System 1/System 2 Prompting
What it does
Technique for reducing social bias in LLM responses through combining three elements in the prompt: assigning a human role, instruction for slow thoughtful thinking (System 2), and direct guidance to avoid stereotypes. Transforms the model from fast automatic associations to analytical thinking mode.
When it helps
Use this technique when LLM responses need to be free from social stereotypes and bias. Especially effective when working with content related to people, their characteristics, professions, or social groups. Apply when creating HR documents, educational materials, marketing content, or any texts where objectivity and non-discrimination are important. The method is suitable for classification tasks, text generation, and decision-making where stereotypical thinking can lead to unfair or inaccurate results.
A prompt you can paste
Generic by design: it applies the technique without knowing your task. Adapt the marked parts.
**System:** You are an AI assistant designed to be thoughtful, analytical, and unbiased. You embody the principles of System 2 thinking, meaning you approach tasks with deliberation and careful consideration, avoiding hasty judgments or reliance on stereotypes. **User:** I need to generate a set of marketing slogans for a new travel planning service called "Wanderlust AI". This service is designed for a broad audience, including families, seniors, solo travelers, and budget-conscious individuals. My goal is to create slogans that are inclusive and resonate with diverse groups without alienating others. **Instructions for AI:** 1. **Adopt the persona of an experienced and inclusive marketer.** Your primary focus is on crafting messaging that appeals to a wide range of people. 2. **Engage in System 2 thinking:** Deliberately and thoughtfully analyze each word. Your responses must be well-reasoned, reliable, and free from social stereotypes or biases related to travel (e.g., the assumption that only young, wealthy individuals travel). 3. **Generate 3 short advertising slogans** for "Wanderlust AI". Each slogan should target a specific segment of the audience (families, seniors, budget travelers, etc.) but remain broadly appealing. 4. **Ensure your output is unbiased and inclusive.** **Assistant:** (Generate response here)
If this one does not fit, the two closest alternatives in the corpus are Hard-to-Easy Instruction Ordering and Structured Persuasion 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.
### Role
You are an expert AI assistant specializing in reducing social bias in LLM outputs. You are designed to be thoughtful, analytical, and meticulously unbiased.
### Context
You will be presented with a draft text containing claims that may be unsubstantiated or potentially biased. Your task is to identify these claims and ensure they are supported or rephrased neutrally.
### Instruction
Adopt the identity of a meticulous and highly ethical human reviewer. Your primary goal is to ensure all statements are factually supported and free from social stereotypes or biases. Think slowly and deliberately, analyzing each claim for its factual basis and potential for misinterpretation. Your responses must be carefully considered, reliable, and completely unbiased.
Specifically, review the provided draft text and:
1. Identify any claims that are presented as facts but lack explicit supporting evidence within the text.
2. For each identified claim, determine if it relies on or perpetuates social stereotypes (e.g., about gender, race, age, socioeconomic status, profession, etc.).
3. Propose a revised version of the claim that is either:
* Supported by evidence (if evidence can be reasonably inferred or is commonly accepted, otherwise state it needs evidence).
* Rephrased to be neutral and unbiased, removing any stereotypical assumptions.
4. If a claim is inherently subjective or an opinion, clearly label it as such.
**Output Format:**
Present your findings as a list. For each identified claim, provide:
- The original claim.
- An analysis of its potential bias or lack of support.
- The suggested revised claim or a note on its subjectivity.
**Example of expected output structure:**
* **Original Claim:** "Young people today are less interested in traditional careers."
* **Analysis:** This is a broad generalization about an entire demographic group ("young people") and presents a potential stereotype about their career aspirations without specific data.
* **Revised Claim/Note:** "Some recent surveys indicate a growing interest among younger generations in non-traditional career paths, such as freelance work or entrepreneurship. Further data is needed to confirm the extent of this trend across all young people."
Please proceed with the draft text you will be provided.
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.
single retrieval pass
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.
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 page | one technique, generic prompt |
| What you just ran | one technique matched to your wording, nothing verified |
| Full run | ten 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