pie beyond price?

Then calculate the final weighted score using:

(Interests × 0.30) + (BATNA × 0.25) + (Scenarios × 0.25) + (Value Creation × 0.20)

Finally, provide:

Below is the actual response generated by ChatGPT-5 using this Quality Assessment prompt.

The analysis does a strong job identifying the seller’s interests beyond price: speed of transaction, avoiding holding costs, minimizing hassle, and limiting renegotiation risk.

It also identifies inferred interests (certainty, neutrality of relationship, maintaining neighborhood stability).

On the buyer side, interests are addressed within the scenarios but could be more deeply differentiated, especially around emotional or lifestyle drivers.

The BATNA is correctly identified as the $17,000 offer already available to the seller.

The analysis also infers alternative paths (waiting for a future buyer, holding the land, potential zoning changes).

Consequences of no agreement are addressed, though the relative strength of each BATNA could be more explicit.

Three realistic, distinct buyer scenarios are articulated:

The Practical Expander

The Strategic Investor

The Reluctant Neighbor

Each includes ranked interests and plausible BATNAs with clear diagnostic questions.

The differentiation is meaningful and grounded.

The analysis proposes solid integrative options (closing flexibility, small concessions, cost sharing).

More creative multi-issue trades could be added (conditional price increases, shared improvements, staged agreements).

= (4 × 0.30) + (4 × 0.25) + (5 × 0.25) + (4 × 0.20)

= 4.25 / 5

A strong, reliable analysis with excellent scenario design and well-structured interests.

Areas for improvement include deeper emotional interests and more creative value-creation options.

Improve the Value-Creation Insight criterion by generating five new integrative options tied to the interests in each scenario.

Follow this simple, repeatable process using the TOAST method — Test, Observe, Analyze, Solve, Test Again:

Run your structured prompt to generate the initial output.

Score the output using the four criteria and identify the weakest one.

Determine why that criterion failed — unclear instructions, missing steps, or insufficient detail.

Refine the prompt by adding clarity or constraints targeted at that specific weak area.

Re-run the prompt and re-score the output.

Repeat until the analysis consistently reaches ≥ 4.2.

This is how you train the AI to think better — not just answer better..