Prompting technique

Confirmation Bias Mitigation Prompting

Unveiling Confirmation Bias in Chain-of-Thought Reasoning, arXiv:2506.12301 (2025) · read the paper

What it does

A technique to combat confirmation bias in LLMs through active prompt formation. It forces the model to doubt obvious answers through explicit instructions to consider alternatives, refute assumptions, and avoid template solutions. It transforms Chain-of-Thought from a justification tool into a mechanism for truth-seeking.

Task typereasoning
Shapechain of thought
Efforttwo or three steps
Addressesinvented facts and citations; answers that are close but wrong; the same prompt giving different answers
Published2025-06-14
AuthorsYue Wan, Xiaowei Jia, Xiang Lorraine Li

When it helps

Apply this technique when working with tasks where the model may have strong biased opinions or tends to give template answers. Particularly effective for common-sense tasks, analysis of contradictory data, generation of non-standard ideas, and fact-checking. Use when objectivity is important and you need to avoid confirmation bias. Not suitable for simple math tasks where CoT already works well. Ideal for situations requiring critical thinking and consideration of alternatives

A prompt you can paste

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

Prompt
# Role:

You are an AI assistant tasked with generating a specific type of output, but you are prone to confirmation bias. Your default tendency is to provide a balanced, objective, or the most statistically probable answer based on your training data.

# Task:

Generate a reusable prompt template that readers can adapt to mitigate confirmation bias when using Chain-of-Thought (CoT) reasoning for complex or subjective tasks. The goal is to force the AI to challenge its initial "belief" or most probable answer.

# Key Instructions for Mitigating Confirmation Bias:

1.  **Identify the "Path of Least Resistance" for the AI:** Before generating the prompt, consider what the AI's most obvious, popular, or simplest answer would be for a given task. This is its potential "belief."
2.  **Directly Attack this "Belief" in the Prompt:** Do not let the AI follow its easiest path. Use explicit instructions to force it to:
    *   **Consider Alternatives:** Directly ask the AI to analyze opposing viewpoints or counterarguments.
    *   **Challenge Assumptions:** Instruct the AI to question the foundational assumptions of the most likely answer.
    *   **Avoid Clichés and Common Tropes:** Explicitly forbid the use of overused phrases, common arguments, or statistically probable but unoriginal ideas.
    *   **Adopt a Contrarian Persona:** Assign a role to the AI that inherently opposes its likely default stance (e.g., a skeptic, an advocate for the underdog, someone with a strong pre-existing opposing opinion).
    *   **Focus on Refutation:** Frame the task not as "prove X," but as "refute Y" (where Y is the common counterargument to X).
3.  **Structure for Clarity and Focus:** Organize the prompt to guide the AI through the process of challenging its bias.

# Prompt Template Structure:

Adapt the following structure for your specific task. Replace bracketed placeholders `[ ]` with your task details.

---

## Role:

You are a `[Describe the AI's persona, ideally one that opposes its default stance on the topic. E.g., "a staunch critic of X," "a contrarian analyst," "an advocate for the underdog position on Y"]`. Your primary goal is to `[State the AI's objective, which should involve exploring a less obvious or opposing perspective]`.

## Context:

The user is seeking information or analysis on `[Briefly describe the topic or problem]`. The most common or statistically probable answer/perspective on this topic is `[Describe the AI's likely default "belief" or the common viewpoint. Be specific.]`.

## Task:

Generate a `[Specify the desired output format, e.g., "detailed analysis," "argumentative essay," "report," "list of ideas"]` that `[State the core objective of the user's request]`.

## Key Instructions for Overcoming Confirmation Bias:

1.  **Explicitly Avoid the Default:** Do NOT provide the common or statistically probable answer described in the Context section. Your primary objective is to explore alternatives.
2.  **Challenge the Dominant Viewpoint:** Focus your analysis on `[Describe the specific counter-argument, opposing perspective, or less common angle the AI should explore. E.g., "the potential downsides of X," "arguments against the prevailing theory," "unconventional solutions to Y"]`.
3.  **Refute Common Counterarguments:** If the task involves persuasion or argumentation, structure your response around directly addressing and refuting the typical objections to your position. Specifically, address:
    *   `[List the 1-3 most common objections or counterarguments the AI should refute. E.g., "The argument that X is too expensive."]`
    *   `[Objection 2]`
    *   `[Objection 3 (optional)]`
4.  **Use Strong, Assertive Language:** Avoid hedging words like "perhaps," "maybe," "it seems," "on the one hand." Use confident and direct language that reflects your assigned role.
5.  **Focus on [Specify a particular aspect to emphasize]:** For example, "focus on the long-term implications," "focus on the ethical considerations," "focus on the user experience."

## Output Format:

Present your response as a `[Specify the desired output format, e.g., "structured report with clear headings," "bulleted list," "narrative essay"]`. Ensure that `[Add any specific formatting requirements, e.g., "each point is supported by reasoning," "the conclusion summarizes the key counterarguments addressed."]`.

---

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: work through a multi-step planning problem. Use it as the pattern for your own case rather than as a finished artefact.

Worked example
# Role:

You are a critical thinking AI, specifically designed to identify and mitigate confirmation bias within Chain-of-Thought (CoT) reasoning processes. Your goal is to challenge initial assumptions and ensure a thorough, unbiased exploration of a problem.

# Task:

Work through the following multi-step planning problem: "Develop a strategy for a small local bookstore to increase foot traffic and sales in the next six months."

# Core Problem Analysis & Bias Mitigation Instructions:

1.  **Identify the "Most Likely" Initial Assumption:**
    *   What is the most common, perhaps cliché, strategy a bookstore might consider first (e.g., "more discounts," "better window displays")? State this initial assumption clearly.
    *   **Instruction:** Do not immediately adopt this assumption. Treat it as a potential bias to overcome.

2.  **Challenge the Initial Assumption (Counter-Argument Generation):**
    *   Generate at least two strong counter-arguments or alternative perspectives that directly challenge the initial assumption identified in Step 1.
    *   **Instruction:** Frame these counter-arguments as if you were a skeptical customer or a competitor. Focus on why the initial assumption might fail or be insufficient.

3.  **Develop Alternative Strategies (Exploration Phase):**
    *   Based on the counter-arguments and a desire to avoid the initial bias, brainstorm at least three distinct, potentially less obvious strategies. These should aim to increase foot traffic and sales.
    *   Consider strategies beyond simple promotions, such as community engagement, unique events, partnerships, or leveraging technology in novel ways.
    *   **Instruction:** For each strategy, explicitly state how it avoids the pitfalls of the initial assumption.

4.  **Chain-of-Thought Reasoning for Chosen Strategy:**
    *   Select ONE of the alternative strategies developed in Step 3 that you believe has the highest potential.
    *   Now, apply a detailed Chain-of-Thought reasoning process to flesh out this chosen strategy. Break it down into actionable steps.
    *   **Instruction:** As you reason through the steps, actively look for potential confirmation bias. For instance, if a step seems too easy or confirms your preference for the chosen strategy, pause and ask: "Is there another way to interpret this? What are the potential downsides I'm overlooking?" Explicitly state any such self-corrections.

5.  **Final Recommendation & Bias Check:**
    *   Present the fully detailed plan for the chosen strategy.
    *   Conclude with a final check: Briefly summarize how this plan actively mitigates the initial confirmation bias identified in Step 1, ensuring a more robust and objective outcome.

# Output Format:

Structure your response clearly using Markdown headers for each step (## Step 1, ## Step 2, etc.). Use bullet points for lists and clear, concise language. Explicitly mention when you are applying a bias mitigation instruction.

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 …Debate-to-Detect (D2D) / Multi-Agent Structured DebateThe method transforms an analysis task into structured debates between multiple AI agents with opposite roles.…

All techniques · Failure modes and fixes