Prompting technique

Debate-to-Detect (D2D) / Multi-Agent Structured Debate

Debate-to-Detect: Reformulating Misinformation Detection as a Real-World Debate with Large Language Models, arXiv:2505.18596 (2025) · read the paper

What it does

The method transforms an analysis task into structured debates between multiple AI agents with opposite roles. Agents go through stages of opening statements, rebuttals, and free debate, after which a judge issues a multidimensional assessment. The adversarial approach forces the model to explore arguments and evidence more deeply, improving the accuracy of conclusions.

Task typereasoning
Shapemulti agent
Efforta pipeline
Addressesinvented facts and citations; answers that are close but wrong; the same prompt giving different answers
Published2025-05-24
AuthorsChen Han, Wenzhen Zheng, Xijin Tang

When it helps

Use this technique when deep fact-checking, critical analysis of claims, or making important decisions based on conflicting information is required. Particularly effective for identifying misinformation, evaluating disputed claims, analyzing complex business decisions or legal arguments. Suitable for situations where a simple answer may be superficial or biased, and the problem needs to be considered from several opposing perspectives. Requires time to set up the structure of debates

A prompt you can paste

Generic by design: it applies the technique without knowing your task. Adapt the marked parts.

Prompt
# Task: Evaluate the potential of a specific claim or statement.

You are an AI system tasked with critically evaluating information using the Debate-to-Detect (D2D) methodology. Your goal is to simulate a structured debate to arrive at a well-reasoned conclusion.

**Core Statement for Evaluation:**
[INSERT THE SPECIFIC CLAIM OR STATEMENT TO BE EVALUATED HERE]

---

### **PHASE 1: ROLE ASSIGNMENT**

Assign the following roles for the debate:

1.  **Agent "Affirmative"**: This agent will argue *in favor* of the core statement. Their profile is: [DESCRIBE THE AFFIRMATIVE AGENT'S PERSPECTIVE OR ROLE, e.g., "A proponent of the idea, focusing on its potential benefits and supporting evidence."]
2.  **Agent "Negative"**: This agent will argue *against* the core statement. Their profile is: [DESCRIBE THE NEGATIVE AGENT'S PERSPECTIVE OR ROLE, e.g., "A skeptic, focusing on counter-arguments, risks, and lack of evidence."]
3.  **Agent "Judge"**: This agent will act as a neutral arbiter, evaluating the arguments and delivering a final verdict.

---

### **PHASE 2: STRUCTURED DEBATE**

Conduct the debate strictly following these stages:

**1. Opening Statements:**
    *   **Affirmative:** Present your initial arguments and evidence supporting the core statement.
    *   **Negative:** Present your initial counter-arguments and evidence refuting the core statement.

**2. Rebuttal Round:**
    *   **Affirmative:** Directly address and refute the arguments made by the Negative agent.
    *   **Negative:** Directly address and refute the arguments made by the Affirmative agent.

**3. Free Debate (1 Round):**
    *   **Affirmative:** Provide one additional argument or clarification in response to the Negative's last point.
    *   **Negative:** Provide one additional argument or clarification in response to the Affirmative's last point.

**4. Closing Statements:**
    *   **Affirmative:** Summarize your strongest points and reiterate your stance.
    *   **Negative:** Summarize your strongest points and reiterate your stance.

---

### **PHASE 3: MULTI-DIMENSIONAL JUDGMENT**

Acting as the **Agent "Judge"**:

1.  **Debate Summary:** Provide a neutral, concise overview of the key arguments presented by both sides.
2.  **Evaluation by Criteria:** Assess the arguments and evidence presented by each agent based on the following criteria. Use a scale of 1-10 for each, where 1 is Poor and 10 is Excellent.
    *   **Factual Accuracy:** How well supported are the claims by verifiable facts?
    *   **Source Reliability:** (If sources were cited) How credible are the cited sources?
    *   **Argument Strength & Logic:** How compelling and logically sound are the arguments?
    *   **Clarity of Presentation:** How easy is it to understand the arguments?
    *   **Ethical Considerations:** Are there any ethical implications to consider regarding the statement or the arguments?
3.  **Final Verdict:** Based on the entire debate and the multi-dimensional evaluation, deliver a final, reasoned conclusion on the core statement. This verdict should clearly state whether the statement is likely true, false, or requires further investigation, with a justification.

**Output Format:** Use Markdown for clear structuring of each phase, agent's contributions, and the final verdict.

If this one does not fit, the two closest alternatives in the corpus are Anti-Bias Prompting and Tabular Prompting, which target the same failure from a different angle.

Worked example

The same technique applied to a concrete job: check a draft for claims that cannot be supported. Use it as the pattern for your own case rather than as a finished artefact.

Worked example
# Role: AI Fact-Checker and Debate Moderator

# Context:
You are tasked with evaluating a draft document. Your primary goal is to identify any claims within the draft that lack sufficient evidence or could be considered unsubstantiated. You will use the Debate-to-Detect (D2D) methodology to rigorously test these claims.

# Task:
Analyze the provided draft text. Identify specific claims that appear to be unsubstantiated or require stronger evidence. For each identified claim, conduct a structured debate using the D2D framework to assess its validity.

# Methodology: Debate-to-Detect (D2D)

You will simulate a debate process for each identified claim. For each claim, follow these stages:

## Stage 1: Role Assignment
Assign the following roles for the debate:
1.  **Agent "Affirmative"**: This agent will defend the claim, presenting arguments and evidence *in favor* of its truthfulness. This agent should adopt a perspective that seeks to validate the claim, potentially as a proponent of the idea or a writer who believes in its accuracy.
2.  **Agent "Negative"**: This agent will challenge the claim, presenting counter-arguments, identifying logical fallacies, and highlighting the lack of evidence or potential for misinformation. This agent should adopt a skeptical perspective, acting as a fact-checker or an editor focused on accuracy.
3.  **Agent "Judge"**: This agent will objectively evaluate the arguments presented by both "Affirmative" and "Negative" agents and render a final verdict on the claim's substantiation.

## Stage 2: Structured Debate
Conduct the debate in the following sequence:

1.  **Opening Statements:**
    *   **Affirmative:** Present the claim and provide initial supporting arguments or evidence.
    *   **Negative:** Present the core challenge to the claim, outlining why it might be unsubstantiated or problematic.

2.  **Rebuttal Round:**
    *   **Affirmative:** Directly address and refute the arguments made by the "Negative" agent.
    *   **Negative:** Directly address and refute the arguments made by the "Affirmative" agent, and further challenge the initial supporting points.

3.  **Free Debate (1 Round):**
    *   Allow each agent to present one additional argument or counter-argument in response to the latest points made by the opponent.

4.  **Closing Statements:**
    *   **Affirmative:** Summarize their strongest points and reiterate why the claim is valid.
    *   **Negative:** Summarize their strongest counter-arguments and reiterate why the claim is unsubstantiated.

## Stage 3: Multi-dimensional Judgment
Act as the "Judge" and provide a final assessment for the claim:

1.  **Debate Summary:** Briefly summarize the main arguments from both sides.
2.  **Evaluation Criteria (Score 1-10 for each):**
    *   **Factual Accuracy:** How well does the claim align with established facts?
    *   **Evidence Strength:** How robust and credible is the evidence presented?
    *   **Argument Quality:** How logical and persuasive were the arguments from both sides?
    *   **Potential for Misinformation:** How likely is this claim to be misleading if presented without further qualification?
3.  **Verdict:** Based on the evaluation, provide a clear verdict:
    *   **Substantiated:** The claim is well-supported by evidence.
    *   **Partially Substantiated:** The claim has some support but requires significant qualification or additional evidence.
    *   **Unsubstantiated:** The claim lacks sufficient evidence and is likely misinformation.
    *   **Contested:** There is significant debate, and a definitive verdict is difficult without more information.

# Output Format:

For each claim identified in the draft:
1.  **Claim:** Clearly state the claim being evaluated.
2.  **Debate Simulation:** Present the dialogue for each stage (Roles, Opening Statements, Rebuttal, Free Debate, Closing Statements). Clearly label each agent's contribution.
3.  **Judge's Verdict:** Present the summary, evaluation scores, and final verdict.

**Begin by reading the draft text and identifying potential claims to evaluate.**

Get this written for your actual task

Paste what you are trying to do and the corpus will be matched against it directly. Free, no account, about ten seconds.

Free · no signup · ~10s
0.00match confidence
single retrieval pass
Prompt for your task

      

That number is low on purpose, and it is real. It is the raw similarity of one retrieval pass: no specialist read the paper, no judge compared anything, the first plausible match won.

6,235techniques in the corpus
one of which is this page

Picking the right one for a specific task is the work, and it is the work GetDecision does.

This pageone technique, generic prompt
What you just ranone technique matched to your wording, nothing verified
Full runten specialists read the papers in full, a judge ranks the top three for your task and shows its reasoning, generation on the model you pick, saved to your history

See the top three for your taskTen specialists, a judge, and the reasoning shown. Free account, first run included.

Run the full analysis

Related techniques

Anti-Bias PromptingThe technique of neutral prompt formulation that excludes cognitive biases and user prejudice. Essence: avoid …Tabular PromptingA method of presenting data as a markdown table before submitting a request to an LLM. The tabular structure h…Multilingual Ensemble PromptingThe method involves translating a complex query into 3-4 different languages, obtaining separate answers from …Guardrail PromptingGuardrail Prompting — adding explicit restriction instructions to the prompt that prevent excessive reasoning …

All techniques · Failure modes and fixes